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);
535 ff_h264_unref_picture(h, &h->last_pic_for_ec);
536 memcpy(h, h1, sizeof(H264Context));
538 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
539 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
541 memset(&h->er, 0, sizeof(h->er));
542 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
543 memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
545 h->slice_ctx = orig_slice_ctx;
547 memset(&h->slice_ctx[0].mb, 0, sizeof(h->slice_ctx[0].mb));
548 memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
549 memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
553 h->qscale_table_pool = NULL;
554 h->mb_type_pool = NULL;
555 h->ref_index_pool = NULL;
556 h->motion_val_pool = NULL;
557 h->intra4x4_pred_mode= NULL;
558 h->non_zero_count = NULL;
559 h->slice_table_base = NULL;
560 h->slice_table = NULL;
562 h->chroma_pred_mode_table = NULL;
563 memset(h->mvd_table, 0, sizeof(h->mvd_table));
564 h->direct_table = NULL;
565 h->list_counts = NULL;
568 h->rbsp_buffer = NULL;
569 h->rbsp_buffer_size = 0;
571 if (h1->context_initialized) {
572 h->context_initialized = 0;
574 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
575 av_frame_unref(&h->cur_pic.f);
576 h->cur_pic.tf.f = &h->cur_pic.f;
578 ret = ff_h264_alloc_tables(h);
580 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
583 ret = ff_h264_context_init(h);
585 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
590 h->bipred_scratchpad = NULL;
591 h->edge_emu_buffer = NULL;
593 h->thread_context[0] = h;
594 h->context_initialized = h1->context_initialized;
597 h->avctx->coded_height = h1->avctx->coded_height;
598 h->avctx->coded_width = h1->avctx->coded_width;
599 h->avctx->width = h1->avctx->width;
600 h->avctx->height = h1->avctx->height;
601 h->coded_picture_number = h1->coded_picture_number;
602 h->first_field = h1->first_field;
603 h->picture_structure = h1->picture_structure;
604 h->droppable = h1->droppable;
605 h->low_delay = h1->low_delay;
607 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
608 ff_h264_unref_picture(h, &h->DPB[i]);
609 if (h1->DPB && h1->DPB[i].f.buf[0] &&
610 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
614 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
615 ff_h264_unref_picture(h, &h->cur_pic);
616 if (h1->cur_pic.f.buf[0]) {
617 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
622 h->workaround_bugs = h1->workaround_bugs;
623 h->low_delay = h1->low_delay;
624 h->droppable = h1->droppable;
626 // extradata/NAL handling
627 h->is_avc = h1->is_avc;
630 if ((ret = copy_parameter_set((void **)h->sps_buffers,
631 (void **)h1->sps_buffers,
632 MAX_SPS_COUNT, sizeof(SPS))) < 0)
635 if ((ret = copy_parameter_set((void **)h->pps_buffers,
636 (void **)h1->pps_buffers,
637 MAX_PPS_COUNT, sizeof(PPS))) < 0)
641 // Dequantization matrices
642 // FIXME these are big - can they be only copied when PPS changes?
643 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
645 for (i = 0; i < 6; i++)
646 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
647 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
649 for (i = 0; i < 6; i++)
650 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
651 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
653 h->dequant_coeff_pps = h1->dequant_coeff_pps;
656 copy_fields(h, h1, poc_lsb, default_ref_list);
659 copy_fields(h, h1, short_ref, thread_context);
661 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
662 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
663 copy_picture_range(h->delayed_pic, h1->delayed_pic,
664 MAX_DELAYED_PIC_COUNT + 2, h, h1);
666 h->frame_recovered = h1->frame_recovered;
668 if (context_reinitialized)
669 ff_h264_set_parameter_from_sps(h);
675 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
676 h->prev_poc_msb = h->poc_msb;
677 h->prev_poc_lsb = h->poc_lsb;
679 h->prev_frame_num_offset = h->frame_num_offset;
680 h->prev_frame_num = h->frame_num;
681 h->outputed_poc = h->next_outputed_poc;
683 h->recovery_frame = h1->recovery_frame;
688 static int h264_frame_start(H264Context *h)
692 const int pixel_shift = h->pixel_shift;
694 1<<(h->sps.bit_depth_luma-1),
695 1<<(h->sps.bit_depth_chroma-1),
696 1<<(h->sps.bit_depth_chroma-1),
700 if (!ff_thread_can_start_frame(h->avctx)) {
701 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
705 release_unused_pictures(h, 1);
706 h->cur_pic_ptr = NULL;
708 i = find_unused_picture(h);
710 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
715 pic->reference = h->droppable ? 0 : h->picture_structure;
716 pic->f.coded_picture_number = h->coded_picture_number++;
717 pic->field_picture = h->picture_structure != PICT_FRAME;
720 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
722 * See decode_nal_units().
724 pic->f.key_frame = 0;
727 pic->invalid_gap = 0;
728 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
730 if ((ret = alloc_picture(h, pic)) < 0)
732 if(!h->frame_recovered && !h->avctx->hwaccel &&
733 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
734 avpriv_color_frame(&pic->f, c);
736 h->cur_pic_ptr = pic;
737 ff_h264_unref_picture(h, &h->cur_pic);
738 if (CONFIG_ERROR_RESILIENCE) {
739 ff_h264_set_erpic(&h->er.cur_pic, NULL);
742 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
745 if (CONFIG_ERROR_RESILIENCE) {
746 ff_er_frame_start(&h->er);
747 ff_h264_set_erpic(&h->er.last_pic, NULL);
748 ff_h264_set_erpic(&h->er.next_pic, NULL);
751 assert(h->linesize && h->uvlinesize);
753 for (i = 0; i < 16; i++) {
754 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
755 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
757 for (i = 0; i < 16; i++) {
758 h->block_offset[16 + i] =
759 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
760 h->block_offset[48 + 16 + i] =
761 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
764 /* We mark the current picture as non-reference after allocating it, so
765 * that if we break out due to an error it can be released automatically
766 * in the next ff_mpv_frame_start().
768 h->cur_pic_ptr->reference = 0;
770 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
772 h->next_output_pic = NULL;
774 assert(h->cur_pic_ptr->long_ref == 0);
779 static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *sl,
781 uint8_t *src_cb, uint8_t *src_cr,
782 int linesize, int uvlinesize,
787 const int pixel_shift = h->pixel_shift;
788 int chroma444 = CHROMA444(h);
789 int chroma422 = CHROMA422(h);
792 src_cb -= uvlinesize;
793 src_cr -= uvlinesize;
795 if (!simple && FRAME_MBAFF(h)) {
798 top_border = h->top_borders[0][sl->mb_x];
799 AV_COPY128(top_border, src_y + 15 * linesize);
801 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
802 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
805 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
806 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
807 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
808 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
810 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
811 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
813 } else if (chroma422) {
815 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
816 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
818 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
819 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
823 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
824 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
826 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
827 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
832 } else if (MB_MBAFF(h)) {
838 top_border = h->top_borders[top_idx][sl->mb_x];
839 /* There are two lines saved, the line above the top macroblock
840 * of a pair, and the line above the bottom macroblock. */
841 AV_COPY128(top_border, src_y + 16 * linesize);
843 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
845 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
848 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
849 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
850 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
851 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
853 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
854 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
856 } else if (chroma422) {
858 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
859 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
861 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
862 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
866 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
867 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
869 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
870 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
877 * Initialize implicit_weight table.
878 * @param field 0/1 initialize the weight for interlaced MBAFF
879 * -1 initializes the rest
881 static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int field)
883 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
885 for (i = 0; i < 2; i++) {
886 sl->luma_weight_flag[i] = 0;
887 sl->chroma_weight_flag[i] = 0;
891 if (h->picture_structure == PICT_FRAME) {
892 cur_poc = h->cur_pic_ptr->poc;
894 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
896 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
897 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
899 sl->use_weight_chroma = 0;
903 ref_count0 = sl->ref_count[0];
904 ref_count1 = sl->ref_count[1];
906 cur_poc = h->cur_pic_ptr->field_poc[field];
908 ref_count0 = 16 + 2 * sl->ref_count[0];
909 ref_count1 = 16 + 2 * sl->ref_count[1];
913 sl->use_weight_chroma = 2;
914 sl->luma_log2_weight_denom = 5;
915 sl->chroma_log2_weight_denom = 5;
917 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
918 int poc0 = sl->ref_list[0][ref0].poc;
919 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
921 if (!sl->ref_list[0][ref0].long_ref && !sl->ref_list[1][ref1].long_ref) {
922 int poc1 = sl->ref_list[1][ref1].poc;
923 int td = av_clip_int8(poc1 - poc0);
925 int tb = av_clip_int8(cur_poc - poc0);
926 int tx = (16384 + (FFABS(td) >> 1)) / td;
927 int dist_scale_factor = (tb * tx + 32) >> 8;
928 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
929 w = 64 - dist_scale_factor;
933 sl->implicit_weight[ref0][ref1][0] =
934 sl->implicit_weight[ref0][ref1][1] = w;
936 sl->implicit_weight[ref0][ref1][field] = w;
943 * initialize scan tables
945 static void init_scan_tables(H264Context *h)
948 for (i = 0; i < 16; i++) {
949 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
950 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
951 h->field_scan[i] = TRANSPOSE(field_scan[i]);
954 for (i = 0; i < 64; i++) {
955 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
956 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
957 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
958 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
959 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
962 if (h->sps.transform_bypass) { // FIXME same ugly
963 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
964 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
965 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
966 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
967 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
968 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
970 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
971 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
972 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
973 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
974 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
975 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
980 * Replicate H264 "master" context to thread contexts.
982 static int clone_slice(H264Context *dst, H264Context *src)
984 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
985 dst->cur_pic_ptr = src->cur_pic_ptr;
986 dst->cur_pic = src->cur_pic;
987 dst->linesize = src->linesize;
988 dst->uvlinesize = src->uvlinesize;
989 dst->first_field = src->first_field;
991 dst->prev_poc_msb = src->prev_poc_msb;
992 dst->prev_poc_lsb = src->prev_poc_lsb;
993 dst->prev_frame_num_offset = src->prev_frame_num_offset;
994 dst->prev_frame_num = src->prev_frame_num;
995 dst->short_ref_count = src->short_ref_count;
997 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
998 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
999 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1001 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1002 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1007 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1009 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
1010 CONFIG_H264_VAAPI_HWACCEL + \
1011 (CONFIG_H264_VDA_HWACCEL * 2) + \
1012 CONFIG_H264_VDPAU_HWACCEL)
1013 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
1014 const enum AVPixelFormat *choices = pix_fmts;
1017 switch (h->sps.bit_depth_luma) {
1020 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1021 *fmt++ = AV_PIX_FMT_GBRP9;
1023 *fmt++ = AV_PIX_FMT_YUV444P9;
1024 } else if (CHROMA422(h))
1025 *fmt++ = AV_PIX_FMT_YUV422P9;
1027 *fmt++ = AV_PIX_FMT_YUV420P9;
1031 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1032 *fmt++ = AV_PIX_FMT_GBRP10;
1034 *fmt++ = AV_PIX_FMT_YUV444P10;
1035 } else if (CHROMA422(h))
1036 *fmt++ = AV_PIX_FMT_YUV422P10;
1038 *fmt++ = AV_PIX_FMT_YUV420P10;
1042 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1043 *fmt++ = AV_PIX_FMT_GBRP12;
1045 *fmt++ = AV_PIX_FMT_YUV444P12;
1046 } else if (CHROMA422(h))
1047 *fmt++ = AV_PIX_FMT_YUV422P12;
1049 *fmt++ = AV_PIX_FMT_YUV420P12;
1053 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1054 *fmt++ = AV_PIX_FMT_GBRP14;
1056 *fmt++ = AV_PIX_FMT_YUV444P14;
1057 } else if (CHROMA422(h))
1058 *fmt++ = AV_PIX_FMT_YUV422P14;
1060 *fmt++ = AV_PIX_FMT_YUV420P14;
1063 #if CONFIG_H264_VDPAU_HWACCEL
1064 *fmt++ = AV_PIX_FMT_VDPAU;
1067 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1068 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1069 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1070 *fmt++ = AV_PIX_FMT_GBRP;
1071 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1072 *fmt++ = AV_PIX_FMT_YUVJ444P;
1074 *fmt++ = AV_PIX_FMT_YUV444P;
1075 } else if (CHROMA422(h)) {
1076 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1077 *fmt++ = AV_PIX_FMT_YUVJ422P;
1079 *fmt++ = AV_PIX_FMT_YUV422P;
1081 #if CONFIG_H264_DXVA2_HWACCEL
1082 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1084 #if CONFIG_H264_VAAPI_HWACCEL
1085 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1087 #if CONFIG_H264_VDA_HWACCEL
1088 *fmt++ = AV_PIX_FMT_VDA_VLD;
1089 *fmt++ = AV_PIX_FMT_VDA;
1091 if (h->avctx->codec->pix_fmts)
1092 choices = h->avctx->codec->pix_fmts;
1093 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1094 *fmt++ = AV_PIX_FMT_YUVJ420P;
1096 *fmt++ = AV_PIX_FMT_YUV420P;
1100 av_log(h->avctx, AV_LOG_ERROR,
1101 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1102 return AVERROR_INVALIDDATA;
1105 *fmt = AV_PIX_FMT_NONE;
1107 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1108 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1110 return ff_thread_get_format(h->avctx, choices);
1113 /* export coded and cropped frame dimensions to AVCodecContext */
1114 static int init_dimensions(H264Context *h)
1116 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1117 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1118 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1119 h->sps.crop_right || h->sps.crop_bottom;
1120 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1121 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1123 /* handle container cropping */
1124 if (!crop_present &&
1125 FFALIGN(h->avctx->width, 16) == h->width &&
1126 FFALIGN(h->avctx->height, 16) == h->height) {
1127 width = h->avctx->width;
1128 height = h->avctx->height;
1131 if (width <= 0 || height <= 0) {
1132 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1134 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1135 return AVERROR_INVALIDDATA;
1137 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1138 h->sps.crop_bottom =
1148 h->avctx->coded_width = h->width;
1149 h->avctx->coded_height = h->height;
1150 h->avctx->width = width;
1151 h->avctx->height = height;
1156 static int h264_slice_header_init(H264Context *h, int reinit)
1158 int nb_slices = (HAVE_THREADS &&
1159 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1160 h->avctx->thread_count : 1;
1163 ff_set_sar(h->avctx, h->sps.sar);
1164 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1165 &h->chroma_x_shift, &h->chroma_y_shift);
1167 if (h->sps.timing_info_present_flag) {
1168 int64_t den = h->sps.time_scale;
1169 if (h->x264_build < 44U)
1171 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1172 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1176 ff_h264_free_tables(h, 0);
1178 h->prev_interlaced_frame = 1;
1180 init_scan_tables(h);
1181 ret = ff_h264_alloc_tables(h);
1183 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1187 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1190 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1192 max_slices = H264_MAX_THREADS;
1193 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1194 " reducing to %d\n", nb_slices, max_slices);
1195 nb_slices = max_slices;
1197 h->slice_context_count = nb_slices;
1199 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1200 ret = ff_h264_context_init(h);
1202 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1206 for (i = 1; i < h->slice_context_count; i++) {
1208 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1210 ret = AVERROR(ENOMEM);
1213 c->avctx = h->avctx;
1215 c->h264dsp = h->h264dsp;
1216 c->h264qpel = h->h264qpel;
1217 c->h264chroma = h->h264chroma;
1220 c->pixel_shift = h->pixel_shift;
1221 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1222 c->width = h->width;
1223 c->height = h->height;
1224 c->linesize = h->linesize;
1225 c->uvlinesize = h->uvlinesize;
1226 c->chroma_x_shift = h->chroma_x_shift;
1227 c->chroma_y_shift = h->chroma_y_shift;
1228 c->droppable = h->droppable;
1229 c->low_delay = h->low_delay;
1230 c->mb_width = h->mb_width;
1231 c->mb_height = h->mb_height;
1232 c->mb_stride = h->mb_stride;
1233 c->mb_num = h->mb_num;
1234 c->flags = h->flags;
1235 c->workaround_bugs = h->workaround_bugs;
1236 c->pict_type = h->pict_type;
1238 h->slice_ctx[i].h264 = c;
1240 init_scan_tables(c);
1241 clone_tables(c, &h->slice_ctx[i], h, i);
1242 c->context_initialized = 1;
1245 for (i = 0; i < h->slice_context_count; i++)
1246 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1247 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1252 h->context_initialized = 1;
1256 ff_h264_free_tables(h, 0);
1257 h->context_initialized = 0;
1261 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1264 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1265 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1266 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1273 * Decode a slice header.
1274 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1276 * @param h h264context
1277 * @param h0 h264 master context (differs from 'h' when doing sliced based
1278 * parallel decoding)
1280 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1282 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Context *h0)
1284 unsigned int first_mb_in_slice;
1285 unsigned int pps_id;
1287 unsigned int slice_type, tmp, i, j;
1288 int last_pic_structure, last_pic_droppable;
1290 int needs_reinit = 0;
1291 int field_pic_flag, bottom_field_flag;
1292 int first_slice = h == h0 && !h0->current_slice;
1293 int frame_num, picture_structure, droppable;
1296 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1297 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1299 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1301 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1302 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1303 ff_h264_field_end(h, sl, 1);
1306 h0->current_slice = 0;
1307 if (!h0->first_field) {
1308 if (h->cur_pic_ptr && !h->droppable) {
1309 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1310 h->picture_structure == PICT_BOTTOM_FIELD);
1312 h->cur_pic_ptr = NULL;
1316 slice_type = get_ue_golomb_31(&sl->gb);
1317 if (slice_type > 9) {
1318 av_log(h->avctx, AV_LOG_ERROR,
1319 "slice type %d too large at %d\n",
1320 slice_type, first_mb_in_slice);
1321 return AVERROR_INVALIDDATA;
1323 if (slice_type > 4) {
1325 sl->slice_type_fixed = 1;
1327 sl->slice_type_fixed = 0;
1329 slice_type = golomb_to_pict_type[slice_type];
1330 sl->slice_type = slice_type;
1331 sl->slice_type_nos = slice_type & 3;
1333 if (h->nal_unit_type == NAL_IDR_SLICE &&
1334 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1335 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1336 return AVERROR_INVALIDDATA;
1340 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1341 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1342 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1343 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1344 h->avctx->skip_frame >= AVDISCARD_ALL) {
1345 return SLICE_SKIPED;
1348 // to make a few old functions happy, it's wrong though
1349 h->pict_type = sl->slice_type;
1351 pps_id = get_ue_golomb(&sl->gb);
1352 if (pps_id >= MAX_PPS_COUNT) {
1353 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1354 return AVERROR_INVALIDDATA;
1356 if (!h0->pps_buffers[pps_id]) {
1357 av_log(h->avctx, AV_LOG_ERROR,
1358 "non-existing PPS %u referenced\n",
1360 return AVERROR_INVALIDDATA;
1362 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1363 av_log(h->avctx, AV_LOG_ERROR,
1364 "PPS change from %d to %d forbidden\n",
1365 h0->au_pps_id, pps_id);
1366 return AVERROR_INVALIDDATA;
1369 pps = h0->pps_buffers[pps_id];
1371 if (!h0->sps_buffers[pps->sps_id]) {
1372 av_log(h->avctx, AV_LOG_ERROR,
1373 "non-existing SPS %u referenced\n",
1375 return AVERROR_INVALIDDATA;
1378 h->pps = *h0->pps_buffers[pps_id];
1380 if (pps->sps_id != h->sps.sps_id ||
1381 pps->sps_id != h->current_sps_id ||
1382 h0->sps_buffers[pps->sps_id]->new) {
1385 av_log(h->avctx, AV_LOG_ERROR,
1386 "SPS changed in the middle of the frame\n");
1387 return AVERROR_INVALIDDATA;
1390 h->sps = *h0->sps_buffers[h->pps.sps_id];
1392 if (h->mb_width != h->sps.mb_width ||
1393 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1394 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1395 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1399 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1400 h->chroma_format_idc != h->sps.chroma_format_idc) {
1401 h->bit_depth_luma = h->sps.bit_depth_luma;
1402 h->chroma_format_idc = h->sps.chroma_format_idc;
1405 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1409 h->avctx->profile = ff_h264_get_profile(&h->sps);
1410 h->avctx->level = h->sps.level_idc;
1411 h->avctx->refs = h->sps.ref_frame_count;
1413 must_reinit = (h->context_initialized &&
1414 ( 16*h->sps.mb_width != h->avctx->coded_width
1415 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1416 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1417 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1418 || h->mb_width != h->sps.mb_width
1419 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1421 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1424 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1427 h->mb_width = h->sps.mb_width;
1428 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1429 h->mb_num = h->mb_width * h->mb_height;
1430 h->mb_stride = h->mb_width + 1;
1432 h->b_stride = h->mb_width * 4;
1434 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1436 h->width = 16 * h->mb_width;
1437 h->height = 16 * h->mb_height;
1439 ret = init_dimensions(h);
1443 if (h->sps.video_signal_type_present_flag) {
1444 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1446 if (h->sps.colour_description_present_flag) {
1447 if (h->avctx->colorspace != h->sps.colorspace)
1449 h->avctx->color_primaries = h->sps.color_primaries;
1450 h->avctx->color_trc = h->sps.color_trc;
1451 h->avctx->colorspace = h->sps.colorspace;
1455 if (h->context_initialized &&
1456 (must_reinit || needs_reinit)) {
1458 av_log(h->avctx, AV_LOG_ERROR,
1459 "changing width %d -> %d / height %d -> %d on "
1461 h->width, h->avctx->coded_width,
1462 h->height, h->avctx->coded_height,
1463 h0->current_slice + 1);
1464 return AVERROR_INVALIDDATA;
1467 av_assert1(first_slice);
1469 ff_h264_flush_change(h);
1471 if ((ret = get_pixel_format(h, 1)) < 0)
1473 h->avctx->pix_fmt = ret;
1475 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1476 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1478 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1479 av_log(h->avctx, AV_LOG_ERROR,
1480 "h264_slice_header_init() failed\n");
1484 if (!h->context_initialized) {
1486 av_log(h->avctx, AV_LOG_ERROR,
1487 "Cannot (re-)initialize context during parallel decoding.\n");
1488 return AVERROR_PATCHWELCOME;
1491 if ((ret = get_pixel_format(h, 1)) < 0)
1493 h->avctx->pix_fmt = ret;
1495 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1496 av_log(h->avctx, AV_LOG_ERROR,
1497 "h264_slice_header_init() failed\n");
1502 if (first_slice && h->dequant_coeff_pps != pps_id) {
1503 h->dequant_coeff_pps = pps_id;
1504 ff_h264_init_dequant_tables(h);
1507 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1509 if (h0->frame_num != frame_num) {
1510 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1511 h0->frame_num, frame_num);
1512 return AVERROR_INVALIDDATA;
1517 h->mb_aff_frame = 0;
1518 last_pic_structure = h0->picture_structure;
1519 last_pic_droppable = h0->droppable;
1520 droppable = h->nal_ref_idc == 0;
1521 if (h->sps.frame_mbs_only_flag) {
1522 picture_structure = PICT_FRAME;
1524 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1525 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1528 field_pic_flag = get_bits1(&sl->gb);
1530 if (field_pic_flag) {
1531 bottom_field_flag = get_bits1(&sl->gb);
1532 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1534 picture_structure = PICT_FRAME;
1535 h->mb_aff_frame = h->sps.mb_aff;
1538 if (h0->current_slice) {
1539 if (last_pic_structure != picture_structure ||
1540 last_pic_droppable != droppable) {
1541 av_log(h->avctx, AV_LOG_ERROR,
1542 "Changing field mode (%d -> %d) between slices is not allowed\n",
1543 last_pic_structure, h->picture_structure);
1544 return AVERROR_INVALIDDATA;
1545 } else if (!h0->cur_pic_ptr) {
1546 av_log(h->avctx, AV_LOG_ERROR,
1547 "unset cur_pic_ptr on slice %d\n",
1548 h0->current_slice + 1);
1549 return AVERROR_INVALIDDATA;
1553 h->picture_structure = picture_structure;
1554 h->droppable = droppable;
1555 h->frame_num = frame_num;
1556 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1558 if (h0->current_slice == 0) {
1559 /* Shorten frame num gaps so we don't have to allocate reference
1560 * frames just to throw them away */
1561 if (h->frame_num != h->prev_frame_num) {
1562 int unwrap_prev_frame_num = h->prev_frame_num;
1563 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1565 if (unwrap_prev_frame_num > h->frame_num)
1566 unwrap_prev_frame_num -= max_frame_num;
1568 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1569 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1570 if (unwrap_prev_frame_num < 0)
1571 unwrap_prev_frame_num += max_frame_num;
1573 h->prev_frame_num = unwrap_prev_frame_num;
1577 /* See if we have a decoded first field looking for a pair...
1578 * Here, we're using that to see if we should mark previously
1579 * decode frames as "finished".
1580 * We have to do that before the "dummy" in-between frame allocation,
1581 * since that can modify h->cur_pic_ptr. */
1582 if (h0->first_field) {
1583 assert(h0->cur_pic_ptr);
1584 assert(h0->cur_pic_ptr->f.buf[0]);
1585 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1587 /* Mark old field/frame as completed */
1588 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1589 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1590 last_pic_structure == PICT_BOTTOM_FIELD);
1593 /* figure out if we have a complementary field pair */
1594 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1595 /* Previous field is unmatched. Don't display it, but let it
1596 * remain for reference if marked as such. */
1597 if (last_pic_structure != PICT_FRAME) {
1598 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1599 last_pic_structure == PICT_TOP_FIELD);
1602 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1603 /* This and previous field were reference, but had
1604 * different frame_nums. Consider this field first in
1605 * pair. Throw away previous field except for reference
1607 if (last_pic_structure != PICT_FRAME) {
1608 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1609 last_pic_structure == PICT_TOP_FIELD);
1612 /* Second field in complementary pair */
1613 if (!((last_pic_structure == PICT_TOP_FIELD &&
1614 h->picture_structure == PICT_BOTTOM_FIELD) ||
1615 (last_pic_structure == PICT_BOTTOM_FIELD &&
1616 h->picture_structure == PICT_TOP_FIELD))) {
1617 av_log(h->avctx, AV_LOG_ERROR,
1618 "Invalid field mode combination %d/%d\n",
1619 last_pic_structure, h->picture_structure);
1620 h->picture_structure = last_pic_structure;
1621 h->droppable = last_pic_droppable;
1622 return AVERROR_INVALIDDATA;
1623 } else if (last_pic_droppable != h->droppable) {
1624 avpriv_request_sample(h->avctx,
1625 "Found reference and non-reference fields in the same frame, which");
1626 h->picture_structure = last_pic_structure;
1627 h->droppable = last_pic_droppable;
1628 return AVERROR_PATCHWELCOME;
1634 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1635 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1636 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1637 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1638 h->frame_num, h->prev_frame_num);
1639 if (!h->sps.gaps_in_frame_num_allowed_flag)
1640 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1641 h->last_pocs[i] = INT_MIN;
1642 ret = h264_frame_start(h);
1644 h0->first_field = 0;
1648 h->prev_frame_num++;
1649 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1650 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1651 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1652 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1653 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1654 ret = ff_generate_sliding_window_mmcos(h, 1);
1655 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1657 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1658 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1660 /* Error concealment: If a ref is missing, copy the previous ref
1662 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1663 * many assumptions about there being no actual duplicates.
1664 * FIXME: This does not copy padding for out-of-frame motion
1665 * vectors. Given we are concealing a lost frame, this probably
1666 * is not noticeable by comparison, but it should be fixed. */
1667 if (h->short_ref_count) {
1669 av_image_copy(h->short_ref[0]->f.data,
1670 h->short_ref[0]->f.linesize,
1671 (const uint8_t **)prev->f.data,
1676 h->short_ref[0]->poc = prev->poc + 2;
1678 h->short_ref[0]->frame_num = h->prev_frame_num;
1682 /* See if we have a decoded first field looking for a pair...
1683 * We're using that to see whether to continue decoding in that
1684 * frame, or to allocate a new one. */
1685 if (h0->first_field) {
1686 assert(h0->cur_pic_ptr);
1687 assert(h0->cur_pic_ptr->f.buf[0]);
1688 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1690 /* figure out if we have a complementary field pair */
1691 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1692 /* Previous field is unmatched. Don't display it, but let it
1693 * remain for reference if marked as such. */
1694 h0->missing_fields ++;
1695 h0->cur_pic_ptr = NULL;
1696 h0->first_field = FIELD_PICTURE(h);
1698 h0->missing_fields = 0;
1699 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1700 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1701 h0->picture_structure==PICT_BOTTOM_FIELD);
1702 /* This and the previous field had different frame_nums.
1703 * Consider this field first in pair. Throw away previous
1704 * one except for reference purposes. */
1705 h0->first_field = 1;
1706 h0->cur_pic_ptr = NULL;
1708 /* Second field in complementary pair */
1709 h0->first_field = 0;
1713 /* Frame or first field in a potentially complementary pair */
1714 h0->first_field = FIELD_PICTURE(h);
1717 if (!FIELD_PICTURE(h) || h0->first_field) {
1718 if (h264_frame_start(h) < 0) {
1719 h0->first_field = 0;
1720 return AVERROR_INVALIDDATA;
1723 release_unused_pictures(h, 0);
1725 /* Some macroblocks can be accessed before they're available in case
1726 * of lost slices, MBAFF or threading. */
1727 if (FIELD_PICTURE(h)) {
1728 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1729 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1731 memset(h->slice_table, -1,
1732 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1734 h0->last_slice_type = -1;
1736 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1739 /* can't be in alloc_tables because linesize isn't known there.
1740 * FIXME: redo bipred weight to not require extra buffer? */
1741 for (i = 0; i < h->slice_context_count; i++)
1742 if (h->thread_context[i]) {
1743 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1748 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1750 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1751 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1752 first_mb_in_slice >= h->mb_num) {
1753 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1754 return AVERROR_INVALIDDATA;
1756 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1757 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1758 FIELD_OR_MBAFF_PICTURE(h);
1759 if (h->picture_structure == PICT_BOTTOM_FIELD)
1760 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1761 av_assert1(sl->mb_y < h->mb_height);
1763 if (h->picture_structure == PICT_FRAME) {
1764 h->curr_pic_num = h->frame_num;
1765 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1767 h->curr_pic_num = 2 * h->frame_num + 1;
1768 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1771 if (h->nal_unit_type == NAL_IDR_SLICE)
1772 get_ue_golomb(&sl->gb); /* idr_pic_id */
1774 if (h->sps.poc_type == 0) {
1775 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1777 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1778 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1781 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1782 h->delta_poc[0] = get_se_golomb(&sl->gb);
1784 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1785 h->delta_poc[1] = get_se_golomb(&sl->gb);
1788 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1790 if (h->pps.redundant_pic_cnt_present)
1791 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1793 ret = ff_set_ref_count(h, sl);
1797 if (slice_type != AV_PICTURE_TYPE_I &&
1798 (h0->current_slice == 0 ||
1799 slice_type != h0->last_slice_type ||
1800 memcmp(h0->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1802 ff_h264_fill_default_ref_list(h, sl);
1805 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1806 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1808 sl->ref_count[1] = sl->ref_count[0] = 0;
1813 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1814 (h->pps.weighted_bipred_idc == 1 &&
1815 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1816 ff_pred_weight_table(h, sl);
1817 else if (h->pps.weighted_bipred_idc == 2 &&
1818 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1819 implicit_weight_table(h, sl, -1);
1822 for (i = 0; i < 2; i++) {
1823 sl->luma_weight_flag[i] = 0;
1824 sl->chroma_weight_flag[i] = 0;
1828 // If frame-mt is enabled, only update mmco tables for the first slice
1829 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1830 // or h->mmco, which will cause ref list mix-ups and decoding errors
1831 // further down the line. This may break decoding if the first slice is
1832 // corrupt, thus we only do this if frame-mt is enabled.
1833 if (h->nal_ref_idc) {
1834 ret = ff_h264_decode_ref_pic_marking(h0, &sl->gb,
1835 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1836 h0->current_slice == 0);
1837 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1838 return AVERROR_INVALIDDATA;
1841 if (FRAME_MBAFF(h)) {
1842 ff_h264_fill_mbaff_ref_list(h, sl);
1844 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1845 implicit_weight_table(h, sl, 0);
1846 implicit_weight_table(h, sl, 1);
1850 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1851 ff_h264_direct_dist_scale_factor(h, sl);
1852 ff_h264_direct_ref_list_init(h, sl);
1854 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1855 tmp = get_ue_golomb_31(&sl->gb);
1857 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1858 return AVERROR_INVALIDDATA;
1860 sl->cabac_init_idc = tmp;
1863 sl->last_qscale_diff = 0;
1864 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1865 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1866 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1867 return AVERROR_INVALIDDATA;
1870 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1871 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1872 // FIXME qscale / qp ... stuff
1873 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1874 get_bits1(&sl->gb); /* sp_for_switch_flag */
1875 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1876 sl->slice_type == AV_PICTURE_TYPE_SI)
1877 get_se_golomb(&sl->gb); /* slice_qs_delta */
1879 sl->deblocking_filter = 1;
1880 sl->slice_alpha_c0_offset = 0;
1881 sl->slice_beta_offset = 0;
1882 if (h->pps.deblocking_filter_parameters_present) {
1883 tmp = get_ue_golomb_31(&sl->gb);
1885 av_log(h->avctx, AV_LOG_ERROR,
1886 "deblocking_filter_idc %u out of range\n", tmp);
1887 return AVERROR_INVALIDDATA;
1889 sl->deblocking_filter = tmp;
1890 if (sl->deblocking_filter < 2)
1891 sl->deblocking_filter ^= 1; // 1<->0
1893 if (sl->deblocking_filter) {
1894 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1895 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1896 if (sl->slice_alpha_c0_offset > 12 ||
1897 sl->slice_alpha_c0_offset < -12 ||
1898 sl->slice_beta_offset > 12 ||
1899 sl->slice_beta_offset < -12) {
1900 av_log(h->avctx, AV_LOG_ERROR,
1901 "deblocking filter parameters %d %d out of range\n",
1902 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1903 return AVERROR_INVALIDDATA;
1908 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1909 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1910 h->nal_unit_type != NAL_IDR_SLICE) ||
1911 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1912 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1913 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1914 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1915 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1916 h->nal_ref_idc == 0))
1917 sl->deblocking_filter = 0;
1919 if (sl->deblocking_filter == 1 && h0->max_contexts > 1) {
1920 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1921 /* Cheat slightly for speed:
1922 * Do not bother to deblock across slices. */
1923 sl->deblocking_filter = 2;
1925 h0->max_contexts = 1;
1926 if (!h0->single_decode_warning) {
1927 av_log(h->avctx, AV_LOG_INFO,
1928 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1929 "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"
1930 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1931 "but will generate non-standard-compliant output.\n");
1932 h0->single_decode_warning = 1;
1935 av_log(h->avctx, AV_LOG_ERROR,
1936 "Deblocking switched inside frame.\n");
1937 return SLICE_SINGLETHREAD;
1941 sl->qp_thresh = 15 -
1942 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1944 h->pps.chroma_qp_index_offset[0],
1945 h->pps.chroma_qp_index_offset[1]) +
1946 6 * (h->sps.bit_depth_luma - 8);
1948 h0->last_slice_type = slice_type;
1949 memcpy(h0->last_ref_count, sl->ref_count, sizeof(h0->last_ref_count));
1950 sl->slice_num = ++h0->current_slice;
1953 h0->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1954 if ( h0->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1955 && h0->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1956 && sl->slice_num >= MAX_SLICES) {
1957 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1958 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);
1961 for (j = 0; j < 2; j++) {
1963 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1964 for (i = 0; i < 16; i++) {
1966 if (j < sl->list_count && i < sl->ref_count[j] &&
1967 sl->ref_list[j][i].f.buf[0]) {
1969 AVBuffer *buf = sl->ref_list[j][i].f.buf[0]->buffer;
1970 for (k = 0; k < h->short_ref_count; k++)
1971 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1975 for (k = 0; k < h->long_ref_count; k++)
1976 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1977 id_list[i] = h->short_ref_count + k;
1985 for (i = 0; i < 16; i++)
1986 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1988 ref2frm[18 + 1] = -1;
1989 for (i = 16; i < 48; i++)
1990 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1991 (sl->ref_list[j][i].reference & 3);
1994 h0->au_pps_id = pps_id;
1996 h0->sps_buffers[h->pps.sps_id]->new = 0;
1997 h->current_sps_id = h->pps.sps_id;
1999 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
2000 av_log(h->avctx, AV_LOG_DEBUG,
2001 "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",
2003 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
2005 av_get_picture_type_char(sl->slice_type),
2006 sl->slice_type_fixed ? " fix" : "",
2007 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2008 pps_id, h->frame_num,
2009 h->cur_pic_ptr->field_poc[0],
2010 h->cur_pic_ptr->field_poc[1],
2011 sl->ref_count[0], sl->ref_count[1],
2013 sl->deblocking_filter,
2014 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
2016 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
2017 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2023 int ff_h264_get_slice_type(const H264SliceContext *sl)
2025 switch (sl->slice_type) {
2026 case AV_PICTURE_TYPE_P:
2028 case AV_PICTURE_TYPE_B:
2030 case AV_PICTURE_TYPE_I:
2032 case AV_PICTURE_TYPE_SP:
2034 case AV_PICTURE_TYPE_SI:
2037 return AVERROR_INVALIDDATA;
2041 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2042 H264SliceContext *sl,
2043 int mb_type, int top_xy,
2044 int left_xy[LEFT_MBS],
2046 int left_type[LEFT_MBS],
2047 int mb_xy, int list)
2049 int b_stride = h->b_stride;
2050 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
2051 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
2052 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2053 if (USES_LIST(top_type, list)) {
2054 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2055 const int b8_xy = 4 * top_xy + 2;
2056 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2057 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2058 ref_cache[0 - 1 * 8] =
2059 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2060 ref_cache[2 - 1 * 8] =
2061 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2063 AV_ZERO128(mv_dst - 1 * 8);
2064 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2067 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2068 if (USES_LIST(left_type[LTOP], list)) {
2069 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2070 const int b8_xy = 4 * left_xy[LTOP] + 1;
2071 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2072 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2073 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2074 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2075 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2077 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2078 ref_cache[-1 + 16] =
2079 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2081 AV_ZERO32(mv_dst - 1 + 0);
2082 AV_ZERO32(mv_dst - 1 + 8);
2083 AV_ZERO32(mv_dst - 1 + 16);
2084 AV_ZERO32(mv_dst - 1 + 24);
2087 ref_cache[-1 + 16] =
2088 ref_cache[-1 + 24] = LIST_NOT_USED;
2093 if (!USES_LIST(mb_type, list)) {
2094 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2095 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2096 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2097 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2098 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2103 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2104 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2105 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2106 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2107 AV_WN32A(&ref_cache[0 * 8], ref01);
2108 AV_WN32A(&ref_cache[1 * 8], ref01);
2109 AV_WN32A(&ref_cache[2 * 8], ref23);
2110 AV_WN32A(&ref_cache[3 * 8], ref23);
2114 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2115 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2116 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2117 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2118 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2124 * @return non zero if the loop filter can be skipped
2126 static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
2128 const int mb_xy = sl->mb_xy;
2129 int top_xy, left_xy[LEFT_MBS];
2130 int top_type, left_type[LEFT_MBS];
2134 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2136 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2137 * stuff, I can't imagine that these complex rules are worth it. */
2139 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2140 if (FRAME_MBAFF(h)) {
2141 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2142 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2144 if (left_mb_field_flag != curr_mb_field_flag)
2145 left_xy[LTOP] -= h->mb_stride;
2147 if (curr_mb_field_flag)
2148 top_xy += h->mb_stride &
2149 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2150 if (left_mb_field_flag != curr_mb_field_flag)
2151 left_xy[LBOT] += h->mb_stride;
2155 sl->top_mb_xy = top_xy;
2156 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2157 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2159 /* For sufficiently low qp, filtering wouldn't do anything.
2160 * This is a conservative estimate: could also check beta_offset
2161 * and more accurate chroma_qp. */
2162 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2163 int qp = h->cur_pic.qscale_table[mb_xy];
2164 if (qp <= qp_thresh &&
2165 (left_xy[LTOP] < 0 ||
2166 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2168 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2169 if (!FRAME_MBAFF(h))
2171 if ((left_xy[LTOP] < 0 ||
2172 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2173 (top_xy < h->mb_stride ||
2174 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2179 top_type = h->cur_pic.mb_type[top_xy];
2180 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2181 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2182 if (sl->deblocking_filter == 2) {
2183 if (h->slice_table[top_xy] != sl->slice_num)
2185 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2186 left_type[LTOP] = left_type[LBOT] = 0;
2188 if (h->slice_table[top_xy] == 0xFFFF)
2190 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2191 left_type[LTOP] = left_type[LBOT] = 0;
2193 sl->top_type = top_type;
2194 sl->left_type[LTOP] = left_type[LTOP];
2195 sl->left_type[LBOT] = left_type[LBOT];
2197 if (IS_INTRA(mb_type))
2200 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2201 top_type, left_type, mb_xy, 0);
2202 if (sl->list_count == 2)
2203 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2204 top_type, left_type, mb_xy, 1);
2206 nnz = h->non_zero_count[mb_xy];
2207 nnz_cache = sl->non_zero_count_cache;
2208 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2209 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2210 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2211 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2212 sl->cbp = h->cbp_table[mb_xy];
2215 nnz = h->non_zero_count[top_xy];
2216 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2219 if (left_type[LTOP]) {
2220 nnz = h->non_zero_count[left_xy[LTOP]];
2221 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2222 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2223 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2224 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2227 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2228 * from what the loop filter needs */
2229 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2230 if (IS_8x8DCT(top_type)) {
2231 nnz_cache[4 + 8 * 0] =
2232 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2233 nnz_cache[6 + 8 * 0] =
2234 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2236 if (IS_8x8DCT(left_type[LTOP])) {
2237 nnz_cache[3 + 8 * 1] =
2238 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2240 if (IS_8x8DCT(left_type[LBOT])) {
2241 nnz_cache[3 + 8 * 3] =
2242 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2245 if (IS_8x8DCT(mb_type)) {
2246 nnz_cache[scan8[0]] =
2247 nnz_cache[scan8[1]] =
2248 nnz_cache[scan8[2]] =
2249 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2251 nnz_cache[scan8[0 + 4]] =
2252 nnz_cache[scan8[1 + 4]] =
2253 nnz_cache[scan8[2 + 4]] =
2254 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2256 nnz_cache[scan8[0 + 8]] =
2257 nnz_cache[scan8[1 + 8]] =
2258 nnz_cache[scan8[2 + 8]] =
2259 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2261 nnz_cache[scan8[0 + 12]] =
2262 nnz_cache[scan8[1 + 12]] =
2263 nnz_cache[scan8[2 + 12]] =
2264 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2271 static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2273 uint8_t *dest_y, *dest_cb, *dest_cr;
2274 int linesize, uvlinesize, mb_x, mb_y;
2275 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2276 const int old_slice_type = sl->slice_type;
2277 const int pixel_shift = h->pixel_shift;
2278 const int block_h = 16 >> h->chroma_y_shift;
2280 if (sl->deblocking_filter) {
2281 for (mb_x = start_x; mb_x < end_x; mb_x++)
2282 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2284 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2285 sl->slice_num = h->slice_table[mb_xy];
2286 mb_type = h->cur_pic.mb_type[mb_xy];
2287 sl->list_count = h->list_counts[mb_xy];
2291 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2295 dest_y = h->cur_pic.f.data[0] +
2296 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2297 dest_cb = h->cur_pic.f.data[1] +
2298 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2299 mb_y * h->uvlinesize * block_h;
2300 dest_cr = h->cur_pic.f.data[2] +
2301 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2302 mb_y * h->uvlinesize * block_h;
2303 // FIXME simplify above
2306 linesize = sl->mb_linesize = h->linesize * 2;
2307 uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
2308 if (mb_y & 1) { // FIXME move out of this function?
2309 dest_y -= h->linesize * 15;
2310 dest_cb -= h->uvlinesize * (block_h - 1);
2311 dest_cr -= h->uvlinesize * (block_h - 1);
2314 linesize = sl->mb_linesize = h->linesize;
2315 uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
2317 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2319 if (fill_filter_caches(h, sl, mb_type))
2321 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2322 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2324 if (FRAME_MBAFF(h)) {
2325 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2326 linesize, uvlinesize);
2328 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2329 dest_cr, linesize, uvlinesize);
2333 sl->slice_type = old_slice_type;
2335 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2336 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2337 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2340 static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
2342 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2343 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2344 h->cur_pic.mb_type[mb_xy - 1] :
2345 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2346 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2347 h->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2351 * Draw edges and report progress for the last MB row.
2353 static void decode_finish_row(H264Context *h, H264SliceContext *sl)
2355 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2356 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2357 int height = 16 << FRAME_MBAFF(h);
2358 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2360 if (sl->deblocking_filter) {
2361 if ((top + height) >= pic_height)
2362 height += deblock_border;
2363 top -= deblock_border;
2366 if (top >= pic_height || (top + height) < 0)
2369 height = FFMIN(height, pic_height - top);
2371 height = top + height;
2375 ff_h264_draw_horiz_band(h, sl, top, height);
2377 if (h->droppable || h->er.error_occurred)
2380 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2381 h->picture_structure == PICT_BOTTOM_FIELD);
2384 static void er_add_slice(H264Context *h, H264SliceContext *sl,
2385 int startx, int starty,
2386 int endx, int endy, int status)
2388 if (CONFIG_ERROR_RESILIENCE) {
2389 ERContext *er = &h->er;
2391 ff_er_add_slice(er, startx, starty, endx, endy, status);
2395 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2397 H264SliceContext *sl = arg;
2398 H264Context *h = sl->h264;
2399 int lf_x_start = sl->mb_x;
2401 sl->mb_skip_run = -1;
2403 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2405 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2406 avctx->codec_id != AV_CODEC_ID_H264 ||
2407 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2409 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2410 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2412 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2413 prev_status &= ~ VP_START;
2414 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2415 h->er.error_occurred = 1;
2421 align_get_bits(&sl->gb);
2424 ff_init_cabac_decoder(&sl->cabac,
2425 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2426 (get_bits_left(&sl->gb) + 7) / 8);
2428 ff_h264_init_cabac_states(h, sl);
2432 int ret = ff_h264_decode_mb_cabac(h, sl);
2434 // STOP_TIMER("decode_mb_cabac")
2437 ff_h264_hl_decode_mb(h, sl);
2439 // FIXME optimal? or let mb_decode decode 16x32 ?
2440 if (ret >= 0 && FRAME_MBAFF(h)) {
2443 ret = ff_h264_decode_mb_cabac(h, sl);
2446 ff_h264_hl_decode_mb(h, sl);
2449 eos = get_cabac_terminate(&sl->cabac);
2451 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2452 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2453 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2454 sl->mb_y, ER_MB_END);
2455 if (sl->mb_x >= lf_x_start)
2456 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2459 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2460 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2461 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2462 av_log(h->avctx, AV_LOG_ERROR,
2463 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2465 sl->cabac.bytestream_end - sl->cabac.bytestream);
2466 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2467 sl->mb_y, ER_MB_ERROR);
2468 return AVERROR_INVALIDDATA;
2471 if (++sl->mb_x >= h->mb_width) {
2472 loop_filter(h, sl, lf_x_start, sl->mb_x);
2473 sl->mb_x = lf_x_start = 0;
2474 decode_finish_row(h, sl);
2476 if (FIELD_OR_MBAFF_PICTURE(h)) {
2478 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2479 predict_field_decoding_flag(h, sl);
2483 if (eos || sl->mb_y >= h->mb_height) {
2484 tprintf(h->avctx, "slice end %d %d\n",
2485 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2486 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2487 sl->mb_y, ER_MB_END);
2488 if (sl->mb_x > lf_x_start)
2489 loop_filter(h, sl, lf_x_start, sl->mb_x);
2495 int ret = ff_h264_decode_mb_cavlc(h, sl);
2498 ff_h264_hl_decode_mb(h, sl);
2500 // FIXME optimal? or let mb_decode decode 16x32 ?
2501 if (ret >= 0 && FRAME_MBAFF(h)) {
2503 ret = ff_h264_decode_mb_cavlc(h, sl);
2506 ff_h264_hl_decode_mb(h, sl);
2511 av_log(h->avctx, AV_LOG_ERROR,
2512 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2513 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2514 sl->mb_y, ER_MB_ERROR);
2518 if (++sl->mb_x >= h->mb_width) {
2519 loop_filter(h, sl, lf_x_start, sl->mb_x);
2520 sl->mb_x = lf_x_start = 0;
2521 decode_finish_row(h, sl);
2523 if (FIELD_OR_MBAFF_PICTURE(h)) {
2525 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2526 predict_field_decoding_flag(h, sl);
2528 if (sl->mb_y >= h->mb_height) {
2529 tprintf(h->avctx, "slice end %d %d\n",
2530 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2532 if ( get_bits_left(&sl->gb) == 0
2533 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2534 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
2535 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2539 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
2540 sl->mb_x, sl->mb_y, ER_MB_END);
2542 return AVERROR_INVALIDDATA;
2547 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2548 tprintf(h->avctx, "slice end %d %d\n",
2549 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2551 if (get_bits_left(&sl->gb) == 0) {
2552 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
2553 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2554 if (sl->mb_x > lf_x_start)
2555 loop_filter(h, sl, lf_x_start, sl->mb_x);
2559 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2560 sl->mb_y, ER_MB_ERROR);
2562 return AVERROR_INVALIDDATA;
2570 * Call decode_slice() for each context.
2572 * @param h h264 master context
2573 * @param context_count number of contexts to execute
2575 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2577 AVCodecContext *const avctx = h->avctx;
2579 H264SliceContext *sl;
2582 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2584 if (h->avctx->hwaccel ||
2585 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2587 if (context_count == 1) {
2588 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2589 h->mb_y = h->slice_ctx[0].mb_y;
2592 av_assert0(context_count > 0);
2593 for (i = 1; i < context_count; i++) {
2594 hx = h->thread_context[i];
2595 if (CONFIG_ERROR_RESILIENCE) {
2596 hx->er.error_count = 0;
2598 hx->x264_build = h->x264_build;
2601 avctx->execute(avctx, decode_slice, h->slice_ctx,
2602 NULL, context_count, sizeof(h->slice_ctx[0]));
2604 /* pull back stuff from slices to master context */
2605 hx = h->thread_context[context_count - 1];
2606 sl = &h->slice_ctx[context_count - 1];
2608 h->droppable = hx->droppable;
2609 h->picture_structure = hx->picture_structure;
2610 if (CONFIG_ERROR_RESILIENCE) {
2611 for (i = 1; i < context_count; i++)
2612 h->er.error_count += h->thread_context[i]->er.error_count;