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(H264SliceContext *sl, int linesize)
166 const H264Context *h = sl->h264;
167 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
169 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
170 // edge emu needs blocksize + filter length - 1
171 // (= 21x21 for h264)
172 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
174 av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
175 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
176 av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
177 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
179 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
180 !sl->top_borders[0] || !sl->top_borders[1]) {
181 av_freep(&sl->bipred_scratchpad);
182 av_freep(&sl->edge_emu_buffer);
183 av_freep(&sl->top_borders[0]);
184 av_freep(&sl->top_borders[1]);
186 sl->bipred_scratchpad_allocated = 0;
187 sl->edge_emu_buffer_allocated = 0;
188 sl->top_borders_allocated[0] = 0;
189 sl->top_borders_allocated[1] = 0;
190 return AVERROR(ENOMEM);
196 static int init_table_pools(H264Context *h)
198 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
199 const int mb_array_size = h->mb_stride * h->mb_height;
200 const int b4_stride = h->mb_width * 4 + 1;
201 const int b4_array_size = b4_stride * h->mb_height * 4;
203 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
205 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
206 sizeof(uint32_t), av_buffer_allocz);
207 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
208 sizeof(int16_t), av_buffer_allocz);
209 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
211 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
212 !h->ref_index_pool) {
213 av_buffer_pool_uninit(&h->qscale_table_pool);
214 av_buffer_pool_uninit(&h->mb_type_pool);
215 av_buffer_pool_uninit(&h->motion_val_pool);
216 av_buffer_pool_uninit(&h->ref_index_pool);
217 return AVERROR(ENOMEM);
223 static int alloc_picture(H264Context *h, H264Picture *pic)
227 av_assert0(!pic->f.data[0]);
230 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
231 AV_GET_BUFFER_FLAG_REF : 0);
235 h->linesize = pic->f.linesize[0];
236 h->uvlinesize = pic->f.linesize[1];
237 pic->crop = h->sps.crop;
238 pic->crop_top = h->sps.crop_top;
239 pic->crop_left= h->sps.crop_left;
241 if (h->avctx->hwaccel) {
242 const AVHWAccel *hwaccel = h->avctx->hwaccel;
243 av_assert0(!pic->hwaccel_picture_private);
244 if (hwaccel->frame_priv_data_size) {
245 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
246 if (!pic->hwaccel_priv_buf)
247 return AVERROR(ENOMEM);
248 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
251 if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
252 int h_chroma_shift, v_chroma_shift;
253 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
254 &h_chroma_shift, &v_chroma_shift);
256 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
257 memset(pic->f.data[1] + pic->f.linesize[1]*i,
258 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
259 memset(pic->f.data[2] + pic->f.linesize[2]*i,
260 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
264 if (!h->qscale_table_pool) {
265 ret = init_table_pools(h);
270 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
271 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
272 if (!pic->qscale_table_buf || !pic->mb_type_buf)
275 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
276 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
278 for (i = 0; i < 2; i++) {
279 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
280 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
281 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
284 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
285 pic->ref_index[i] = pic->ref_index_buf[i]->data;
290 ff_h264_unref_picture(h, pic);
291 return (ret < 0) ? ret : AVERROR(ENOMEM);
294 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
298 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
303 static int find_unused_picture(H264Context *h)
307 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
308 if (pic_is_unused(h, &h->DPB[i]))
311 if (i == H264_MAX_PICTURE_COUNT)
312 return AVERROR_INVALIDDATA;
314 if (h->DPB[i].needs_realloc) {
315 h->DPB[i].needs_realloc = 0;
316 ff_h264_unref_picture(h, &h->DPB[i]);
323 static void init_dequant8_coeff_table(H264Context *h)
326 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
328 for (i = 0; i < 6; i++) {
329 h->dequant8_coeff[i] = h->dequant8_buffer[i];
330 for (j = 0; j < i; j++)
331 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
332 64 * sizeof(uint8_t))) {
333 h->dequant8_coeff[i] = h->dequant8_buffer[j];
339 for (q = 0; q < max_qp + 1; q++) {
342 for (x = 0; x < 64; x++)
343 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
344 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
345 h->pps.scaling_matrix8[i][x]) << shift;
350 static void init_dequant4_coeff_table(H264Context *h)
353 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
354 for (i = 0; i < 6; i++) {
355 h->dequant4_coeff[i] = h->dequant4_buffer[i];
356 for (j = 0; j < i; j++)
357 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
358 16 * sizeof(uint8_t))) {
359 h->dequant4_coeff[i] = h->dequant4_buffer[j];
365 for (q = 0; q < max_qp + 1; q++) {
366 int shift = div6[q] + 2;
368 for (x = 0; x < 16; x++)
369 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
370 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
371 h->pps.scaling_matrix4[i][x]) << shift;
376 void ff_h264_init_dequant_tables(H264Context *h)
379 init_dequant4_coeff_table(h);
380 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
382 if (h->pps.transform_8x8_mode)
383 init_dequant8_coeff_table(h);
384 if (h->sps.transform_bypass) {
385 for (i = 0; i < 6; i++)
386 for (x = 0; x < 16; x++)
387 h->dequant4_coeff[i][0][x] = 1 << 6;
388 if (h->pps.transform_8x8_mode)
389 for (i = 0; i < 6; i++)
390 for (x = 0; x < 64; x++)
391 h->dequant8_coeff[i][0][x] = 1 << 6;
396 * Mimic alloc_tables(), but for every context thread.
398 static void clone_tables(H264Context *dst, H264SliceContext *sl,
399 H264Context *src, int i)
401 sl->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
402 sl->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
403 sl->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
405 dst->non_zero_count = src->non_zero_count;
406 dst->slice_table = src->slice_table;
407 dst->cbp_table = src->cbp_table;
408 dst->mb2b_xy = src->mb2b_xy;
409 dst->mb2br_xy = src->mb2br_xy;
410 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
411 dst->direct_table = src->direct_table;
412 dst->list_counts = src->list_counts;
414 dst->cur_pic_ptr = src->cur_pic_ptr;
415 dst->cur_pic = src->cur_pic;
416 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
417 src->sps.chroma_format_idc);
420 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
422 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
423 (((pic) && (pic) >= (old_ctx)->DPB && \
424 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
425 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
427 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
428 H264Context *new_base,
429 H264Context *old_base)
433 for (i = 0; i < count; i++) {
434 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
435 IN_RANGE(from[i], old_base->DPB,
436 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
438 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
442 static int copy_parameter_set(void **to, void **from, int count, int size)
446 for (i = 0; i < count; i++) {
447 if (to[i] && !from[i]) {
449 } else if (from[i] && !to[i]) {
450 to[i] = av_malloc(size);
452 return AVERROR(ENOMEM);
456 memcpy(to[i], from[i], size);
462 #define copy_fields(to, from, start_field, end_field) \
463 memcpy(&(to)->start_field, &(from)->start_field, \
464 (char *)&(to)->end_field - (char *)&(to)->start_field)
466 static int h264_slice_header_init(H264Context *h, int reinit);
468 int ff_h264_update_thread_context(AVCodecContext *dst,
469 const AVCodecContext *src)
471 H264Context *h = dst->priv_data, *h1 = src->priv_data;
472 int inited = h->context_initialized, err = 0;
473 int context_reinitialized = 0;
480 (h->width != h1->width ||
481 h->height != h1->height ||
482 h->mb_width != h1->mb_width ||
483 h->mb_height != h1->mb_height ||
484 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
485 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
486 h->sps.colorspace != h1->sps.colorspace)) {
488 /* set bits_per_raw_sample to the previous value. the check for changed
489 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
490 * the current value */
491 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
493 h->width = h1->width;
494 h->height = h1->height;
495 h->mb_height = h1->mb_height;
496 h->mb_width = h1->mb_width;
497 h->mb_num = h1->mb_num;
498 h->mb_stride = h1->mb_stride;
499 h->b_stride = h1->b_stride;
501 if ((ret = copy_parameter_set((void **)h->sps_buffers,
502 (void **)h1->sps_buffers,
503 MAX_SPS_COUNT, sizeof(SPS))) < 0)
506 if ((ret = copy_parameter_set((void **)h->pps_buffers,
507 (void **)h1->pps_buffers,
508 MAX_PPS_COUNT, sizeof(PPS))) < 0)
512 if ((err = h264_slice_header_init(h, 1)) < 0) {
513 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
516 context_reinitialized = 1;
519 h264_set_parameter_from_sps(h);
520 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
521 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
524 /* update linesize on resize for h264. The h264 decoder doesn't
525 * necessarily call ff_mpv_frame_start in the new thread */
526 h->linesize = h1->linesize;
527 h->uvlinesize = h1->uvlinesize;
529 /* copy block_offset since frame_start may not be called */
530 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
533 H264SliceContext *orig_slice_ctx = h->slice_ctx;
535 for (i = 0; i < MAX_SPS_COUNT; i++)
536 av_freep(h->sps_buffers + i);
538 for (i = 0; i < MAX_PPS_COUNT; i++)
539 av_freep(h->pps_buffers + i);
541 ff_h264_unref_picture(h, &h->last_pic_for_ec);
542 memcpy(h, h1, sizeof(H264Context));
544 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
545 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
547 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
548 memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
550 h->slice_ctx = orig_slice_ctx;
552 memset(&h->slice_ctx[0].er, 0, sizeof(h->slice_ctx[0].er));
553 memset(&h->slice_ctx[0].mb, 0, sizeof(h->slice_ctx[0].mb));
554 memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
555 memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
559 h->qscale_table_pool = NULL;
560 h->mb_type_pool = NULL;
561 h->ref_index_pool = NULL;
562 h->motion_val_pool = NULL;
563 h->intra4x4_pred_mode= NULL;
564 h->non_zero_count = NULL;
565 h->slice_table_base = NULL;
566 h->slice_table = NULL;
568 h->chroma_pred_mode_table = NULL;
569 memset(h->mvd_table, 0, sizeof(h->mvd_table));
570 h->direct_table = NULL;
571 h->list_counts = NULL;
575 if (h1->context_initialized) {
576 h->context_initialized = 0;
578 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
579 av_frame_unref(&h->cur_pic.f);
580 h->cur_pic.tf.f = &h->cur_pic.f;
582 ret = ff_h264_alloc_tables(h);
584 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
587 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
589 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
594 h->thread_context[0] = h;
595 h->context_initialized = h1->context_initialized;
598 h->avctx->coded_height = h1->avctx->coded_height;
599 h->avctx->coded_width = h1->avctx->coded_width;
600 h->avctx->width = h1->avctx->width;
601 h->avctx->height = h1->avctx->height;
602 h->coded_picture_number = h1->coded_picture_number;
603 h->first_field = h1->first_field;
604 h->picture_structure = h1->picture_structure;
605 h->droppable = h1->droppable;
606 h->low_delay = h1->low_delay;
608 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
609 ff_h264_unref_picture(h, &h->DPB[i]);
610 if (h1->DPB && h1->DPB[i].f.buf[0] &&
611 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
615 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
616 ff_h264_unref_picture(h, &h->cur_pic);
617 if (h1->cur_pic.f.buf[0]) {
618 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
623 h->workaround_bugs = h1->workaround_bugs;
624 h->low_delay = h1->low_delay;
625 h->droppable = h1->droppable;
627 // extradata/NAL handling
628 h->is_avc = h1->is_avc;
631 if ((ret = copy_parameter_set((void **)h->sps_buffers,
632 (void **)h1->sps_buffers,
633 MAX_SPS_COUNT, sizeof(SPS))) < 0)
636 if ((ret = copy_parameter_set((void **)h->pps_buffers,
637 (void **)h1->pps_buffers,
638 MAX_PPS_COUNT, sizeof(PPS))) < 0)
642 // Dequantization matrices
643 // FIXME these are big - can they be only copied when PPS changes?
644 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
646 for (i = 0; i < 6; i++)
647 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
648 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
650 for (i = 0; i < 6; i++)
651 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
652 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
654 h->dequant_coeff_pps = h1->dequant_coeff_pps;
657 copy_fields(h, h1, poc_lsb, default_ref_list);
660 copy_fields(h, h1, short_ref, thread_context);
662 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
663 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
664 copy_picture_range(h->delayed_pic, h1->delayed_pic,
665 MAX_DELAYED_PIC_COUNT + 2, h, h1);
667 h->frame_recovered = h1->frame_recovered;
669 if (context_reinitialized)
670 ff_h264_set_parameter_from_sps(h);
676 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
677 h->prev_poc_msb = h->poc_msb;
678 h->prev_poc_lsb = h->poc_lsb;
680 h->prev_frame_num_offset = h->frame_num_offset;
681 h->prev_frame_num = h->frame_num;
682 h->outputed_poc = h->next_outputed_poc;
684 h->recovery_frame = h1->recovery_frame;
689 static int h264_frame_start(H264Context *h)
693 const int pixel_shift = h->pixel_shift;
695 1<<(h->sps.bit_depth_luma-1),
696 1<<(h->sps.bit_depth_chroma-1),
697 1<<(h->sps.bit_depth_chroma-1),
701 if (!ff_thread_can_start_frame(h->avctx)) {
702 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
706 release_unused_pictures(h, 1);
707 h->cur_pic_ptr = NULL;
709 i = find_unused_picture(h);
711 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
716 pic->reference = h->droppable ? 0 : h->picture_structure;
717 pic->f.coded_picture_number = h->coded_picture_number++;
718 pic->field_picture = h->picture_structure != PICT_FRAME;
721 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
723 * See decode_nal_units().
725 pic->f.key_frame = 0;
728 pic->invalid_gap = 0;
729 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
731 if ((ret = alloc_picture(h, pic)) < 0)
733 if(!h->frame_recovered && !h->avctx->hwaccel &&
734 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
735 avpriv_color_frame(&pic->f, c);
737 h->cur_pic_ptr = pic;
738 ff_h264_unref_picture(h, &h->cur_pic);
739 if (CONFIG_ERROR_RESILIENCE) {
740 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
743 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
746 if (CONFIG_ERROR_RESILIENCE) {
747 ff_er_frame_start(&h->slice_ctx[0].er);
748 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
749 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
752 assert(h->linesize && h->uvlinesize);
754 for (i = 0; i < 16; i++) {
755 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
756 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
758 for (i = 0; i < 16; i++) {
759 h->block_offset[16 + i] =
760 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
761 h->block_offset[48 + 16 + i] =
762 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
765 /* We mark the current picture as non-reference after allocating it, so
766 * that if we break out due to an error it can be released automatically
767 * in the next ff_mpv_frame_start().
769 h->cur_pic_ptr->reference = 0;
771 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
773 h->next_output_pic = NULL;
775 assert(h->cur_pic_ptr->long_ref == 0);
780 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
782 uint8_t *src_cb, uint8_t *src_cr,
783 int linesize, int uvlinesize,
788 const int pixel_shift = h->pixel_shift;
789 int chroma444 = CHROMA444(h);
790 int chroma422 = CHROMA422(h);
793 src_cb -= uvlinesize;
794 src_cr -= uvlinesize;
796 if (!simple && FRAME_MBAFF(h)) {
799 top_border = sl->top_borders[0][sl->mb_x];
800 AV_COPY128(top_border, src_y + 15 * linesize);
802 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
803 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
806 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
807 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
808 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
809 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
811 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
812 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
814 } else if (chroma422) {
816 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
817 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
819 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
820 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
824 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
825 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
827 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
828 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
833 } else if (MB_MBAFF(sl)) {
839 top_border = sl->top_borders[top_idx][sl->mb_x];
840 /* There are two lines saved, the line above the top macroblock
841 * of a pair, and the line above the bottom macroblock. */
842 AV_COPY128(top_border, src_y + 16 * linesize);
844 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
846 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
849 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
850 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
851 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
852 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
854 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
855 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
857 } else if (chroma422) {
859 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
860 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
862 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
863 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
867 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
868 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
870 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
871 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
878 * Initialize implicit_weight table.
879 * @param field 0/1 initialize the weight for interlaced MBAFF
880 * -1 initializes the rest
882 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
884 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
886 for (i = 0; i < 2; i++) {
887 sl->luma_weight_flag[i] = 0;
888 sl->chroma_weight_flag[i] = 0;
892 if (h->picture_structure == PICT_FRAME) {
893 cur_poc = h->cur_pic_ptr->poc;
895 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
897 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
898 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
900 sl->use_weight_chroma = 0;
904 ref_count0 = sl->ref_count[0];
905 ref_count1 = sl->ref_count[1];
907 cur_poc = h->cur_pic_ptr->field_poc[field];
909 ref_count0 = 16 + 2 * sl->ref_count[0];
910 ref_count1 = 16 + 2 * sl->ref_count[1];
914 sl->use_weight_chroma = 2;
915 sl->luma_log2_weight_denom = 5;
916 sl->chroma_log2_weight_denom = 5;
918 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
919 int poc0 = sl->ref_list[0][ref0].poc;
920 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
922 if (!sl->ref_list[0][ref0].long_ref && !sl->ref_list[1][ref1].long_ref) {
923 int poc1 = sl->ref_list[1][ref1].poc;
924 int td = av_clip_int8(poc1 - poc0);
926 int tb = av_clip_int8(cur_poc - poc0);
927 int tx = (16384 + (FFABS(td) >> 1)) / td;
928 int dist_scale_factor = (tb * tx + 32) >> 8;
929 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
930 w = 64 - dist_scale_factor;
934 sl->implicit_weight[ref0][ref1][0] =
935 sl->implicit_weight[ref0][ref1][1] = w;
937 sl->implicit_weight[ref0][ref1][field] = w;
944 * initialize scan tables
946 static void init_scan_tables(H264Context *h)
949 for (i = 0; i < 16; i++) {
950 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
951 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
952 h->field_scan[i] = TRANSPOSE(field_scan[i]);
955 for (i = 0; i < 64; i++) {
956 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
957 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
958 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
959 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
960 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
963 if (h->sps.transform_bypass) { // FIXME same ugly
964 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
965 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
966 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
967 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
968 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
969 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
971 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
972 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
973 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
974 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
975 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
976 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
981 * Replicate H264 "master" context to thread contexts.
983 static int clone_slice(H264Context *dst, H264Context *src)
985 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
986 dst->cur_pic_ptr = src->cur_pic_ptr;
987 dst->cur_pic = src->cur_pic;
988 dst->linesize = src->linesize;
989 dst->uvlinesize = src->uvlinesize;
990 dst->first_field = src->first_field;
992 dst->prev_poc_msb = src->prev_poc_msb;
993 dst->prev_poc_lsb = src->prev_poc_lsb;
994 dst->prev_frame_num_offset = src->prev_frame_num_offset;
995 dst->prev_frame_num = src->prev_frame_num;
996 dst->short_ref_count = src->short_ref_count;
998 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
999 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1000 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1002 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1003 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1008 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1010 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
1011 CONFIG_H264_VAAPI_HWACCEL + \
1012 (CONFIG_H264_VDA_HWACCEL * 2) + \
1013 CONFIG_H264_VDPAU_HWACCEL)
1014 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
1015 const enum AVPixelFormat *choices = pix_fmts;
1018 switch (h->sps.bit_depth_luma) {
1021 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1022 *fmt++ = AV_PIX_FMT_GBRP9;
1024 *fmt++ = AV_PIX_FMT_YUV444P9;
1025 } else if (CHROMA422(h))
1026 *fmt++ = AV_PIX_FMT_YUV422P9;
1028 *fmt++ = AV_PIX_FMT_YUV420P9;
1032 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1033 *fmt++ = AV_PIX_FMT_GBRP10;
1035 *fmt++ = AV_PIX_FMT_YUV444P10;
1036 } else if (CHROMA422(h))
1037 *fmt++ = AV_PIX_FMT_YUV422P10;
1039 *fmt++ = AV_PIX_FMT_YUV420P10;
1043 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1044 *fmt++ = AV_PIX_FMT_GBRP12;
1046 *fmt++ = AV_PIX_FMT_YUV444P12;
1047 } else if (CHROMA422(h))
1048 *fmt++ = AV_PIX_FMT_YUV422P12;
1050 *fmt++ = AV_PIX_FMT_YUV420P12;
1054 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1055 *fmt++ = AV_PIX_FMT_GBRP14;
1057 *fmt++ = AV_PIX_FMT_YUV444P14;
1058 } else if (CHROMA422(h))
1059 *fmt++ = AV_PIX_FMT_YUV422P14;
1061 *fmt++ = AV_PIX_FMT_YUV420P14;
1064 #if CONFIG_H264_VDPAU_HWACCEL
1065 *fmt++ = AV_PIX_FMT_VDPAU;
1068 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1069 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1070 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1071 *fmt++ = AV_PIX_FMT_GBRP;
1072 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1073 *fmt++ = AV_PIX_FMT_YUVJ444P;
1075 *fmt++ = AV_PIX_FMT_YUV444P;
1076 } else if (CHROMA422(h)) {
1077 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1078 *fmt++ = AV_PIX_FMT_YUVJ422P;
1080 *fmt++ = AV_PIX_FMT_YUV422P;
1082 #if CONFIG_H264_DXVA2_HWACCEL
1083 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1085 #if CONFIG_H264_VAAPI_HWACCEL
1086 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1088 #if CONFIG_H264_VDA_HWACCEL
1089 *fmt++ = AV_PIX_FMT_VDA_VLD;
1090 *fmt++ = AV_PIX_FMT_VDA;
1092 if (h->avctx->codec->pix_fmts)
1093 choices = h->avctx->codec->pix_fmts;
1094 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1095 *fmt++ = AV_PIX_FMT_YUVJ420P;
1097 *fmt++ = AV_PIX_FMT_YUV420P;
1101 av_log(h->avctx, AV_LOG_ERROR,
1102 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1103 return AVERROR_INVALIDDATA;
1106 *fmt = AV_PIX_FMT_NONE;
1108 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1109 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1111 return ff_thread_get_format(h->avctx, choices);
1114 /* export coded and cropped frame dimensions to AVCodecContext */
1115 static int init_dimensions(H264Context *h)
1117 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1118 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1119 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1120 h->sps.crop_right || h->sps.crop_bottom;
1121 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1122 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1124 /* handle container cropping */
1125 if (!crop_present &&
1126 FFALIGN(h->avctx->width, 16) == h->width &&
1127 FFALIGN(h->avctx->height, 16) == h->height) {
1128 width = h->avctx->width;
1129 height = h->avctx->height;
1132 if (width <= 0 || height <= 0) {
1133 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1135 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1136 return AVERROR_INVALIDDATA;
1138 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1139 h->sps.crop_bottom =
1149 h->avctx->coded_width = h->width;
1150 h->avctx->coded_height = h->height;
1151 h->avctx->width = width;
1152 h->avctx->height = height;
1157 static int h264_slice_header_init(H264Context *h, int reinit)
1159 int nb_slices = (HAVE_THREADS &&
1160 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1161 h->avctx->thread_count : 1;
1164 ff_set_sar(h->avctx, h->sps.sar);
1165 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1166 &h->chroma_x_shift, &h->chroma_y_shift);
1168 if (h->sps.timing_info_present_flag) {
1169 int64_t den = h->sps.time_scale;
1170 if (h->x264_build < 44U)
1172 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1173 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1177 ff_h264_free_tables(h, 0);
1179 h->prev_interlaced_frame = 1;
1181 init_scan_tables(h);
1182 ret = ff_h264_alloc_tables(h);
1184 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1188 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1191 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1193 max_slices = H264_MAX_THREADS;
1194 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1195 " reducing to %d\n", nb_slices, max_slices);
1196 nb_slices = max_slices;
1198 h->slice_context_count = nb_slices;
1200 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1201 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1203 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1207 for (i = 1; i < h->slice_context_count; i++) {
1209 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1211 ret = AVERROR(ENOMEM);
1214 c->avctx = h->avctx;
1216 c->h264dsp = h->h264dsp;
1217 c->h264qpel = h->h264qpel;
1218 c->h264chroma = h->h264chroma;
1221 c->pixel_shift = h->pixel_shift;
1222 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1223 c->width = h->width;
1224 c->height = h->height;
1225 c->linesize = h->linesize;
1226 c->uvlinesize = h->uvlinesize;
1227 c->chroma_x_shift = h->chroma_x_shift;
1228 c->chroma_y_shift = h->chroma_y_shift;
1229 c->droppable = h->droppable;
1230 c->low_delay = h->low_delay;
1231 c->mb_width = h->mb_width;
1232 c->mb_height = h->mb_height;
1233 c->mb_stride = h->mb_stride;
1234 c->mb_num = h->mb_num;
1235 c->flags = h->flags;
1236 c->workaround_bugs = h->workaround_bugs;
1237 c->pict_type = h->pict_type;
1239 h->slice_ctx[i].h264 = c;
1241 init_scan_tables(c);
1242 clone_tables(c, &h->slice_ctx[i], h, i);
1243 c->context_initialized = 1;
1246 for (i = 0; i < h->slice_context_count; i++)
1247 if ((ret = ff_h264_slice_context_init(h, &h->slice_ctx[i])) < 0) {
1248 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1253 h->context_initialized = 1;
1257 ff_h264_free_tables(h, 0);
1258 h->context_initialized = 0;
1262 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1265 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1266 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1267 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1274 * Decode a slice header.
1275 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1277 * @param h h264context
1278 * @param h0 h264 master context (differs from 'h' when doing sliced based
1279 * parallel decoding)
1281 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1283 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Context *h0)
1285 unsigned int first_mb_in_slice;
1286 unsigned int pps_id;
1288 unsigned int slice_type, tmp, i, j;
1289 int last_pic_structure, last_pic_droppable;
1291 int needs_reinit = 0;
1292 int field_pic_flag, bottom_field_flag;
1293 int first_slice = h == h0 && !h0->current_slice;
1294 int frame_num, picture_structure, droppable;
1297 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1298 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1300 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1302 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1303 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1304 ff_h264_field_end(h, sl, 1);
1307 h0->current_slice = 0;
1308 if (!h0->first_field) {
1309 if (h->cur_pic_ptr && !h->droppable) {
1310 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1311 h->picture_structure == PICT_BOTTOM_FIELD);
1313 h->cur_pic_ptr = NULL;
1317 slice_type = get_ue_golomb_31(&sl->gb);
1318 if (slice_type > 9) {
1319 av_log(h->avctx, AV_LOG_ERROR,
1320 "slice type %d too large at %d\n",
1321 slice_type, first_mb_in_slice);
1322 return AVERROR_INVALIDDATA;
1324 if (slice_type > 4) {
1326 sl->slice_type_fixed = 1;
1328 sl->slice_type_fixed = 0;
1330 slice_type = golomb_to_pict_type[slice_type];
1331 sl->slice_type = slice_type;
1332 sl->slice_type_nos = slice_type & 3;
1334 if (h->nal_unit_type == NAL_IDR_SLICE &&
1335 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1336 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1337 return AVERROR_INVALIDDATA;
1341 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1342 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1343 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1344 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1345 h->avctx->skip_frame >= AVDISCARD_ALL) {
1346 return SLICE_SKIPED;
1349 // to make a few old functions happy, it's wrong though
1350 h->pict_type = sl->slice_type;
1352 pps_id = get_ue_golomb(&sl->gb);
1353 if (pps_id >= MAX_PPS_COUNT) {
1354 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1355 return AVERROR_INVALIDDATA;
1357 if (!h0->pps_buffers[pps_id]) {
1358 av_log(h->avctx, AV_LOG_ERROR,
1359 "non-existing PPS %u referenced\n",
1361 return AVERROR_INVALIDDATA;
1363 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1364 av_log(h->avctx, AV_LOG_ERROR,
1365 "PPS change from %d to %d forbidden\n",
1366 h0->au_pps_id, pps_id);
1367 return AVERROR_INVALIDDATA;
1370 pps = h0->pps_buffers[pps_id];
1372 if (!h0->sps_buffers[pps->sps_id]) {
1373 av_log(h->avctx, AV_LOG_ERROR,
1374 "non-existing SPS %u referenced\n",
1376 return AVERROR_INVALIDDATA;
1379 h->pps = *h0->pps_buffers[pps_id];
1381 if (pps->sps_id != h->sps.sps_id ||
1382 pps->sps_id != h->current_sps_id ||
1383 h0->sps_buffers[pps->sps_id]->new) {
1386 av_log(h->avctx, AV_LOG_ERROR,
1387 "SPS changed in the middle of the frame\n");
1388 return AVERROR_INVALIDDATA;
1391 h->sps = *h0->sps_buffers[h->pps.sps_id];
1393 if (h->mb_width != h->sps.mb_width ||
1394 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1395 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1396 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1400 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1401 h->chroma_format_idc != h->sps.chroma_format_idc) {
1402 h->bit_depth_luma = h->sps.bit_depth_luma;
1403 h->chroma_format_idc = h->sps.chroma_format_idc;
1406 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1410 h->avctx->profile = ff_h264_get_profile(&h->sps);
1411 h->avctx->level = h->sps.level_idc;
1412 h->avctx->refs = h->sps.ref_frame_count;
1414 must_reinit = (h->context_initialized &&
1415 ( 16*h->sps.mb_width != h->avctx->coded_width
1416 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1417 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1418 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1419 || h->mb_width != h->sps.mb_width
1420 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1422 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1425 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1428 h->mb_width = h->sps.mb_width;
1429 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1430 h->mb_num = h->mb_width * h->mb_height;
1431 h->mb_stride = h->mb_width + 1;
1433 h->b_stride = h->mb_width * 4;
1435 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1437 h->width = 16 * h->mb_width;
1438 h->height = 16 * h->mb_height;
1440 ret = init_dimensions(h);
1444 if (h->sps.video_signal_type_present_flag) {
1445 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1447 if (h->sps.colour_description_present_flag) {
1448 if (h->avctx->colorspace != h->sps.colorspace)
1450 h->avctx->color_primaries = h->sps.color_primaries;
1451 h->avctx->color_trc = h->sps.color_trc;
1452 h->avctx->colorspace = h->sps.colorspace;
1456 if (h->context_initialized &&
1457 (must_reinit || needs_reinit)) {
1459 av_log(h->avctx, AV_LOG_ERROR,
1460 "changing width %d -> %d / height %d -> %d on "
1462 h->width, h->avctx->coded_width,
1463 h->height, h->avctx->coded_height,
1464 h0->current_slice + 1);
1465 return AVERROR_INVALIDDATA;
1468 av_assert1(first_slice);
1470 ff_h264_flush_change(h);
1472 if ((ret = get_pixel_format(h, 1)) < 0)
1474 h->avctx->pix_fmt = ret;
1476 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1477 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1479 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1480 av_log(h->avctx, AV_LOG_ERROR,
1481 "h264_slice_header_init() failed\n");
1485 if (!h->context_initialized) {
1487 av_log(h->avctx, AV_LOG_ERROR,
1488 "Cannot (re-)initialize context during parallel decoding.\n");
1489 return AVERROR_PATCHWELCOME;
1492 if ((ret = get_pixel_format(h, 1)) < 0)
1494 h->avctx->pix_fmt = ret;
1496 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1497 av_log(h->avctx, AV_LOG_ERROR,
1498 "h264_slice_header_init() failed\n");
1503 if (first_slice && h->dequant_coeff_pps != pps_id) {
1504 h->dequant_coeff_pps = pps_id;
1505 ff_h264_init_dequant_tables(h);
1508 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1510 if (h0->frame_num != frame_num) {
1511 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1512 h0->frame_num, frame_num);
1513 return AVERROR_INVALIDDATA;
1518 h->mb_aff_frame = 0;
1519 last_pic_structure = h0->picture_structure;
1520 last_pic_droppable = h0->droppable;
1521 droppable = h->nal_ref_idc == 0;
1522 if (h->sps.frame_mbs_only_flag) {
1523 picture_structure = PICT_FRAME;
1525 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1526 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1529 field_pic_flag = get_bits1(&sl->gb);
1531 if (field_pic_flag) {
1532 bottom_field_flag = get_bits1(&sl->gb);
1533 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1535 picture_structure = PICT_FRAME;
1536 h->mb_aff_frame = h->sps.mb_aff;
1539 if (h0->current_slice) {
1540 if (last_pic_structure != picture_structure ||
1541 last_pic_droppable != droppable) {
1542 av_log(h->avctx, AV_LOG_ERROR,
1543 "Changing field mode (%d -> %d) between slices is not allowed\n",
1544 last_pic_structure, h->picture_structure);
1545 return AVERROR_INVALIDDATA;
1546 } else if (!h0->cur_pic_ptr) {
1547 av_log(h->avctx, AV_LOG_ERROR,
1548 "unset cur_pic_ptr on slice %d\n",
1549 h0->current_slice + 1);
1550 return AVERROR_INVALIDDATA;
1554 h->picture_structure = picture_structure;
1555 h->droppable = droppable;
1556 h->frame_num = frame_num;
1557 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1559 if (h0->current_slice == 0) {
1560 /* Shorten frame num gaps so we don't have to allocate reference
1561 * frames just to throw them away */
1562 if (h->frame_num != h->prev_frame_num) {
1563 int unwrap_prev_frame_num = h->prev_frame_num;
1564 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1566 if (unwrap_prev_frame_num > h->frame_num)
1567 unwrap_prev_frame_num -= max_frame_num;
1569 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1570 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1571 if (unwrap_prev_frame_num < 0)
1572 unwrap_prev_frame_num += max_frame_num;
1574 h->prev_frame_num = unwrap_prev_frame_num;
1578 /* See if we have a decoded first field looking for a pair...
1579 * Here, we're using that to see if we should mark previously
1580 * decode frames as "finished".
1581 * We have to do that before the "dummy" in-between frame allocation,
1582 * since that can modify h->cur_pic_ptr. */
1583 if (h0->first_field) {
1584 assert(h0->cur_pic_ptr);
1585 assert(h0->cur_pic_ptr->f.buf[0]);
1586 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1588 /* Mark old field/frame as completed */
1589 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1590 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1591 last_pic_structure == PICT_BOTTOM_FIELD);
1594 /* figure out if we have a complementary field pair */
1595 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1596 /* Previous field is unmatched. Don't display it, but let it
1597 * remain for reference if marked as such. */
1598 if (last_pic_structure != PICT_FRAME) {
1599 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1600 last_pic_structure == PICT_TOP_FIELD);
1603 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1604 /* This and previous field were reference, but had
1605 * different frame_nums. Consider this field first in
1606 * pair. Throw away previous field except for reference
1608 if (last_pic_structure != PICT_FRAME) {
1609 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1610 last_pic_structure == PICT_TOP_FIELD);
1613 /* Second field in complementary pair */
1614 if (!((last_pic_structure == PICT_TOP_FIELD &&
1615 h->picture_structure == PICT_BOTTOM_FIELD) ||
1616 (last_pic_structure == PICT_BOTTOM_FIELD &&
1617 h->picture_structure == PICT_TOP_FIELD))) {
1618 av_log(h->avctx, AV_LOG_ERROR,
1619 "Invalid field mode combination %d/%d\n",
1620 last_pic_structure, h->picture_structure);
1621 h->picture_structure = last_pic_structure;
1622 h->droppable = last_pic_droppable;
1623 return AVERROR_INVALIDDATA;
1624 } else if (last_pic_droppable != h->droppable) {
1625 avpriv_request_sample(h->avctx,
1626 "Found reference and non-reference fields in the same frame, which");
1627 h->picture_structure = last_pic_structure;
1628 h->droppable = last_pic_droppable;
1629 return AVERROR_PATCHWELCOME;
1635 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1636 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1637 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1638 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1639 h->frame_num, h->prev_frame_num);
1640 if (!h->sps.gaps_in_frame_num_allowed_flag)
1641 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1642 h->last_pocs[i] = INT_MIN;
1643 ret = h264_frame_start(h);
1645 h0->first_field = 0;
1649 h->prev_frame_num++;
1650 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1651 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1652 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1653 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1654 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1655 ret = ff_generate_sliding_window_mmcos(h, 1);
1656 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1658 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1659 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1661 /* Error concealment: If a ref is missing, copy the previous ref
1663 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1664 * many assumptions about there being no actual duplicates.
1665 * FIXME: This does not copy padding for out-of-frame motion
1666 * vectors. Given we are concealing a lost frame, this probably
1667 * is not noticeable by comparison, but it should be fixed. */
1668 if (h->short_ref_count) {
1670 av_image_copy(h->short_ref[0]->f.data,
1671 h->short_ref[0]->f.linesize,
1672 (const uint8_t **)prev->f.data,
1677 h->short_ref[0]->poc = prev->poc + 2;
1679 h->short_ref[0]->frame_num = h->prev_frame_num;
1683 /* See if we have a decoded first field looking for a pair...
1684 * We're using that to see whether to continue decoding in that
1685 * frame, or to allocate a new one. */
1686 if (h0->first_field) {
1687 assert(h0->cur_pic_ptr);
1688 assert(h0->cur_pic_ptr->f.buf[0]);
1689 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1691 /* figure out if we have a complementary field pair */
1692 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1693 /* Previous field is unmatched. Don't display it, but let it
1694 * remain for reference if marked as such. */
1695 h0->missing_fields ++;
1696 h0->cur_pic_ptr = NULL;
1697 h0->first_field = FIELD_PICTURE(h);
1699 h0->missing_fields = 0;
1700 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1701 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1702 h0->picture_structure==PICT_BOTTOM_FIELD);
1703 /* This and the previous field had different frame_nums.
1704 * Consider this field first in pair. Throw away previous
1705 * one except for reference purposes. */
1706 h0->first_field = 1;
1707 h0->cur_pic_ptr = NULL;
1709 /* Second field in complementary pair */
1710 h0->first_field = 0;
1714 /* Frame or first field in a potentially complementary pair */
1715 h0->first_field = FIELD_PICTURE(h);
1718 if (!FIELD_PICTURE(h) || h0->first_field) {
1719 if (h264_frame_start(h) < 0) {
1720 h0->first_field = 0;
1721 return AVERROR_INVALIDDATA;
1724 release_unused_pictures(h, 0);
1726 /* Some macroblocks can be accessed before they're available in case
1727 * of lost slices, MBAFF or threading. */
1728 if (FIELD_PICTURE(h)) {
1729 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1730 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1732 memset(h->slice_table, -1,
1733 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1735 h0->last_slice_type = -1;
1737 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1741 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1743 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1744 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1745 first_mb_in_slice >= h->mb_num) {
1746 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1747 return AVERROR_INVALIDDATA;
1749 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1750 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1751 FIELD_OR_MBAFF_PICTURE(h);
1752 if (h->picture_structure == PICT_BOTTOM_FIELD)
1753 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1754 av_assert1(sl->mb_y < h->mb_height);
1756 if (h->picture_structure == PICT_FRAME) {
1757 h->curr_pic_num = h->frame_num;
1758 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1760 h->curr_pic_num = 2 * h->frame_num + 1;
1761 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1764 if (h->nal_unit_type == NAL_IDR_SLICE)
1765 get_ue_golomb(&sl->gb); /* idr_pic_id */
1767 if (h->sps.poc_type == 0) {
1768 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1770 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1771 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1774 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1775 h->delta_poc[0] = get_se_golomb(&sl->gb);
1777 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1778 h->delta_poc[1] = get_se_golomb(&sl->gb);
1781 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1783 if (h->pps.redundant_pic_cnt_present)
1784 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1786 ret = ff_set_ref_count(h, sl);
1790 if (slice_type != AV_PICTURE_TYPE_I &&
1791 (h0->current_slice == 0 ||
1792 slice_type != h0->last_slice_type ||
1793 memcmp(h0->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1795 ff_h264_fill_default_ref_list(h, sl);
1798 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1799 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1801 sl->ref_count[1] = sl->ref_count[0] = 0;
1806 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1807 (h->pps.weighted_bipred_idc == 1 &&
1808 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1809 ff_pred_weight_table(h, sl);
1810 else if (h->pps.weighted_bipred_idc == 2 &&
1811 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1812 implicit_weight_table(h, sl, -1);
1815 for (i = 0; i < 2; i++) {
1816 sl->luma_weight_flag[i] = 0;
1817 sl->chroma_weight_flag[i] = 0;
1821 // If frame-mt is enabled, only update mmco tables for the first slice
1822 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1823 // or h->mmco, which will cause ref list mix-ups and decoding errors
1824 // further down the line. This may break decoding if the first slice is
1825 // corrupt, thus we only do this if frame-mt is enabled.
1826 if (h->nal_ref_idc) {
1827 ret = ff_h264_decode_ref_pic_marking(h0, &sl->gb,
1828 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1829 h0->current_slice == 0);
1830 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1831 return AVERROR_INVALIDDATA;
1834 if (FRAME_MBAFF(h)) {
1835 ff_h264_fill_mbaff_ref_list(h, sl);
1837 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1838 implicit_weight_table(h, sl, 0);
1839 implicit_weight_table(h, sl, 1);
1843 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1844 ff_h264_direct_dist_scale_factor(h, sl);
1845 ff_h264_direct_ref_list_init(h, sl);
1847 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1848 tmp = get_ue_golomb_31(&sl->gb);
1850 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1851 return AVERROR_INVALIDDATA;
1853 sl->cabac_init_idc = tmp;
1856 sl->last_qscale_diff = 0;
1857 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1858 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1859 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1860 return AVERROR_INVALIDDATA;
1863 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1864 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1865 // FIXME qscale / qp ... stuff
1866 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1867 get_bits1(&sl->gb); /* sp_for_switch_flag */
1868 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1869 sl->slice_type == AV_PICTURE_TYPE_SI)
1870 get_se_golomb(&sl->gb); /* slice_qs_delta */
1872 sl->deblocking_filter = 1;
1873 sl->slice_alpha_c0_offset = 0;
1874 sl->slice_beta_offset = 0;
1875 if (h->pps.deblocking_filter_parameters_present) {
1876 tmp = get_ue_golomb_31(&sl->gb);
1878 av_log(h->avctx, AV_LOG_ERROR,
1879 "deblocking_filter_idc %u out of range\n", tmp);
1880 return AVERROR_INVALIDDATA;
1882 sl->deblocking_filter = tmp;
1883 if (sl->deblocking_filter < 2)
1884 sl->deblocking_filter ^= 1; // 1<->0
1886 if (sl->deblocking_filter) {
1887 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1888 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1889 if (sl->slice_alpha_c0_offset > 12 ||
1890 sl->slice_alpha_c0_offset < -12 ||
1891 sl->slice_beta_offset > 12 ||
1892 sl->slice_beta_offset < -12) {
1893 av_log(h->avctx, AV_LOG_ERROR,
1894 "deblocking filter parameters %d %d out of range\n",
1895 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1896 return AVERROR_INVALIDDATA;
1901 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1902 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1903 h->nal_unit_type != NAL_IDR_SLICE) ||
1904 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1905 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1906 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1907 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1908 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1909 h->nal_ref_idc == 0))
1910 sl->deblocking_filter = 0;
1912 if (sl->deblocking_filter == 1 && h0->max_contexts > 1) {
1913 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1914 /* Cheat slightly for speed:
1915 * Do not bother to deblock across slices. */
1916 sl->deblocking_filter = 2;
1918 h0->max_contexts = 1;
1919 if (!h0->single_decode_warning) {
1920 av_log(h->avctx, AV_LOG_INFO,
1921 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1922 "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"
1923 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1924 "but will generate non-standard-compliant output.\n");
1925 h0->single_decode_warning = 1;
1928 av_log(h->avctx, AV_LOG_ERROR,
1929 "Deblocking switched inside frame.\n");
1930 return SLICE_SINGLETHREAD;
1934 sl->qp_thresh = 15 -
1935 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1937 h->pps.chroma_qp_index_offset[0],
1938 h->pps.chroma_qp_index_offset[1]) +
1939 6 * (h->sps.bit_depth_luma - 8);
1941 h0->last_slice_type = slice_type;
1942 memcpy(h0->last_ref_count, sl->ref_count, sizeof(h0->last_ref_count));
1943 sl->slice_num = ++h0->current_slice;
1946 h0->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1947 if ( h0->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1948 && h0->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1949 && sl->slice_num >= MAX_SLICES) {
1950 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1951 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);
1954 for (j = 0; j < 2; j++) {
1956 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1957 for (i = 0; i < 16; i++) {
1959 if (j < sl->list_count && i < sl->ref_count[j] &&
1960 sl->ref_list[j][i].f.buf[0]) {
1962 AVBuffer *buf = sl->ref_list[j][i].f.buf[0]->buffer;
1963 for (k = 0; k < h->short_ref_count; k++)
1964 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1968 for (k = 0; k < h->long_ref_count; k++)
1969 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1970 id_list[i] = h->short_ref_count + k;
1978 for (i = 0; i < 16; i++)
1979 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1981 ref2frm[18 + 1] = -1;
1982 for (i = 16; i < 48; i++)
1983 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1984 (sl->ref_list[j][i].reference & 3);
1987 h0->au_pps_id = pps_id;
1989 h0->sps_buffers[h->pps.sps_id]->new = 0;
1990 h->current_sps_id = h->pps.sps_id;
1992 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1993 av_log(h->avctx, AV_LOG_DEBUG,
1994 "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",
1996 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1998 av_get_picture_type_char(sl->slice_type),
1999 sl->slice_type_fixed ? " fix" : "",
2000 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2001 pps_id, h->frame_num,
2002 h->cur_pic_ptr->field_poc[0],
2003 h->cur_pic_ptr->field_poc[1],
2004 sl->ref_count[0], sl->ref_count[1],
2006 sl->deblocking_filter,
2007 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
2009 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
2010 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2016 int ff_h264_get_slice_type(const H264SliceContext *sl)
2018 switch (sl->slice_type) {
2019 case AV_PICTURE_TYPE_P:
2021 case AV_PICTURE_TYPE_B:
2023 case AV_PICTURE_TYPE_I:
2025 case AV_PICTURE_TYPE_SP:
2027 case AV_PICTURE_TYPE_SI:
2030 return AVERROR_INVALIDDATA;
2034 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
2035 H264SliceContext *sl,
2036 int mb_type, int top_xy,
2037 int left_xy[LEFT_MBS],
2039 int left_type[LEFT_MBS],
2040 int mb_xy, int list)
2042 int b_stride = h->b_stride;
2043 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
2044 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
2045 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2046 if (USES_LIST(top_type, list)) {
2047 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2048 const int b8_xy = 4 * top_xy + 2;
2049 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2050 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2051 ref_cache[0 - 1 * 8] =
2052 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2053 ref_cache[2 - 1 * 8] =
2054 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2056 AV_ZERO128(mv_dst - 1 * 8);
2057 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2060 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2061 if (USES_LIST(left_type[LTOP], list)) {
2062 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2063 const int b8_xy = 4 * left_xy[LTOP] + 1;
2064 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2065 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2066 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2067 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2068 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2070 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2071 ref_cache[-1 + 16] =
2072 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2074 AV_ZERO32(mv_dst - 1 + 0);
2075 AV_ZERO32(mv_dst - 1 + 8);
2076 AV_ZERO32(mv_dst - 1 + 16);
2077 AV_ZERO32(mv_dst - 1 + 24);
2080 ref_cache[-1 + 16] =
2081 ref_cache[-1 + 24] = LIST_NOT_USED;
2086 if (!USES_LIST(mb_type, list)) {
2087 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2088 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2089 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2090 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2091 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2096 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2097 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2098 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2099 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2100 AV_WN32A(&ref_cache[0 * 8], ref01);
2101 AV_WN32A(&ref_cache[1 * 8], ref01);
2102 AV_WN32A(&ref_cache[2 * 8], ref23);
2103 AV_WN32A(&ref_cache[3 * 8], ref23);
2107 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2108 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2109 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2110 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2111 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2117 * @return non zero if the loop filter can be skipped
2119 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2121 const int mb_xy = sl->mb_xy;
2122 int top_xy, left_xy[LEFT_MBS];
2123 int top_type, left_type[LEFT_MBS];
2127 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2129 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2130 * stuff, I can't imagine that these complex rules are worth it. */
2132 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2133 if (FRAME_MBAFF(h)) {
2134 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2135 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2137 if (left_mb_field_flag != curr_mb_field_flag)
2138 left_xy[LTOP] -= h->mb_stride;
2140 if (curr_mb_field_flag)
2141 top_xy += h->mb_stride &
2142 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2143 if (left_mb_field_flag != curr_mb_field_flag)
2144 left_xy[LBOT] += h->mb_stride;
2148 sl->top_mb_xy = top_xy;
2149 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2150 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2152 /* For sufficiently low qp, filtering wouldn't do anything.
2153 * This is a conservative estimate: could also check beta_offset
2154 * and more accurate chroma_qp. */
2155 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2156 int qp = h->cur_pic.qscale_table[mb_xy];
2157 if (qp <= qp_thresh &&
2158 (left_xy[LTOP] < 0 ||
2159 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2161 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2162 if (!FRAME_MBAFF(h))
2164 if ((left_xy[LTOP] < 0 ||
2165 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2166 (top_xy < h->mb_stride ||
2167 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2172 top_type = h->cur_pic.mb_type[top_xy];
2173 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2174 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2175 if (sl->deblocking_filter == 2) {
2176 if (h->slice_table[top_xy] != sl->slice_num)
2178 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2179 left_type[LTOP] = left_type[LBOT] = 0;
2181 if (h->slice_table[top_xy] == 0xFFFF)
2183 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2184 left_type[LTOP] = left_type[LBOT] = 0;
2186 sl->top_type = top_type;
2187 sl->left_type[LTOP] = left_type[LTOP];
2188 sl->left_type[LBOT] = left_type[LBOT];
2190 if (IS_INTRA(mb_type))
2193 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2194 top_type, left_type, mb_xy, 0);
2195 if (sl->list_count == 2)
2196 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2197 top_type, left_type, mb_xy, 1);
2199 nnz = h->non_zero_count[mb_xy];
2200 nnz_cache = sl->non_zero_count_cache;
2201 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2202 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2203 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2204 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2205 sl->cbp = h->cbp_table[mb_xy];
2208 nnz = h->non_zero_count[top_xy];
2209 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2212 if (left_type[LTOP]) {
2213 nnz = h->non_zero_count[left_xy[LTOP]];
2214 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2215 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2216 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2217 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2220 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2221 * from what the loop filter needs */
2222 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2223 if (IS_8x8DCT(top_type)) {
2224 nnz_cache[4 + 8 * 0] =
2225 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2226 nnz_cache[6 + 8 * 0] =
2227 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2229 if (IS_8x8DCT(left_type[LTOP])) {
2230 nnz_cache[3 + 8 * 1] =
2231 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2233 if (IS_8x8DCT(left_type[LBOT])) {
2234 nnz_cache[3 + 8 * 3] =
2235 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2238 if (IS_8x8DCT(mb_type)) {
2239 nnz_cache[scan8[0]] =
2240 nnz_cache[scan8[1]] =
2241 nnz_cache[scan8[2]] =
2242 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2244 nnz_cache[scan8[0 + 4]] =
2245 nnz_cache[scan8[1 + 4]] =
2246 nnz_cache[scan8[2 + 4]] =
2247 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2249 nnz_cache[scan8[0 + 8]] =
2250 nnz_cache[scan8[1 + 8]] =
2251 nnz_cache[scan8[2 + 8]] =
2252 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2254 nnz_cache[scan8[0 + 12]] =
2255 nnz_cache[scan8[1 + 12]] =
2256 nnz_cache[scan8[2 + 12]] =
2257 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2264 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2266 uint8_t *dest_y, *dest_cb, *dest_cr;
2267 int linesize, uvlinesize, mb_x, mb_y;
2268 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2269 const int old_slice_type = sl->slice_type;
2270 const int pixel_shift = h->pixel_shift;
2271 const int block_h = 16 >> h->chroma_y_shift;
2273 if (sl->deblocking_filter) {
2274 for (mb_x = start_x; mb_x < end_x; mb_x++)
2275 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2277 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2278 sl->slice_num = h->slice_table[mb_xy];
2279 mb_type = h->cur_pic.mb_type[mb_xy];
2280 sl->list_count = h->list_counts[mb_xy];
2284 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2288 dest_y = h->cur_pic.f.data[0] +
2289 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2290 dest_cb = h->cur_pic.f.data[1] +
2291 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2292 mb_y * h->uvlinesize * block_h;
2293 dest_cr = h->cur_pic.f.data[2] +
2294 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2295 mb_y * h->uvlinesize * block_h;
2296 // FIXME simplify above
2299 linesize = sl->mb_linesize = h->linesize * 2;
2300 uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
2301 if (mb_y & 1) { // FIXME move out of this function?
2302 dest_y -= h->linesize * 15;
2303 dest_cb -= h->uvlinesize * (block_h - 1);
2304 dest_cr -= h->uvlinesize * (block_h - 1);
2307 linesize = sl->mb_linesize = h->linesize;
2308 uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
2310 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2312 if (fill_filter_caches(h, sl, mb_type))
2314 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2315 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2317 if (FRAME_MBAFF(h)) {
2318 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2319 linesize, uvlinesize);
2321 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2322 dest_cr, linesize, uvlinesize);
2326 sl->slice_type = old_slice_type;
2328 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2329 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2330 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2333 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2335 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2336 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2337 h->cur_pic.mb_type[mb_xy - 1] :
2338 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2339 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2340 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2344 * Draw edges and report progress for the last MB row.
2346 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2348 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2349 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2350 int height = 16 << FRAME_MBAFF(h);
2351 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2353 if (sl->deblocking_filter) {
2354 if ((top + height) >= pic_height)
2355 height += deblock_border;
2356 top -= deblock_border;
2359 if (top >= pic_height || (top + height) < 0)
2362 height = FFMIN(height, pic_height - top);
2364 height = top + height;
2368 ff_h264_draw_horiz_band(h, sl, top, height);
2370 if (h->droppable || sl->er.error_occurred)
2373 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2374 h->picture_structure == PICT_BOTTOM_FIELD);
2377 static void er_add_slice(H264SliceContext *sl,
2378 int startx, int starty,
2379 int endx, int endy, int status)
2381 if (CONFIG_ERROR_RESILIENCE) {
2382 ERContext *er = &sl->er;
2384 ff_er_add_slice(er, startx, starty, endx, endy, status);
2388 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2390 H264SliceContext *sl = arg;
2391 const H264Context *h = sl->h264;
2392 int lf_x_start = sl->mb_x;
2395 ret = alloc_scratch_buffers(sl, h->linesize);
2399 sl->mb_skip_run = -1;
2401 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2403 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2404 avctx->codec_id != AV_CODEC_ID_H264 ||
2405 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2407 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && sl->er.error_status_table) {
2408 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2410 int prev_status = sl->er.error_status_table[sl->er.mb_index2xy[start_i - 1]];
2411 prev_status &= ~ VP_START;
2412 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2413 sl->er.error_occurred = 1;
2419 align_get_bits(&sl->gb);
2422 ff_init_cabac_decoder(&sl->cabac,
2423 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2424 (get_bits_left(&sl->gb) + 7) / 8);
2426 ff_h264_init_cabac_states(h, sl);
2430 int ret = ff_h264_decode_mb_cabac(h, sl);
2432 // STOP_TIMER("decode_mb_cabac")
2435 ff_h264_hl_decode_mb(h, sl);
2437 // FIXME optimal? or let mb_decode decode 16x32 ?
2438 if (ret >= 0 && FRAME_MBAFF(h)) {
2441 ret = ff_h264_decode_mb_cabac(h, sl);
2444 ff_h264_hl_decode_mb(h, sl);
2447 eos = get_cabac_terminate(&sl->cabac);
2449 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2450 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2451 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2452 sl->mb_y, ER_MB_END);
2453 if (sl->mb_x >= lf_x_start)
2454 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2457 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2458 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2459 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2460 av_log(h->avctx, AV_LOG_ERROR,
2461 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2463 sl->cabac.bytestream_end - sl->cabac.bytestream);
2464 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2465 sl->mb_y, ER_MB_ERROR);
2466 return AVERROR_INVALIDDATA;
2469 if (++sl->mb_x >= h->mb_width) {
2470 loop_filter(h, sl, lf_x_start, sl->mb_x);
2471 sl->mb_x = lf_x_start = 0;
2472 decode_finish_row(h, sl);
2474 if (FIELD_OR_MBAFF_PICTURE(h)) {
2476 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2477 predict_field_decoding_flag(h, sl);
2481 if (eos || sl->mb_y >= h->mb_height) {
2482 tprintf(h->avctx, "slice end %d %d\n",
2483 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2484 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2485 sl->mb_y, ER_MB_END);
2486 if (sl->mb_x > lf_x_start)
2487 loop_filter(h, sl, lf_x_start, sl->mb_x);
2493 int ret = ff_h264_decode_mb_cavlc(h, sl);
2496 ff_h264_hl_decode_mb(h, sl);
2498 // FIXME optimal? or let mb_decode decode 16x32 ?
2499 if (ret >= 0 && FRAME_MBAFF(h)) {
2501 ret = ff_h264_decode_mb_cavlc(h, sl);
2504 ff_h264_hl_decode_mb(h, sl);
2509 av_log(h->avctx, AV_LOG_ERROR,
2510 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2511 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2512 sl->mb_y, ER_MB_ERROR);
2516 if (++sl->mb_x >= h->mb_width) {
2517 loop_filter(h, sl, lf_x_start, sl->mb_x);
2518 sl->mb_x = lf_x_start = 0;
2519 decode_finish_row(h, sl);
2521 if (FIELD_OR_MBAFF_PICTURE(h)) {
2523 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2524 predict_field_decoding_flag(h, sl);
2526 if (sl->mb_y >= h->mb_height) {
2527 tprintf(h->avctx, "slice end %d %d\n",
2528 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2530 if ( get_bits_left(&sl->gb) == 0
2531 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2532 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2533 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2537 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2538 sl->mb_x, sl->mb_y, ER_MB_END);
2540 return AVERROR_INVALIDDATA;
2545 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2546 tprintf(h->avctx, "slice end %d %d\n",
2547 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2549 if (get_bits_left(&sl->gb) == 0) {
2550 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2551 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2552 if (sl->mb_x > lf_x_start)
2553 loop_filter(h, sl, lf_x_start, sl->mb_x);
2557 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2558 sl->mb_y, ER_MB_ERROR);
2560 return AVERROR_INVALIDDATA;
2568 * Call decode_slice() for each context.
2570 * @param h h264 master context
2571 * @param context_count number of contexts to execute
2573 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2575 AVCodecContext *const avctx = h->avctx;
2577 H264SliceContext *sl;
2580 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2582 if (h->avctx->hwaccel ||
2583 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2585 if (context_count == 1) {
2586 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2587 h->mb_y = h->slice_ctx[0].mb_y;
2590 av_assert0(context_count > 0);
2591 for (i = 1; i < context_count; i++) {
2592 hx = h->thread_context[i];
2593 sl = &h->slice_ctx[i];
2594 if (CONFIG_ERROR_RESILIENCE) {
2595 sl->er.error_count = 0;
2597 hx->x264_build = h->x264_build;
2600 avctx->execute(avctx, decode_slice, h->slice_ctx,
2601 NULL, context_count, sizeof(h->slice_ctx[0]));
2603 /* pull back stuff from slices to master context */
2604 hx = h->thread_context[context_count - 1];
2605 sl = &h->slice_ctx[context_count - 1];
2607 h->droppable = hx->droppable;
2608 h->picture_structure = hx->picture_structure;
2609 if (CONFIG_ERROR_RESILIENCE) {
2610 for (i = 1; i < context_count; i++)
2611 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;