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, H264Context *src, int i)
389 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
390 dst->non_zero_count = src->non_zero_count;
391 dst->slice_table = src->slice_table;
392 dst->cbp_table = src->cbp_table;
393 dst->mb2b_xy = src->mb2b_xy;
394 dst->mb2br_xy = src->mb2br_xy;
395 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
396 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
397 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
398 dst->direct_table = src->direct_table;
399 dst->list_counts = src->list_counts;
401 dst->cur_pic_ptr = src->cur_pic_ptr;
402 dst->cur_pic = src->cur_pic;
403 dst->bipred_scratchpad = NULL;
404 dst->edge_emu_buffer = NULL;
405 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
406 src->sps.chroma_format_idc);
409 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
411 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
412 (((pic) && (pic) >= (old_ctx)->DPB && \
413 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
414 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
416 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
417 H264Context *new_base,
418 H264Context *old_base)
422 for (i = 0; i < count; i++) {
423 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
424 IN_RANGE(from[i], old_base->DPB,
425 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
427 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
431 static int copy_parameter_set(void **to, void **from, int count, int size)
435 for (i = 0; i < count; i++) {
436 if (to[i] && !from[i]) {
438 } else if (from[i] && !to[i]) {
439 to[i] = av_malloc(size);
441 return AVERROR(ENOMEM);
445 memcpy(to[i], from[i], size);
451 #define copy_fields(to, from, start_field, end_field) \
452 memcpy(&(to)->start_field, &(from)->start_field, \
453 (char *)&(to)->end_field - (char *)&(to)->start_field)
455 static int h264_slice_header_init(H264Context *h, int reinit);
457 int ff_h264_update_thread_context(AVCodecContext *dst,
458 const AVCodecContext *src)
460 H264Context *h = dst->priv_data, *h1 = src->priv_data;
461 int inited = h->context_initialized, err = 0;
462 int context_reinitialized = 0;
469 (h->width != h1->width ||
470 h->height != h1->height ||
471 h->mb_width != h1->mb_width ||
472 h->mb_height != h1->mb_height ||
473 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
474 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
475 h->sps.colorspace != h1->sps.colorspace)) {
477 /* set bits_per_raw_sample to the previous value. the check for changed
478 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
479 * the current value */
480 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
482 av_freep(&h->bipred_scratchpad);
484 h->width = h1->width;
485 h->height = h1->height;
486 h->mb_height = h1->mb_height;
487 h->mb_width = h1->mb_width;
488 h->mb_num = h1->mb_num;
489 h->mb_stride = h1->mb_stride;
490 h->b_stride = h1->b_stride;
492 if ((ret = copy_parameter_set((void **)h->sps_buffers,
493 (void **)h1->sps_buffers,
494 MAX_SPS_COUNT, sizeof(SPS))) < 0)
497 if ((ret = copy_parameter_set((void **)h->pps_buffers,
498 (void **)h1->pps_buffers,
499 MAX_PPS_COUNT, sizeof(PPS))) < 0)
503 if ((err = h264_slice_header_init(h, 1)) < 0) {
504 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
507 context_reinitialized = 1;
510 h264_set_parameter_from_sps(h);
511 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
512 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
515 /* update linesize on resize for h264. The h264 decoder doesn't
516 * necessarily call ff_mpv_frame_start in the new thread */
517 h->linesize = h1->linesize;
518 h->uvlinesize = h1->uvlinesize;
520 /* copy block_offset since frame_start may not be called */
521 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
524 for (i = 0; i < MAX_SPS_COUNT; i++)
525 av_freep(h->sps_buffers + i);
527 for (i = 0; i < MAX_PPS_COUNT; i++)
528 av_freep(h->pps_buffers + i);
530 av_freep(&h->rbsp_buffer[0]);
531 av_freep(&h->rbsp_buffer[1]);
532 ff_h264_unref_picture(h, &h->last_pic_for_ec);
533 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
534 memcpy(&h->cabac, &h1->cabac,
535 sizeof(H264Context) - offsetof(H264Context, cabac));
536 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
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->mb, 0, sizeof(h->mb));
543 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
544 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
545 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
546 memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
550 h->qscale_table_pool = NULL;
551 h->mb_type_pool = NULL;
552 h->ref_index_pool = NULL;
553 h->motion_val_pool = NULL;
554 h->intra4x4_pred_mode= NULL;
555 h->non_zero_count = NULL;
556 h->slice_table_base = NULL;
557 h->slice_table = NULL;
559 h->chroma_pred_mode_table = NULL;
560 memset(h->mvd_table, 0, sizeof(h->mvd_table));
561 h->direct_table = NULL;
562 h->list_counts = NULL;
565 for (i = 0; i < 2; i++) {
566 h->rbsp_buffer[i] = NULL;
567 h->rbsp_buffer_size[i] = 0;
570 if (h1->context_initialized) {
571 h->context_initialized = 0;
573 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
574 av_frame_unref(&h->cur_pic.f);
575 h->cur_pic.tf.f = &h->cur_pic.f;
577 ret = ff_h264_alloc_tables(h);
579 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
582 ret = ff_h264_context_init(h);
584 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
589 h->bipred_scratchpad = NULL;
590 h->edge_emu_buffer = NULL;
592 h->thread_context[0] = h;
593 h->context_initialized = h1->context_initialized;
596 h->avctx->coded_height = h1->avctx->coded_height;
597 h->avctx->coded_width = h1->avctx->coded_width;
598 h->avctx->width = h1->avctx->width;
599 h->avctx->height = h1->avctx->height;
600 h->coded_picture_number = h1->coded_picture_number;
601 h->first_field = h1->first_field;
602 h->picture_structure = h1->picture_structure;
603 h->qscale = h1->qscale;
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, redundant_pic_count);
659 copy_fields(h, h1, short_ref, cabac_init_idc);
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, uint8_t *src_y,
780 uint8_t *src_cb, uint8_t *src_cr,
781 int linesize, int uvlinesize,
786 const int pixel_shift = h->pixel_shift;
787 int chroma444 = CHROMA444(h);
788 int chroma422 = CHROMA422(h);
791 src_cb -= uvlinesize;
792 src_cr -= uvlinesize;
794 if (!simple && FRAME_MBAFF(h)) {
797 top_border = h->top_borders[0][h->mb_x];
798 AV_COPY128(top_border, src_y + 15 * linesize);
800 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
801 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
804 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
805 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
806 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
807 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
809 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
810 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
812 } else if (chroma422) {
814 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
815 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
817 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
818 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
822 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
823 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
825 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
826 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
831 } else if (MB_MBAFF(h)) {
837 top_border = h->top_borders[top_idx][h->mb_x];
838 /* There are two lines saved, the line above the top macroblock
839 * of a pair, and the line above the bottom macroblock. */
840 AV_COPY128(top_border, src_y + 16 * linesize);
842 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
844 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
847 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
848 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
849 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
850 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
852 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
853 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
855 } else if (chroma422) {
857 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
858 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
860 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
861 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
865 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
866 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
868 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
869 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
876 * Initialize implicit_weight table.
877 * @param field 0/1 initialize the weight for interlaced MBAFF
878 * -1 initializes the rest
880 static void implicit_weight_table(H264Context *h, int field)
882 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
884 for (i = 0; i < 2; i++) {
885 h->luma_weight_flag[i] = 0;
886 h->chroma_weight_flag[i] = 0;
890 if (h->picture_structure == PICT_FRAME) {
891 cur_poc = h->cur_pic_ptr->poc;
893 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
895 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
896 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
898 h->use_weight_chroma = 0;
902 ref_count0 = h->ref_count[0];
903 ref_count1 = h->ref_count[1];
905 cur_poc = h->cur_pic_ptr->field_poc[field];
907 ref_count0 = 16 + 2 * h->ref_count[0];
908 ref_count1 = 16 + 2 * h->ref_count[1];
912 h->use_weight_chroma = 2;
913 h->luma_log2_weight_denom = 5;
914 h->chroma_log2_weight_denom = 5;
916 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
917 int poc0 = h->ref_list[0][ref0].poc;
918 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
920 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
921 int poc1 = h->ref_list[1][ref1].poc;
922 int td = av_clip_int8(poc1 - poc0);
924 int tb = av_clip_int8(cur_poc - poc0);
925 int tx = (16384 + (FFABS(td) >> 1)) / td;
926 int dist_scale_factor = (tb * tx + 32) >> 8;
927 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
928 w = 64 - dist_scale_factor;
932 h->implicit_weight[ref0][ref1][0] =
933 h->implicit_weight[ref0][ref1][1] = w;
935 h->implicit_weight[ref0][ref1][field] = w;
942 * initialize scan tables
944 static void init_scan_tables(H264Context *h)
947 for (i = 0; i < 16; i++) {
948 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
949 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
950 h->field_scan[i] = TRANSPOSE(field_scan[i]);
953 for (i = 0; i < 64; i++) {
954 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
955 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
956 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
957 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
958 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
961 if (h->sps.transform_bypass) { // FIXME same ugly
962 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
963 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
964 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
965 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
966 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
967 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
969 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
970 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
971 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
972 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
973 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
974 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
979 * Replicate H264 "master" context to thread contexts.
981 static int clone_slice(H264Context *dst, H264Context *src)
983 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
984 dst->cur_pic_ptr = src->cur_pic_ptr;
985 dst->cur_pic = src->cur_pic;
986 dst->linesize = src->linesize;
987 dst->uvlinesize = src->uvlinesize;
988 dst->first_field = src->first_field;
990 dst->prev_poc_msb = src->prev_poc_msb;
991 dst->prev_poc_lsb = src->prev_poc_lsb;
992 dst->prev_frame_num_offset = src->prev_frame_num_offset;
993 dst->prev_frame_num = src->prev_frame_num;
994 dst->short_ref_count = src->short_ref_count;
996 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
997 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
998 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1000 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1001 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1006 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1008 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
1009 CONFIG_H264_VAAPI_HWACCEL + \
1010 (CONFIG_H264_VDA_HWACCEL * 2) + \
1011 CONFIG_H264_VDPAU_HWACCEL)
1012 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
1013 const enum AVPixelFormat *choices = pix_fmts;
1016 switch (h->sps.bit_depth_luma) {
1019 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1020 *fmt++ = AV_PIX_FMT_GBRP9;
1022 *fmt++ = AV_PIX_FMT_YUV444P9;
1023 } else if (CHROMA422(h))
1024 *fmt++ = AV_PIX_FMT_YUV422P9;
1026 *fmt++ = AV_PIX_FMT_YUV420P9;
1030 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1031 *fmt++ = AV_PIX_FMT_GBRP10;
1033 *fmt++ = AV_PIX_FMT_YUV444P10;
1034 } else if (CHROMA422(h))
1035 *fmt++ = AV_PIX_FMT_YUV422P10;
1037 *fmt++ = AV_PIX_FMT_YUV420P10;
1041 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1042 *fmt++ = AV_PIX_FMT_GBRP12;
1044 *fmt++ = AV_PIX_FMT_YUV444P12;
1045 } else if (CHROMA422(h))
1046 *fmt++ = AV_PIX_FMT_YUV422P12;
1048 *fmt++ = AV_PIX_FMT_YUV420P12;
1052 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1053 *fmt++ = AV_PIX_FMT_GBRP14;
1055 *fmt++ = AV_PIX_FMT_YUV444P14;
1056 } else if (CHROMA422(h))
1057 *fmt++ = AV_PIX_FMT_YUV422P14;
1059 *fmt++ = AV_PIX_FMT_YUV420P14;
1062 #if CONFIG_H264_VDPAU_HWACCEL
1063 *fmt++ = AV_PIX_FMT_VDPAU;
1066 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1067 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1068 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1069 *fmt++ = AV_PIX_FMT_GBRP;
1070 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1071 *fmt++ = AV_PIX_FMT_YUVJ444P;
1073 *fmt++ = AV_PIX_FMT_YUV444P;
1074 } else if (CHROMA422(h)) {
1075 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1076 *fmt++ = AV_PIX_FMT_YUVJ422P;
1078 *fmt++ = AV_PIX_FMT_YUV422P;
1080 #if CONFIG_H264_DXVA2_HWACCEL
1081 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1083 #if CONFIG_H264_VAAPI_HWACCEL
1084 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1086 #if CONFIG_H264_VDA_HWACCEL
1087 *fmt++ = AV_PIX_FMT_VDA_VLD;
1088 *fmt++ = AV_PIX_FMT_VDA;
1090 if (h->avctx->codec->pix_fmts)
1091 choices = h->avctx->codec->pix_fmts;
1092 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1093 *fmt++ = AV_PIX_FMT_YUVJ420P;
1095 *fmt++ = AV_PIX_FMT_YUV420P;
1099 av_log(h->avctx, AV_LOG_ERROR,
1100 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1101 return AVERROR_INVALIDDATA;
1104 *fmt = AV_PIX_FMT_NONE;
1106 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1107 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1109 return ff_thread_get_format(h->avctx, choices);
1112 /* export coded and cropped frame dimensions to AVCodecContext */
1113 static int init_dimensions(H264Context *h)
1115 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1116 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1117 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1118 h->sps.crop_right || h->sps.crop_bottom;
1119 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1120 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1122 /* handle container cropping */
1123 if (!crop_present &&
1124 FFALIGN(h->avctx->width, 16) == h->width &&
1125 FFALIGN(h->avctx->height, 16) == h->height) {
1126 width = h->avctx->width;
1127 height = h->avctx->height;
1130 if (width <= 0 || height <= 0) {
1131 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1133 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1134 return AVERROR_INVALIDDATA;
1136 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1137 h->sps.crop_bottom =
1147 h->avctx->coded_width = h->width;
1148 h->avctx->coded_height = h->height;
1149 h->avctx->width = width;
1150 h->avctx->height = height;
1155 static int h264_slice_header_init(H264Context *h, int reinit)
1157 int nb_slices = (HAVE_THREADS &&
1158 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1159 h->avctx->thread_count : 1;
1162 ff_set_sar(h->avctx, h->sps.sar);
1163 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1164 &h->chroma_x_shift, &h->chroma_y_shift);
1166 if (h->sps.timing_info_present_flag) {
1167 int64_t den = h->sps.time_scale;
1168 if (h->x264_build < 44U)
1170 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1171 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1175 ff_h264_free_tables(h, 0);
1177 h->prev_interlaced_frame = 1;
1179 init_scan_tables(h);
1180 ret = ff_h264_alloc_tables(h);
1182 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1186 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1189 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1191 max_slices = H264_MAX_THREADS;
1192 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1193 " reducing to %d\n", nb_slices, max_slices);
1194 nb_slices = max_slices;
1196 h->slice_context_count = nb_slices;
1198 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1199 ret = ff_h264_context_init(h);
1201 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1205 for (i = 1; i < h->slice_context_count; i++) {
1207 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1209 ret = AVERROR(ENOMEM);
1212 c->avctx = h->avctx;
1214 c->h264dsp = h->h264dsp;
1215 c->h264qpel = h->h264qpel;
1216 c->h264chroma = h->h264chroma;
1219 c->pixel_shift = h->pixel_shift;
1220 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1221 c->width = h->width;
1222 c->height = h->height;
1223 c->linesize = h->linesize;
1224 c->uvlinesize = h->uvlinesize;
1225 c->chroma_x_shift = h->chroma_x_shift;
1226 c->chroma_y_shift = h->chroma_y_shift;
1227 c->qscale = h->qscale;
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 init_scan_tables(c);
1239 clone_tables(c, h, i);
1240 c->context_initialized = 1;
1243 for (i = 0; i < h->slice_context_count; i++)
1244 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1245 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1250 h->context_initialized = 1;
1254 ff_h264_free_tables(h, 0);
1255 h->context_initialized = 0;
1259 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1262 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1263 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1264 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1271 * Decode a slice header.
1272 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1274 * @param h h264context
1275 * @param h0 h264 master context (differs from 'h' when doing sliced based
1276 * parallel decoding)
1278 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1280 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1282 unsigned int first_mb_in_slice;
1283 unsigned int pps_id;
1285 unsigned int slice_type, tmp, i, j;
1286 int last_pic_structure, last_pic_droppable;
1288 int needs_reinit = 0;
1289 int field_pic_flag, bottom_field_flag;
1290 int first_slice = h == h0 && !h0->current_slice;
1291 int frame_num, picture_structure, droppable;
1294 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1295 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1297 first_mb_in_slice = get_ue_golomb_long(&h->gb);
1299 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1300 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1301 ff_h264_field_end(h, 1);
1304 h0->current_slice = 0;
1305 if (!h0->first_field) {
1306 if (h->cur_pic_ptr && !h->droppable) {
1307 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1308 h->picture_structure == PICT_BOTTOM_FIELD);
1310 h->cur_pic_ptr = NULL;
1314 slice_type = get_ue_golomb_31(&h->gb);
1315 if (slice_type > 9) {
1316 av_log(h->avctx, AV_LOG_ERROR,
1317 "slice type %d too large at %d %d\n",
1318 slice_type, h->mb_x, h->mb_y);
1319 return AVERROR_INVALIDDATA;
1321 if (slice_type > 4) {
1323 h->slice_type_fixed = 1;
1325 h->slice_type_fixed = 0;
1327 slice_type = golomb_to_pict_type[slice_type];
1328 h->slice_type = slice_type;
1329 h->slice_type_nos = slice_type & 3;
1331 if (h->nal_unit_type == NAL_IDR_SLICE &&
1332 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1333 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1334 return AVERROR_INVALIDDATA;
1338 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1339 (h->avctx->skip_frame >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1340 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1341 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1342 h->avctx->skip_frame >= AVDISCARD_ALL) {
1343 return SLICE_SKIPED;
1346 // to make a few old functions happy, it's wrong though
1347 h->pict_type = h->slice_type;
1349 pps_id = get_ue_golomb(&h->gb);
1350 if (pps_id >= MAX_PPS_COUNT) {
1351 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1352 return AVERROR_INVALIDDATA;
1354 if (!h0->pps_buffers[pps_id]) {
1355 av_log(h->avctx, AV_LOG_ERROR,
1356 "non-existing PPS %u referenced\n",
1358 return AVERROR_INVALIDDATA;
1360 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1361 av_log(h->avctx, AV_LOG_ERROR,
1362 "PPS change from %d to %d forbidden\n",
1363 h0->au_pps_id, pps_id);
1364 return AVERROR_INVALIDDATA;
1367 pps = h0->pps_buffers[pps_id];
1369 if (!h0->sps_buffers[pps->sps_id]) {
1370 av_log(h->avctx, AV_LOG_ERROR,
1371 "non-existing SPS %u referenced\n",
1373 return AVERROR_INVALIDDATA;
1376 h->pps = *h0->pps_buffers[pps_id];
1378 if (pps->sps_id != h->sps.sps_id ||
1379 pps->sps_id != h->current_sps_id ||
1380 h0->sps_buffers[pps->sps_id]->new) {
1383 av_log(h->avctx, AV_LOG_ERROR,
1384 "SPS changed in the middle of the frame\n");
1385 return AVERROR_INVALIDDATA;
1388 h->sps = *h0->sps_buffers[h->pps.sps_id];
1390 if (h->mb_width != h->sps.mb_width ||
1391 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1392 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1393 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1397 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1398 h->chroma_format_idc != h->sps.chroma_format_idc) {
1399 h->bit_depth_luma = h->sps.bit_depth_luma;
1400 h->chroma_format_idc = h->sps.chroma_format_idc;
1403 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1407 h->avctx->profile = ff_h264_get_profile(&h->sps);
1408 h->avctx->level = h->sps.level_idc;
1409 h->avctx->refs = h->sps.ref_frame_count;
1411 must_reinit = (h->context_initialized &&
1412 ( 16*h->sps.mb_width != h->avctx->coded_width
1413 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1414 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1415 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1416 || h->mb_width != h->sps.mb_width
1417 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1419 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1422 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1425 h->mb_width = h->sps.mb_width;
1426 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1427 h->mb_num = h->mb_width * h->mb_height;
1428 h->mb_stride = h->mb_width + 1;
1430 h->b_stride = h->mb_width * 4;
1432 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1434 h->width = 16 * h->mb_width;
1435 h->height = 16 * h->mb_height;
1437 ret = init_dimensions(h);
1441 if (h->sps.video_signal_type_present_flag) {
1442 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1444 if (h->sps.colour_description_present_flag) {
1445 if (h->avctx->colorspace != h->sps.colorspace)
1447 h->avctx->color_primaries = h->sps.color_primaries;
1448 h->avctx->color_trc = h->sps.color_trc;
1449 h->avctx->colorspace = h->sps.colorspace;
1453 if (h->context_initialized &&
1454 (must_reinit || needs_reinit)) {
1456 av_log(h->avctx, AV_LOG_ERROR,
1457 "changing width %d -> %d / height %d -> %d on "
1459 h->width, h->avctx->coded_width,
1460 h->height, h->avctx->coded_height,
1461 h0->current_slice + 1);
1462 return AVERROR_INVALIDDATA;
1465 av_assert1(first_slice);
1467 ff_h264_flush_change(h);
1469 if ((ret = get_pixel_format(h, 1)) < 0)
1471 h->avctx->pix_fmt = ret;
1473 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1474 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1476 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1477 av_log(h->avctx, AV_LOG_ERROR,
1478 "h264_slice_header_init() failed\n");
1482 if (!h->context_initialized) {
1484 av_log(h->avctx, AV_LOG_ERROR,
1485 "Cannot (re-)initialize context during parallel decoding.\n");
1486 return AVERROR_PATCHWELCOME;
1489 if ((ret = get_pixel_format(h, 1)) < 0)
1491 h->avctx->pix_fmt = ret;
1493 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1494 av_log(h->avctx, AV_LOG_ERROR,
1495 "h264_slice_header_init() failed\n");
1500 if (first_slice && h->dequant_coeff_pps != pps_id) {
1501 h->dequant_coeff_pps = pps_id;
1502 ff_h264_init_dequant_tables(h);
1505 frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1507 if (h0->frame_num != frame_num) {
1508 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1509 h0->frame_num, frame_num);
1510 return AVERROR_INVALIDDATA;
1515 h->mb_aff_frame = 0;
1516 last_pic_structure = h0->picture_structure;
1517 last_pic_droppable = h0->droppable;
1518 droppable = h->nal_ref_idc == 0;
1519 if (h->sps.frame_mbs_only_flag) {
1520 picture_structure = PICT_FRAME;
1522 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1523 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1526 field_pic_flag = get_bits1(&h->gb);
1528 if (field_pic_flag) {
1529 bottom_field_flag = get_bits1(&h->gb);
1530 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1532 picture_structure = PICT_FRAME;
1533 h->mb_aff_frame = h->sps.mb_aff;
1536 if (h0->current_slice) {
1537 if (last_pic_structure != picture_structure ||
1538 last_pic_droppable != droppable) {
1539 av_log(h->avctx, AV_LOG_ERROR,
1540 "Changing field mode (%d -> %d) between slices is not allowed\n",
1541 last_pic_structure, h->picture_structure);
1542 return AVERROR_INVALIDDATA;
1543 } else if (!h0->cur_pic_ptr) {
1544 av_log(h->avctx, AV_LOG_ERROR,
1545 "unset cur_pic_ptr on slice %d\n",
1546 h0->current_slice + 1);
1547 return AVERROR_INVALIDDATA;
1551 h->picture_structure = picture_structure;
1552 h->droppable = droppable;
1553 h->frame_num = frame_num;
1554 h->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1556 if (h0->current_slice == 0) {
1557 /* Shorten frame num gaps so we don't have to allocate reference
1558 * frames just to throw them away */
1559 if (h->frame_num != h->prev_frame_num) {
1560 int unwrap_prev_frame_num = h->prev_frame_num;
1561 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1563 if (unwrap_prev_frame_num > h->frame_num)
1564 unwrap_prev_frame_num -= max_frame_num;
1566 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1567 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1568 if (unwrap_prev_frame_num < 0)
1569 unwrap_prev_frame_num += max_frame_num;
1571 h->prev_frame_num = unwrap_prev_frame_num;
1575 /* See if we have a decoded first field looking for a pair...
1576 * Here, we're using that to see if we should mark previously
1577 * decode frames as "finished".
1578 * We have to do that before the "dummy" in-between frame allocation,
1579 * since that can modify h->cur_pic_ptr. */
1580 if (h0->first_field) {
1581 assert(h0->cur_pic_ptr);
1582 assert(h0->cur_pic_ptr->f.buf[0]);
1583 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1585 /* Mark old field/frame as completed */
1586 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1587 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1588 last_pic_structure == PICT_BOTTOM_FIELD);
1591 /* figure out if we have a complementary field pair */
1592 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1593 /* Previous field is unmatched. Don't display it, but let it
1594 * remain for reference if marked as such. */
1595 if (last_pic_structure != PICT_FRAME) {
1596 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1597 last_pic_structure == PICT_TOP_FIELD);
1600 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1601 /* This and previous field were reference, but had
1602 * different frame_nums. Consider this field first in
1603 * pair. Throw away previous field except for reference
1605 if (last_pic_structure != PICT_FRAME) {
1606 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1607 last_pic_structure == PICT_TOP_FIELD);
1610 /* Second field in complementary pair */
1611 if (!((last_pic_structure == PICT_TOP_FIELD &&
1612 h->picture_structure == PICT_BOTTOM_FIELD) ||
1613 (last_pic_structure == PICT_BOTTOM_FIELD &&
1614 h->picture_structure == PICT_TOP_FIELD))) {
1615 av_log(h->avctx, AV_LOG_ERROR,
1616 "Invalid field mode combination %d/%d\n",
1617 last_pic_structure, h->picture_structure);
1618 h->picture_structure = last_pic_structure;
1619 h->droppable = last_pic_droppable;
1620 return AVERROR_INVALIDDATA;
1621 } else if (last_pic_droppable != h->droppable) {
1622 avpriv_request_sample(h->avctx,
1623 "Found reference and non-reference fields in the same frame, which");
1624 h->picture_structure = last_pic_structure;
1625 h->droppable = last_pic_droppable;
1626 return AVERROR_PATCHWELCOME;
1632 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1633 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1634 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1635 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1636 h->frame_num, h->prev_frame_num);
1637 if (!h->sps.gaps_in_frame_num_allowed_flag)
1638 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1639 h->last_pocs[i] = INT_MIN;
1640 ret = h264_frame_start(h);
1642 h0->first_field = 0;
1646 h->prev_frame_num++;
1647 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1648 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1649 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1650 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1651 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1652 ret = ff_generate_sliding_window_mmcos(h, 1);
1653 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1655 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1656 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1658 /* Error concealment: If a ref is missing, copy the previous ref
1660 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1661 * many assumptions about there being no actual duplicates.
1662 * FIXME: This does not copy padding for out-of-frame motion
1663 * vectors. Given we are concealing a lost frame, this probably
1664 * is not noticeable by comparison, but it should be fixed. */
1665 if (h->short_ref_count) {
1667 av_image_copy(h->short_ref[0]->f.data,
1668 h->short_ref[0]->f.linesize,
1669 (const uint8_t **)prev->f.data,
1674 h->short_ref[0]->poc = prev->poc + 2;
1676 h->short_ref[0]->frame_num = h->prev_frame_num;
1680 /* See if we have a decoded first field looking for a pair...
1681 * We're using that to see whether to continue decoding in that
1682 * frame, or to allocate a new one. */
1683 if (h0->first_field) {
1684 assert(h0->cur_pic_ptr);
1685 assert(h0->cur_pic_ptr->f.buf[0]);
1686 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1688 /* figure out if we have a complementary field pair */
1689 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1690 /* Previous field is unmatched. Don't display it, but let it
1691 * remain for reference if marked as such. */
1692 h0->missing_fields ++;
1693 h0->cur_pic_ptr = NULL;
1694 h0->first_field = FIELD_PICTURE(h);
1696 h0->missing_fields = 0;
1697 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1698 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1699 h0->picture_structure==PICT_BOTTOM_FIELD);
1700 /* This and the previous field had different frame_nums.
1701 * Consider this field first in pair. Throw away previous
1702 * one except for reference purposes. */
1703 h0->first_field = 1;
1704 h0->cur_pic_ptr = NULL;
1706 /* Second field in complementary pair */
1707 h0->first_field = 0;
1711 /* Frame or first field in a potentially complementary pair */
1712 h0->first_field = FIELD_PICTURE(h);
1715 if (!FIELD_PICTURE(h) || h0->first_field) {
1716 if (h264_frame_start(h) < 0) {
1717 h0->first_field = 0;
1718 return AVERROR_INVALIDDATA;
1721 release_unused_pictures(h, 0);
1723 /* Some macroblocks can be accessed before they're available in case
1724 * of lost slices, MBAFF or threading. */
1725 if (FIELD_PICTURE(h)) {
1726 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1727 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1729 memset(h->slice_table, -1,
1730 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1732 h0->last_slice_type = -1;
1734 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1737 /* can't be in alloc_tables because linesize isn't known there.
1738 * FIXME: redo bipred weight to not require extra buffer? */
1739 for (i = 0; i < h->slice_context_count; i++)
1740 if (h->thread_context[i]) {
1741 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1746 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1748 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1749 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1750 first_mb_in_slice >= h->mb_num) {
1751 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1752 return AVERROR_INVALIDDATA;
1754 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1755 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1756 FIELD_OR_MBAFF_PICTURE(h);
1757 if (h->picture_structure == PICT_BOTTOM_FIELD)
1758 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1759 av_assert1(h->mb_y < h->mb_height);
1761 if (h->picture_structure == PICT_FRAME) {
1762 h->curr_pic_num = h->frame_num;
1763 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1765 h->curr_pic_num = 2 * h->frame_num + 1;
1766 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1769 if (h->nal_unit_type == NAL_IDR_SLICE)
1770 get_ue_golomb(&h->gb); /* idr_pic_id */
1772 if (h->sps.poc_type == 0) {
1773 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1775 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1776 h->delta_poc_bottom = get_se_golomb(&h->gb);
1779 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1780 h->delta_poc[0] = get_se_golomb(&h->gb);
1782 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1783 h->delta_poc[1] = get_se_golomb(&h->gb);
1786 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1788 if (h->pps.redundant_pic_cnt_present)
1789 h->redundant_pic_count = get_ue_golomb(&h->gb);
1791 ret = ff_set_ref_count(h);
1795 if (slice_type != AV_PICTURE_TYPE_I &&
1796 (h0->current_slice == 0 ||
1797 slice_type != h0->last_slice_type ||
1798 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
1800 ff_h264_fill_default_ref_list(h);
1803 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1804 ret = ff_h264_decode_ref_pic_list_reordering(h);
1806 h->ref_count[1] = h->ref_count[0] = 0;
1811 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1812 (h->pps.weighted_bipred_idc == 1 &&
1813 h->slice_type_nos == AV_PICTURE_TYPE_B))
1814 ff_pred_weight_table(h);
1815 else if (h->pps.weighted_bipred_idc == 2 &&
1816 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1817 implicit_weight_table(h, -1);
1820 for (i = 0; i < 2; i++) {
1821 h->luma_weight_flag[i] = 0;
1822 h->chroma_weight_flag[i] = 0;
1826 // If frame-mt is enabled, only update mmco tables for the first slice
1827 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1828 // or h->mmco, which will cause ref list mix-ups and decoding errors
1829 // further down the line. This may break decoding if the first slice is
1830 // corrupt, thus we only do this if frame-mt is enabled.
1831 if (h->nal_ref_idc) {
1832 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1833 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1834 h0->current_slice == 0);
1835 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1836 return AVERROR_INVALIDDATA;
1839 if (FRAME_MBAFF(h)) {
1840 ff_h264_fill_mbaff_ref_list(h);
1842 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1843 implicit_weight_table(h, 0);
1844 implicit_weight_table(h, 1);
1848 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1849 ff_h264_direct_dist_scale_factor(h);
1850 ff_h264_direct_ref_list_init(h);
1852 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1853 tmp = get_ue_golomb_31(&h->gb);
1855 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1856 return AVERROR_INVALIDDATA;
1858 h->cabac_init_idc = tmp;
1861 h->last_qscale_diff = 0;
1862 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1863 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1864 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1865 return AVERROR_INVALIDDATA;
1868 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1869 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1870 // FIXME qscale / qp ... stuff
1871 if (h->slice_type == AV_PICTURE_TYPE_SP)
1872 get_bits1(&h->gb); /* sp_for_switch_flag */
1873 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1874 h->slice_type == AV_PICTURE_TYPE_SI)
1875 get_se_golomb(&h->gb); /* slice_qs_delta */
1877 h->deblocking_filter = 1;
1878 h->slice_alpha_c0_offset = 0;
1879 h->slice_beta_offset = 0;
1880 if (h->pps.deblocking_filter_parameters_present) {
1881 tmp = get_ue_golomb_31(&h->gb);
1883 av_log(h->avctx, AV_LOG_ERROR,
1884 "deblocking_filter_idc %u out of range\n", tmp);
1885 return AVERROR_INVALIDDATA;
1887 h->deblocking_filter = tmp;
1888 if (h->deblocking_filter < 2)
1889 h->deblocking_filter ^= 1; // 1<->0
1891 if (h->deblocking_filter) {
1892 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1893 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1894 if (h->slice_alpha_c0_offset > 12 ||
1895 h->slice_alpha_c0_offset < -12 ||
1896 h->slice_beta_offset > 12 ||
1897 h->slice_beta_offset < -12) {
1898 av_log(h->avctx, AV_LOG_ERROR,
1899 "deblocking filter parameters %d %d out of range\n",
1900 h->slice_alpha_c0_offset, h->slice_beta_offset);
1901 return AVERROR_INVALIDDATA;
1906 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1907 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1908 h->nal_unit_type != NAL_IDR_SLICE) ||
1909 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1910 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1911 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1912 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1913 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1914 h->nal_ref_idc == 0))
1915 h->deblocking_filter = 0;
1917 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1918 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1919 /* Cheat slightly for speed:
1920 * Do not bother to deblock across slices. */
1921 h->deblocking_filter = 2;
1923 h0->max_contexts = 1;
1924 if (!h0->single_decode_warning) {
1925 av_log(h->avctx, AV_LOG_INFO,
1926 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1927 "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"
1928 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1929 "but will generate non-standard-compliant output.\n");
1930 h0->single_decode_warning = 1;
1933 av_log(h->avctx, AV_LOG_ERROR,
1934 "Deblocking switched inside frame.\n");
1935 return SLICE_SINGLETHREAD;
1940 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1942 h->pps.chroma_qp_index_offset[0],
1943 h->pps.chroma_qp_index_offset[1]) +
1944 6 * (h->sps.bit_depth_luma - 8);
1946 h0->last_slice_type = slice_type;
1947 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
1948 h->slice_num = ++h0->current_slice;
1951 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
1952 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
1953 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
1954 && h->slice_num >= MAX_SLICES) {
1955 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1956 av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
1959 for (j = 0; j < 2; j++) {
1961 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1962 for (i = 0; i < 16; i++) {
1964 if (j < h->list_count && i < h->ref_count[j] &&
1965 h->ref_list[j][i].f.buf[0]) {
1967 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1968 for (k = 0; k < h->short_ref_count; k++)
1969 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1973 for (k = 0; k < h->long_ref_count; k++)
1974 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1975 id_list[i] = h->short_ref_count + k;
1983 for (i = 0; i < 16; i++)
1984 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1986 ref2frm[18 + 1] = -1;
1987 for (i = 16; i < 48; i++)
1988 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1989 (h->ref_list[j][i].reference & 3);
1992 h0->au_pps_id = pps_id;
1994 h0->sps_buffers[h->pps.sps_id]->new = 0;
1995 h->current_sps_id = h->pps.sps_id;
1997 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1998 av_log(h->avctx, AV_LOG_DEBUG,
1999 "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",
2001 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
2003 av_get_picture_type_char(h->slice_type),
2004 h->slice_type_fixed ? " fix" : "",
2005 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2006 pps_id, h->frame_num,
2007 h->cur_pic_ptr->field_poc[0],
2008 h->cur_pic_ptr->field_poc[1],
2009 h->ref_count[0], h->ref_count[1],
2011 h->deblocking_filter,
2012 h->slice_alpha_c0_offset, h->slice_beta_offset,
2014 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
2015 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2021 int ff_h264_get_slice_type(const H264Context *h)
2023 switch (h->slice_type) {
2024 case AV_PICTURE_TYPE_P:
2026 case AV_PICTURE_TYPE_B:
2028 case AV_PICTURE_TYPE_I:
2030 case AV_PICTURE_TYPE_SP:
2032 case AV_PICTURE_TYPE_SI:
2035 return AVERROR_INVALIDDATA;
2039 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2040 int mb_type, int top_xy,
2041 int left_xy[LEFT_MBS],
2043 int left_type[LEFT_MBS],
2044 int mb_xy, int list)
2046 int b_stride = h->b_stride;
2047 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
2048 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
2049 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2050 if (USES_LIST(top_type, list)) {
2051 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2052 const int b8_xy = 4 * top_xy + 2;
2053 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2054 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2055 ref_cache[0 - 1 * 8] =
2056 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2057 ref_cache[2 - 1 * 8] =
2058 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2060 AV_ZERO128(mv_dst - 1 * 8);
2061 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2064 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2065 if (USES_LIST(left_type[LTOP], list)) {
2066 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2067 const int b8_xy = 4 * left_xy[LTOP] + 1;
2068 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2069 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2070 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2071 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2072 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2074 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2075 ref_cache[-1 + 16] =
2076 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2078 AV_ZERO32(mv_dst - 1 + 0);
2079 AV_ZERO32(mv_dst - 1 + 8);
2080 AV_ZERO32(mv_dst - 1 + 16);
2081 AV_ZERO32(mv_dst - 1 + 24);
2084 ref_cache[-1 + 16] =
2085 ref_cache[-1 + 24] = LIST_NOT_USED;
2090 if (!USES_LIST(mb_type, list)) {
2091 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2092 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2093 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2094 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2095 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2100 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2101 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2102 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2103 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2104 AV_WN32A(&ref_cache[0 * 8], ref01);
2105 AV_WN32A(&ref_cache[1 * 8], ref01);
2106 AV_WN32A(&ref_cache[2 * 8], ref23);
2107 AV_WN32A(&ref_cache[3 * 8], ref23);
2111 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
2112 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2113 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2114 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2115 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2121 * @return non zero if the loop filter can be skipped
2123 static int fill_filter_caches(H264Context *h, int mb_type)
2125 const int mb_xy = h->mb_xy;
2126 int top_xy, left_xy[LEFT_MBS];
2127 int top_type, left_type[LEFT_MBS];
2131 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
2133 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2134 * stuff, I can't imagine that these complex rules are worth it. */
2136 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2137 if (FRAME_MBAFF(h)) {
2138 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2139 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2141 if (left_mb_field_flag != curr_mb_field_flag)
2142 left_xy[LTOP] -= h->mb_stride;
2144 if (curr_mb_field_flag)
2145 top_xy += h->mb_stride &
2146 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2147 if (left_mb_field_flag != curr_mb_field_flag)
2148 left_xy[LBOT] += h->mb_stride;
2152 h->top_mb_xy = top_xy;
2153 h->left_mb_xy[LTOP] = left_xy[LTOP];
2154 h->left_mb_xy[LBOT] = left_xy[LBOT];
2156 /* For sufficiently low qp, filtering wouldn't do anything.
2157 * This is a conservative estimate: could also check beta_offset
2158 * and more accurate chroma_qp. */
2159 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2160 int qp = h->cur_pic.qscale_table[mb_xy];
2161 if (qp <= qp_thresh &&
2162 (left_xy[LTOP] < 0 ||
2163 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2165 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2166 if (!FRAME_MBAFF(h))
2168 if ((left_xy[LTOP] < 0 ||
2169 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2170 (top_xy < h->mb_stride ||
2171 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2176 top_type = h->cur_pic.mb_type[top_xy];
2177 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2178 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2179 if (h->deblocking_filter == 2) {
2180 if (h->slice_table[top_xy] != h->slice_num)
2182 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
2183 left_type[LTOP] = left_type[LBOT] = 0;
2185 if (h->slice_table[top_xy] == 0xFFFF)
2187 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2188 left_type[LTOP] = left_type[LBOT] = 0;
2190 h->top_type = top_type;
2191 h->left_type[LTOP] = left_type[LTOP];
2192 h->left_type[LBOT] = left_type[LBOT];
2194 if (IS_INTRA(mb_type))
2197 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2198 top_type, left_type, mb_xy, 0);
2199 if (h->list_count == 2)
2200 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2201 top_type, left_type, mb_xy, 1);
2203 nnz = h->non_zero_count[mb_xy];
2204 nnz_cache = h->non_zero_count_cache;
2205 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2206 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2207 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2208 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2209 h->cbp = h->cbp_table[mb_xy];
2212 nnz = h->non_zero_count[top_xy];
2213 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2216 if (left_type[LTOP]) {
2217 nnz = h->non_zero_count[left_xy[LTOP]];
2218 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2219 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2220 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2221 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2224 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2225 * from what the loop filter needs */
2226 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2227 if (IS_8x8DCT(top_type)) {
2228 nnz_cache[4 + 8 * 0] =
2229 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2230 nnz_cache[6 + 8 * 0] =
2231 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2233 if (IS_8x8DCT(left_type[LTOP])) {
2234 nnz_cache[3 + 8 * 1] =
2235 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2237 if (IS_8x8DCT(left_type[LBOT])) {
2238 nnz_cache[3 + 8 * 3] =
2239 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2242 if (IS_8x8DCT(mb_type)) {
2243 nnz_cache[scan8[0]] =
2244 nnz_cache[scan8[1]] =
2245 nnz_cache[scan8[2]] =
2246 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2248 nnz_cache[scan8[0 + 4]] =
2249 nnz_cache[scan8[1 + 4]] =
2250 nnz_cache[scan8[2 + 4]] =
2251 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2253 nnz_cache[scan8[0 + 8]] =
2254 nnz_cache[scan8[1 + 8]] =
2255 nnz_cache[scan8[2 + 8]] =
2256 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2258 nnz_cache[scan8[0 + 12]] =
2259 nnz_cache[scan8[1 + 12]] =
2260 nnz_cache[scan8[2 + 12]] =
2261 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2268 static void loop_filter(H264Context *h, int start_x, int end_x)
2270 uint8_t *dest_y, *dest_cb, *dest_cr;
2271 int linesize, uvlinesize, mb_x, mb_y;
2272 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2273 const int old_slice_type = h->slice_type;
2274 const int pixel_shift = h->pixel_shift;
2275 const int block_h = 16 >> h->chroma_y_shift;
2277 if (h->deblocking_filter) {
2278 for (mb_x = start_x; mb_x < end_x; mb_x++)
2279 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2281 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2282 h->slice_num = h->slice_table[mb_xy];
2283 mb_type = h->cur_pic.mb_type[mb_xy];
2284 h->list_count = h->list_counts[mb_xy];
2288 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2292 dest_y = h->cur_pic.f.data[0] +
2293 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2294 dest_cb = h->cur_pic.f.data[1] +
2295 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2296 mb_y * h->uvlinesize * block_h;
2297 dest_cr = h->cur_pic.f.data[2] +
2298 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2299 mb_y * h->uvlinesize * block_h;
2300 // FIXME simplify above
2303 linesize = h->mb_linesize = h->linesize * 2;
2304 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2305 if (mb_y & 1) { // FIXME move out of this function?
2306 dest_y -= h->linesize * 15;
2307 dest_cb -= h->uvlinesize * (block_h - 1);
2308 dest_cr -= h->uvlinesize * (block_h - 1);
2311 linesize = h->mb_linesize = h->linesize;
2312 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2314 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2316 if (fill_filter_caches(h, mb_type))
2318 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2319 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2321 if (FRAME_MBAFF(h)) {
2322 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2323 linesize, uvlinesize);
2325 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2326 dest_cr, linesize, uvlinesize);
2330 h->slice_type = old_slice_type;
2332 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2333 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2334 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2337 static void predict_field_decoding_flag(H264Context *h)
2339 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2340 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2341 h->cur_pic.mb_type[mb_xy - 1] :
2342 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2343 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2344 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2348 * Draw edges and report progress for the last MB row.
2350 static void decode_finish_row(H264Context *h)
2352 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2353 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2354 int height = 16 << FRAME_MBAFF(h);
2355 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2357 if (h->deblocking_filter) {
2358 if ((top + height) >= pic_height)
2359 height += deblock_border;
2360 top -= deblock_border;
2363 if (top >= pic_height || (top + height) < 0)
2366 height = FFMIN(height, pic_height - top);
2368 height = top + height;
2372 ff_h264_draw_horiz_band(h, top, height);
2374 if (h->droppable || h->er.error_occurred)
2377 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2378 h->picture_structure == PICT_BOTTOM_FIELD);
2381 static void er_add_slice(H264Context *h, int startx, int starty,
2382 int endx, int endy, int status)
2384 if (CONFIG_ERROR_RESILIENCE) {
2385 ERContext *er = &h->er;
2387 ff_er_add_slice(er, startx, starty, endx, endy, status);
2391 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2393 H264Context *h = *(void **)arg;
2394 int lf_x_start = h->mb_x;
2396 h->mb_skip_run = -1;
2398 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2400 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2401 avctx->codec_id != AV_CODEC_ID_H264 ||
2402 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2404 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2405 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2407 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2408 prev_status &= ~ VP_START;
2409 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2410 h->er.error_occurred = 1;
2416 align_get_bits(&h->gb);
2419 ff_init_cabac_decoder(&h->cabac,
2420 h->gb.buffer + get_bits_count(&h->gb) / 8,
2421 (get_bits_left(&h->gb) + 7) / 8);
2423 ff_h264_init_cabac_states(h);
2427 int ret = ff_h264_decode_mb_cabac(h);
2429 // STOP_TIMER("decode_mb_cabac")
2432 ff_h264_hl_decode_mb(h);
2434 // FIXME optimal? or let mb_decode decode 16x32 ?
2435 if (ret >= 0 && FRAME_MBAFF(h)) {
2438 ret = ff_h264_decode_mb_cabac(h);
2441 ff_h264_hl_decode_mb(h);
2444 eos = get_cabac_terminate(&h->cabac);
2446 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2447 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2448 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2449 h->mb_y, ER_MB_END);
2450 if (h->mb_x >= lf_x_start)
2451 loop_filter(h, lf_x_start, h->mb_x + 1);
2454 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
2455 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", h->cabac.bytestream_end - h->cabac.bytestream);
2456 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
2457 av_log(h->avctx, AV_LOG_ERROR,
2458 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2460 h->cabac.bytestream_end - h->cabac.bytestream);
2461 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2462 h->mb_y, ER_MB_ERROR);
2463 return AVERROR_INVALIDDATA;
2466 if (++h->mb_x >= h->mb_width) {
2467 loop_filter(h, lf_x_start, h->mb_x);
2468 h->mb_x = lf_x_start = 0;
2469 decode_finish_row(h);
2471 if (FIELD_OR_MBAFF_PICTURE(h)) {
2473 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2474 predict_field_decoding_flag(h);
2478 if (eos || h->mb_y >= h->mb_height) {
2479 tprintf(h->avctx, "slice end %d %d\n",
2480 get_bits_count(&h->gb), h->gb.size_in_bits);
2481 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2482 h->mb_y, ER_MB_END);
2483 if (h->mb_x > lf_x_start)
2484 loop_filter(h, lf_x_start, h->mb_x);
2490 int ret = ff_h264_decode_mb_cavlc(h);
2493 ff_h264_hl_decode_mb(h);
2495 // FIXME optimal? or let mb_decode decode 16x32 ?
2496 if (ret >= 0 && FRAME_MBAFF(h)) {
2498 ret = ff_h264_decode_mb_cavlc(h);
2501 ff_h264_hl_decode_mb(h);
2506 av_log(h->avctx, AV_LOG_ERROR,
2507 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2508 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2509 h->mb_y, ER_MB_ERROR);
2513 if (++h->mb_x >= h->mb_width) {
2514 loop_filter(h, lf_x_start, h->mb_x);
2515 h->mb_x = lf_x_start = 0;
2516 decode_finish_row(h);
2518 if (FIELD_OR_MBAFF_PICTURE(h)) {
2520 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2521 predict_field_decoding_flag(h);
2523 if (h->mb_y >= h->mb_height) {
2524 tprintf(h->avctx, "slice end %d %d\n",
2525 get_bits_count(&h->gb), h->gb.size_in_bits);
2527 if ( get_bits_left(&h->gb) == 0
2528 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2529 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2530 h->mb_x - 1, h->mb_y, ER_MB_END);
2534 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2535 h->mb_x, h->mb_y, ER_MB_END);
2537 return AVERROR_INVALIDDATA;
2542 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2543 tprintf(h->avctx, "slice end %d %d\n",
2544 get_bits_count(&h->gb), h->gb.size_in_bits);
2546 if (get_bits_left(&h->gb) == 0) {
2547 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2548 h->mb_x - 1, h->mb_y, ER_MB_END);
2549 if (h->mb_x > lf_x_start)
2550 loop_filter(h, lf_x_start, h->mb_x);
2554 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2555 h->mb_y, ER_MB_ERROR);
2557 return AVERROR_INVALIDDATA;
2565 * Call decode_slice() for each context.
2567 * @param h h264 master context
2568 * @param context_count number of contexts to execute
2570 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2572 AVCodecContext *const avctx = h->avctx;
2576 av_assert0(h->mb_y < h->mb_height);
2578 if (h->avctx->hwaccel ||
2579 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2581 if (context_count == 1) {
2582 return decode_slice(avctx, &h);
2584 av_assert0(context_count > 0);
2585 for (i = 1; i < context_count; i++) {
2586 hx = h->thread_context[i];
2587 if (CONFIG_ERROR_RESILIENCE) {
2588 hx->er.error_count = 0;
2590 hx->x264_build = h->x264_build;
2593 avctx->execute(avctx, decode_slice, h->thread_context,
2594 NULL, context_count, sizeof(void *));
2596 /* pull back stuff from slices to master context */
2597 hx = h->thread_context[context_count - 1];
2600 h->droppable = hx->droppable;
2601 h->picture_structure = hx->picture_structure;
2602 if (CONFIG_ERROR_RESILIENCE) {
2603 for (i = 1; i < context_count; i++)
2604 h->er.error_count += h->thread_context[i]->er.error_count;