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 pic->crop = h->sps.crop;
236 pic->crop_top = h->sps.crop_top;
237 pic->crop_left= h->sps.crop_left;
239 if (h->avctx->hwaccel) {
240 const AVHWAccel *hwaccel = h->avctx->hwaccel;
241 av_assert0(!pic->hwaccel_picture_private);
242 if (hwaccel->frame_priv_data_size) {
243 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
244 if (!pic->hwaccel_priv_buf)
245 return AVERROR(ENOMEM);
246 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
249 if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
250 int h_chroma_shift, v_chroma_shift;
251 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
252 &h_chroma_shift, &v_chroma_shift);
254 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
255 memset(pic->f.data[1] + pic->f.linesize[1]*i,
256 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
257 memset(pic->f.data[2] + pic->f.linesize[2]*i,
258 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
262 if (!h->qscale_table_pool) {
263 ret = init_table_pools(h);
268 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
269 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
270 if (!pic->qscale_table_buf || !pic->mb_type_buf)
273 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
274 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
276 for (i = 0; i < 2; i++) {
277 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
278 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
279 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
282 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
283 pic->ref_index[i] = pic->ref_index_buf[i]->data;
288 ff_h264_unref_picture(h, pic);
289 return (ret < 0) ? ret : AVERROR(ENOMEM);
292 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
296 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
301 static int find_unused_picture(H264Context *h)
305 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
306 if (pic_is_unused(h, &h->DPB[i]))
309 if (i == H264_MAX_PICTURE_COUNT)
310 return AVERROR_INVALIDDATA;
312 if (h->DPB[i].needs_realloc) {
313 h->DPB[i].needs_realloc = 0;
314 ff_h264_unref_picture(h, &h->DPB[i]);
321 static void init_dequant8_coeff_table(H264Context *h)
324 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
326 for (i = 0; i < 6; i++) {
327 h->dequant8_coeff[i] = h->dequant8_buffer[i];
328 for (j = 0; j < i; j++)
329 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
330 64 * sizeof(uint8_t))) {
331 h->dequant8_coeff[i] = h->dequant8_buffer[j];
337 for (q = 0; q < max_qp + 1; q++) {
340 for (x = 0; x < 64; x++)
341 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
342 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
343 h->pps.scaling_matrix8[i][x]) << shift;
348 static void init_dequant4_coeff_table(H264Context *h)
351 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
352 for (i = 0; i < 6; i++) {
353 h->dequant4_coeff[i] = h->dequant4_buffer[i];
354 for (j = 0; j < i; j++)
355 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
356 16 * sizeof(uint8_t))) {
357 h->dequant4_coeff[i] = h->dequant4_buffer[j];
363 for (q = 0; q < max_qp + 1; q++) {
364 int shift = div6[q] + 2;
366 for (x = 0; x < 16; x++)
367 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
368 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
369 h->pps.scaling_matrix4[i][x]) << shift;
374 void ff_h264_init_dequant_tables(H264Context *h)
377 init_dequant4_coeff_table(h);
378 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
380 if (h->pps.transform_8x8_mode)
381 init_dequant8_coeff_table(h);
382 if (h->sps.transform_bypass) {
383 for (i = 0; i < 6; i++)
384 for (x = 0; x < 16; x++)
385 h->dequant4_coeff[i][0][x] = 1 << 6;
386 if (h->pps.transform_8x8_mode)
387 for (i = 0; i < 6; i++)
388 for (x = 0; x < 64; x++)
389 h->dequant8_coeff[i][0][x] = 1 << 6;
393 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
395 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
396 (((pic) && (pic) >= (old_ctx)->DPB && \
397 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
398 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
400 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
401 H264Context *new_base,
402 H264Context *old_base)
406 for (i = 0; i < count; i++) {
407 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
408 IN_RANGE(from[i], old_base->DPB,
409 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
411 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
415 static int copy_parameter_set(void **to, void **from, int count, int size)
419 for (i = 0; i < count; i++) {
420 if (to[i] && !from[i]) {
422 } else if (from[i] && !to[i]) {
423 to[i] = av_malloc(size);
425 return AVERROR(ENOMEM);
429 memcpy(to[i], from[i], size);
435 #define copy_fields(to, from, start_field, end_field) \
436 memcpy(&(to)->start_field, &(from)->start_field, \
437 (char *)&(to)->end_field - (char *)&(to)->start_field)
439 static int h264_slice_header_init(H264Context *h, int reinit);
441 int ff_h264_update_thread_context(AVCodecContext *dst,
442 const AVCodecContext *src)
444 H264Context *h = dst->priv_data, *h1 = src->priv_data;
445 int inited = h->context_initialized, err = 0;
446 int context_reinitialized = 0;
453 (h->width != h1->width ||
454 h->height != h1->height ||
455 h->mb_width != h1->mb_width ||
456 h->mb_height != h1->mb_height ||
457 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
458 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
459 h->sps.colorspace != h1->sps.colorspace)) {
461 h->width = h1->width;
462 h->height = h1->height;
463 h->mb_height = h1->mb_height;
464 h->mb_width = h1->mb_width;
465 h->mb_num = h1->mb_num;
466 h->mb_stride = h1->mb_stride;
467 h->b_stride = h1->b_stride;
469 if ((ret = copy_parameter_set((void **)h->sps_buffers,
470 (void **)h1->sps_buffers,
471 MAX_SPS_COUNT, sizeof(SPS))) < 0)
474 if ((ret = copy_parameter_set((void **)h->pps_buffers,
475 (void **)h1->pps_buffers,
476 MAX_PPS_COUNT, sizeof(PPS))) < 0)
480 if ((err = h264_slice_header_init(h, 1)) < 0) {
481 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
484 context_reinitialized = 1;
487 h264_set_parameter_from_sps(h);
488 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
489 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
493 /* copy block_offset since frame_start may not be called */
494 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
497 H264SliceContext *orig_slice_ctx = h->slice_ctx;
499 for (i = 0; i < MAX_SPS_COUNT; i++)
500 av_freep(h->sps_buffers + i);
502 for (i = 0; i < MAX_PPS_COUNT; i++)
503 av_freep(h->pps_buffers + i);
505 ff_h264_unref_picture(h, &h->last_pic_for_ec);
506 memcpy(h, h1, sizeof(H264Context));
508 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
509 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
511 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
512 memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
514 h->slice_ctx = orig_slice_ctx;
516 memset(&h->slice_ctx[0].er, 0, sizeof(h->slice_ctx[0].er));
517 memset(&h->slice_ctx[0].mb, 0, sizeof(h->slice_ctx[0].mb));
518 memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
519 memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
523 h->qscale_table_pool = NULL;
524 h->mb_type_pool = NULL;
525 h->ref_index_pool = NULL;
526 h->motion_val_pool = NULL;
527 h->intra4x4_pred_mode= NULL;
528 h->non_zero_count = NULL;
529 h->slice_table_base = NULL;
530 h->slice_table = NULL;
532 h->chroma_pred_mode_table = NULL;
533 memset(h->mvd_table, 0, sizeof(h->mvd_table));
534 h->direct_table = NULL;
535 h->list_counts = NULL;
539 if (h1->context_initialized) {
540 h->context_initialized = 0;
542 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
543 av_frame_unref(&h->cur_pic.f);
544 h->cur_pic.tf.f = &h->cur_pic.f;
546 ret = ff_h264_alloc_tables(h);
548 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
551 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
553 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
558 h->context_initialized = h1->context_initialized;
561 h->avctx->coded_height = h1->avctx->coded_height;
562 h->avctx->coded_width = h1->avctx->coded_width;
563 h->avctx->width = h1->avctx->width;
564 h->avctx->height = h1->avctx->height;
565 h->coded_picture_number = h1->coded_picture_number;
566 h->first_field = h1->first_field;
567 h->picture_structure = h1->picture_structure;
568 h->droppable = h1->droppable;
569 h->low_delay = h1->low_delay;
571 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
572 ff_h264_unref_picture(h, &h->DPB[i]);
573 if (h1->DPB && h1->DPB[i].f.buf[0] &&
574 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
578 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
579 ff_h264_unref_picture(h, &h->cur_pic);
580 if (h1->cur_pic.f.buf[0]) {
581 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
586 h->enable_er = h1->enable_er;
587 h->workaround_bugs = h1->workaround_bugs;
588 h->low_delay = h1->low_delay;
589 h->droppable = h1->droppable;
591 // extradata/NAL handling
592 h->is_avc = h1->is_avc;
595 if ((ret = copy_parameter_set((void **)h->sps_buffers,
596 (void **)h1->sps_buffers,
597 MAX_SPS_COUNT, sizeof(SPS))) < 0)
600 if ((ret = copy_parameter_set((void **)h->pps_buffers,
601 (void **)h1->pps_buffers,
602 MAX_PPS_COUNT, sizeof(PPS))) < 0)
606 // Dequantization matrices
607 // FIXME these are big - can they be only copied when PPS changes?
608 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
610 for (i = 0; i < 6; i++)
611 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
612 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
614 for (i = 0; i < 6; i++)
615 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
616 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
618 h->dequant_coeff_pps = h1->dequant_coeff_pps;
621 copy_fields(h, h1, poc_lsb, default_ref_list);
624 copy_fields(h, h1, short_ref, current_slice);
626 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
627 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
628 copy_picture_range(h->delayed_pic, h1->delayed_pic,
629 MAX_DELAYED_PIC_COUNT + 2, h, h1);
631 h->frame_recovered = h1->frame_recovered;
633 if (context_reinitialized)
634 ff_h264_set_parameter_from_sps(h);
640 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
641 h->prev_poc_msb = h->poc_msb;
642 h->prev_poc_lsb = h->poc_lsb;
644 h->prev_frame_num_offset = h->frame_num_offset;
645 h->prev_frame_num = h->frame_num;
646 h->outputed_poc = h->next_outputed_poc;
648 h->recovery_frame = h1->recovery_frame;
653 static int h264_frame_start(H264Context *h)
657 const int pixel_shift = h->pixel_shift;
659 1<<(h->sps.bit_depth_luma-1),
660 1<<(h->sps.bit_depth_chroma-1),
661 1<<(h->sps.bit_depth_chroma-1),
665 if (!ff_thread_can_start_frame(h->avctx)) {
666 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
670 release_unused_pictures(h, 1);
671 h->cur_pic_ptr = NULL;
673 i = find_unused_picture(h);
675 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
680 pic->reference = h->droppable ? 0 : h->picture_structure;
681 pic->f.coded_picture_number = h->coded_picture_number++;
682 pic->field_picture = h->picture_structure != PICT_FRAME;
685 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
687 * See decode_nal_units().
689 pic->f.key_frame = 0;
692 pic->invalid_gap = 0;
693 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
695 if ((ret = alloc_picture(h, pic)) < 0)
697 if(!h->frame_recovered && !h->avctx->hwaccel &&
698 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
699 avpriv_color_frame(&pic->f, c);
701 h->cur_pic_ptr = pic;
702 ff_h264_unref_picture(h, &h->cur_pic);
703 if (CONFIG_ERROR_RESILIENCE) {
704 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
707 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
710 for (i = 0; i < h->nb_slice_ctx; i++) {
711 h->slice_ctx[i].linesize = h->cur_pic_ptr->f.linesize[0];
712 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f.linesize[1];
715 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
716 ff_er_frame_start(&h->slice_ctx[0].er);
717 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
718 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
721 for (i = 0; i < 16; i++) {
722 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f.linesize[0] * ((scan8[i] - scan8[0]) >> 3);
723 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f.linesize[0] * ((scan8[i] - scan8[0]) >> 3);
725 for (i = 0; i < 16; i++) {
726 h->block_offset[16 + i] =
727 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f.linesize[1] * ((scan8[i] - scan8[0]) >> 3);
728 h->block_offset[48 + 16 + i] =
729 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f.linesize[1] * ((scan8[i] - scan8[0]) >> 3);
732 /* We mark the current picture as non-reference after allocating it, so
733 * that if we break out due to an error it can be released automatically
734 * in the next ff_mpv_frame_start().
736 h->cur_pic_ptr->reference = 0;
738 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
740 h->next_output_pic = NULL;
742 assert(h->cur_pic_ptr->long_ref == 0);
747 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
749 uint8_t *src_cb, uint8_t *src_cr,
750 int linesize, int uvlinesize,
755 const int pixel_shift = h->pixel_shift;
756 int chroma444 = CHROMA444(h);
757 int chroma422 = CHROMA422(h);
760 src_cb -= uvlinesize;
761 src_cr -= uvlinesize;
763 if (!simple && FRAME_MBAFF(h)) {
766 top_border = sl->top_borders[0][sl->mb_x];
767 AV_COPY128(top_border, src_y + 15 * linesize);
769 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
770 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
773 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
774 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
775 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
776 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
778 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
779 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
781 } else if (chroma422) {
783 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
784 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
786 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
787 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
791 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
792 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
794 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
795 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
800 } else if (MB_MBAFF(sl)) {
806 top_border = sl->top_borders[top_idx][sl->mb_x];
807 /* There are two lines saved, the line above the top macroblock
808 * of a pair, and the line above the bottom macroblock. */
809 AV_COPY128(top_border, src_y + 16 * linesize);
811 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
813 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
816 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
817 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
818 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
819 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
821 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
822 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
824 } else if (chroma422) {
826 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
827 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
829 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
830 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
834 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
835 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
837 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
838 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
845 * Initialize implicit_weight table.
846 * @param field 0/1 initialize the weight for interlaced MBAFF
847 * -1 initializes the rest
849 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
851 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
853 for (i = 0; i < 2; i++) {
854 sl->luma_weight_flag[i] = 0;
855 sl->chroma_weight_flag[i] = 0;
859 if (h->picture_structure == PICT_FRAME) {
860 cur_poc = h->cur_pic_ptr->poc;
862 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
864 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
865 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
867 sl->use_weight_chroma = 0;
871 ref_count0 = sl->ref_count[0];
872 ref_count1 = sl->ref_count[1];
874 cur_poc = h->cur_pic_ptr->field_poc[field];
876 ref_count0 = 16 + 2 * sl->ref_count[0];
877 ref_count1 = 16 + 2 * sl->ref_count[1];
881 sl->use_weight_chroma = 2;
882 sl->luma_log2_weight_denom = 5;
883 sl->chroma_log2_weight_denom = 5;
885 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
886 int poc0 = sl->ref_list[0][ref0].poc;
887 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
889 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
890 int poc1 = sl->ref_list[1][ref1].poc;
891 int td = av_clip_int8(poc1 - poc0);
893 int tb = av_clip_int8(cur_poc - poc0);
894 int tx = (16384 + (FFABS(td) >> 1)) / td;
895 int dist_scale_factor = (tb * tx + 32) >> 8;
896 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
897 w = 64 - dist_scale_factor;
901 sl->implicit_weight[ref0][ref1][0] =
902 sl->implicit_weight[ref0][ref1][1] = w;
904 sl->implicit_weight[ref0][ref1][field] = w;
911 * initialize scan tables
913 static void init_scan_tables(H264Context *h)
916 for (i = 0; i < 16; i++) {
917 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
918 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
919 h->field_scan[i] = TRANSPOSE(field_scan[i]);
922 for (i = 0; i < 64; i++) {
923 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
924 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
925 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
926 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
927 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
930 if (h->sps.transform_bypass) { // FIXME same ugly
931 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
932 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
933 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
934 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
935 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
936 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
938 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
939 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
940 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
941 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
942 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
943 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
947 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
949 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
950 CONFIG_H264_VAAPI_HWACCEL + \
951 (CONFIG_H264_VDA_HWACCEL * 2) + \
952 CONFIG_H264_VDPAU_HWACCEL)
953 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
954 const enum AVPixelFormat *choices = pix_fmts;
957 switch (h->sps.bit_depth_luma) {
960 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
961 *fmt++ = AV_PIX_FMT_GBRP9;
963 *fmt++ = AV_PIX_FMT_YUV444P9;
964 } else if (CHROMA422(h))
965 *fmt++ = AV_PIX_FMT_YUV422P9;
967 *fmt++ = AV_PIX_FMT_YUV420P9;
971 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
972 *fmt++ = AV_PIX_FMT_GBRP10;
974 *fmt++ = AV_PIX_FMT_YUV444P10;
975 } else if (CHROMA422(h))
976 *fmt++ = AV_PIX_FMT_YUV422P10;
978 *fmt++ = AV_PIX_FMT_YUV420P10;
982 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
983 *fmt++ = AV_PIX_FMT_GBRP12;
985 *fmt++ = AV_PIX_FMT_YUV444P12;
986 } else if (CHROMA422(h))
987 *fmt++ = AV_PIX_FMT_YUV422P12;
989 *fmt++ = AV_PIX_FMT_YUV420P12;
993 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
994 *fmt++ = AV_PIX_FMT_GBRP14;
996 *fmt++ = AV_PIX_FMT_YUV444P14;
997 } else if (CHROMA422(h))
998 *fmt++ = AV_PIX_FMT_YUV422P14;
1000 *fmt++ = AV_PIX_FMT_YUV420P14;
1003 #if CONFIG_H264_VDPAU_HWACCEL
1004 *fmt++ = AV_PIX_FMT_VDPAU;
1007 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1008 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1009 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1010 *fmt++ = AV_PIX_FMT_GBRP;
1011 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1012 *fmt++ = AV_PIX_FMT_YUVJ444P;
1014 *fmt++ = AV_PIX_FMT_YUV444P;
1015 } else if (CHROMA422(h)) {
1016 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1017 *fmt++ = AV_PIX_FMT_YUVJ422P;
1019 *fmt++ = AV_PIX_FMT_YUV422P;
1021 #if CONFIG_H264_DXVA2_HWACCEL
1022 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1024 #if CONFIG_H264_VAAPI_HWACCEL
1025 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1027 #if CONFIG_H264_VDA_HWACCEL
1028 *fmt++ = AV_PIX_FMT_VDA_VLD;
1029 *fmt++ = AV_PIX_FMT_VDA;
1031 if (h->avctx->codec->pix_fmts)
1032 choices = h->avctx->codec->pix_fmts;
1033 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1034 *fmt++ = AV_PIX_FMT_YUVJ420P;
1036 *fmt++ = AV_PIX_FMT_YUV420P;
1040 av_log(h->avctx, AV_LOG_ERROR,
1041 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1042 return AVERROR_INVALIDDATA;
1045 *fmt = AV_PIX_FMT_NONE;
1047 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1048 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1050 return ff_thread_get_format(h->avctx, choices);
1053 /* export coded and cropped frame dimensions to AVCodecContext */
1054 static int init_dimensions(H264Context *h)
1056 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1057 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1058 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1059 h->sps.crop_right || h->sps.crop_bottom;
1060 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1061 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1063 /* handle container cropping */
1064 if (!crop_present &&
1065 FFALIGN(h->avctx->width, 16) == h->width &&
1066 FFALIGN(h->avctx->height, 16) == h->height) {
1067 width = h->avctx->width;
1068 height = h->avctx->height;
1071 if (width <= 0 || height <= 0) {
1072 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1074 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1075 return AVERROR_INVALIDDATA;
1077 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1078 h->sps.crop_bottom =
1088 h->avctx->coded_width = h->width;
1089 h->avctx->coded_height = h->height;
1090 h->avctx->width = width;
1091 h->avctx->height = height;
1096 static int h264_slice_header_init(H264Context *h, int reinit)
1098 int nb_slices = (HAVE_THREADS &&
1099 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1100 h->avctx->thread_count : 1;
1103 ff_set_sar(h->avctx, h->sps.sar);
1104 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1105 &h->chroma_x_shift, &h->chroma_y_shift);
1107 if (h->sps.timing_info_present_flag) {
1108 int64_t den = h->sps.time_scale;
1109 if (h->x264_build < 44U)
1111 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1112 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1116 ff_h264_free_tables(h, 0);
1118 h->prev_interlaced_frame = 1;
1120 init_scan_tables(h);
1121 ret = ff_h264_alloc_tables(h);
1123 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1127 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1130 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1132 max_slices = H264_MAX_THREADS;
1133 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1134 " reducing to %d\n", nb_slices, max_slices);
1135 nb_slices = max_slices;
1137 h->slice_context_count = nb_slices;
1139 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1140 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1142 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1146 for (i = 0; i < h->slice_context_count; i++) {
1147 H264SliceContext *sl = &h->slice_ctx[i];
1150 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1151 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1152 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1154 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1155 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1161 h->context_initialized = 1;
1165 ff_h264_free_tables(h, 0);
1166 h->context_initialized = 0;
1170 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1173 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1174 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1175 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1182 * Decode a slice header.
1183 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1185 * @param h h264context
1187 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1189 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1191 unsigned int first_mb_in_slice;
1192 unsigned int pps_id;
1194 unsigned int slice_type, tmp, i, j;
1195 int last_pic_structure, last_pic_droppable;
1197 int needs_reinit = 0;
1198 int field_pic_flag, bottom_field_flag;
1199 int first_slice = sl == h->slice_ctx && !h->current_slice;
1200 int frame_num, picture_structure, droppable;
1201 int mb_aff_frame, last_mb_aff_frame;
1204 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1205 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1207 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1209 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1210 if (h->current_slice) {
1211 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1212 ff_h264_field_end(h, h->slice_ctx, 1);
1213 h->current_slice = 0;
1214 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == NAL_IDR_SLICE) {
1215 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1216 ff_h264_field_end(h, h->slice_ctx, 1);
1217 h->current_slice = 0;
1218 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1219 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1220 h->cur_pic_ptr = NULL;
1222 return AVERROR_INVALIDDATA;
1225 if (!h->first_field) {
1226 if (h->cur_pic_ptr && !h->droppable) {
1227 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1228 h->picture_structure == PICT_BOTTOM_FIELD);
1230 h->cur_pic_ptr = NULL;
1234 slice_type = get_ue_golomb_31(&sl->gb);
1235 if (slice_type > 9) {
1236 av_log(h->avctx, AV_LOG_ERROR,
1237 "slice type %d too large at %d\n",
1238 slice_type, first_mb_in_slice);
1239 return AVERROR_INVALIDDATA;
1241 if (slice_type > 4) {
1243 sl->slice_type_fixed = 1;
1245 sl->slice_type_fixed = 0;
1247 slice_type = golomb_to_pict_type[slice_type];
1249 sl->slice_type = slice_type;
1250 sl->slice_type_nos = slice_type & 3;
1252 if (h->nal_unit_type == NAL_IDR_SLICE &&
1253 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1254 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1255 return AVERROR_INVALIDDATA;
1259 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1260 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1261 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1262 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1263 h->avctx->skip_frame >= AVDISCARD_ALL) {
1264 return SLICE_SKIPED;
1267 // to make a few old functions happy, it's wrong though
1268 h->pict_type = sl->slice_type;
1270 pps_id = get_ue_golomb(&sl->gb);
1271 if (pps_id >= MAX_PPS_COUNT) {
1272 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1273 return AVERROR_INVALIDDATA;
1275 if (!h->pps_buffers[pps_id]) {
1276 av_log(h->avctx, AV_LOG_ERROR,
1277 "non-existing PPS %u referenced\n",
1279 return AVERROR_INVALIDDATA;
1281 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1282 av_log(h->avctx, AV_LOG_ERROR,
1283 "PPS change from %d to %d forbidden\n",
1284 h->au_pps_id, pps_id);
1285 return AVERROR_INVALIDDATA;
1288 pps = h->pps_buffers[pps_id];
1290 if (!h->sps_buffers[pps->sps_id]) {
1291 av_log(h->avctx, AV_LOG_ERROR,
1292 "non-existing SPS %u referenced\n",
1294 return AVERROR_INVALIDDATA;
1297 h->pps = *h->pps_buffers[pps_id];
1299 if (pps->sps_id != h->sps.sps_id ||
1300 pps->sps_id != h->current_sps_id ||
1301 h->sps_buffers[pps->sps_id]->new) {
1304 av_log(h->avctx, AV_LOG_ERROR,
1305 "SPS changed in the middle of the frame\n");
1306 return AVERROR_INVALIDDATA;
1309 h->sps = *h->sps_buffers[h->pps.sps_id];
1311 if (h->mb_width != h->sps.mb_width ||
1312 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1313 h->cur_bit_depth_luma != h->sps.bit_depth_luma ||
1314 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1318 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1319 h->chroma_format_idc != h->sps.chroma_format_idc) {
1320 h->bit_depth_luma = h->sps.bit_depth_luma;
1321 h->chroma_format_idc = h->sps.chroma_format_idc;
1324 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1328 h->avctx->profile = ff_h264_get_profile(&h->sps);
1329 h->avctx->level = h->sps.level_idc;
1330 h->avctx->refs = h->sps.ref_frame_count;
1332 must_reinit = (h->context_initialized &&
1333 ( 16*h->sps.mb_width != h->avctx->coded_width
1334 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1335 || h->cur_bit_depth_luma != h->sps.bit_depth_luma
1336 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1337 || h->mb_width != h->sps.mb_width
1338 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1340 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1341 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1344 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1347 h->mb_width = h->sps.mb_width;
1348 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1349 h->mb_num = h->mb_width * h->mb_height;
1350 h->mb_stride = h->mb_width + 1;
1352 h->b_stride = h->mb_width * 4;
1354 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1356 h->width = 16 * h->mb_width;
1357 h->height = 16 * h->mb_height;
1359 ret = init_dimensions(h);
1363 if (h->sps.video_signal_type_present_flag) {
1364 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1366 if (h->sps.colour_description_present_flag) {
1367 if (h->avctx->colorspace != h->sps.colorspace)
1369 h->avctx->color_primaries = h->sps.color_primaries;
1370 h->avctx->color_trc = h->sps.color_trc;
1371 h->avctx->colorspace = h->sps.colorspace;
1375 if (h->context_initialized &&
1376 (must_reinit || needs_reinit)) {
1377 if (sl != h->slice_ctx) {
1378 av_log(h->avctx, AV_LOG_ERROR,
1379 "changing width %d -> %d / height %d -> %d on "
1381 h->width, h->avctx->coded_width,
1382 h->height, h->avctx->coded_height,
1383 h->current_slice + 1);
1384 return AVERROR_INVALIDDATA;
1387 av_assert1(first_slice);
1389 ff_h264_flush_change(h);
1391 if ((ret = get_pixel_format(h, 1)) < 0)
1393 h->avctx->pix_fmt = ret;
1395 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1396 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1398 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1399 av_log(h->avctx, AV_LOG_ERROR,
1400 "h264_slice_header_init() failed\n");
1404 if (!h->context_initialized) {
1405 if (sl != h->slice_ctx) {
1406 av_log(h->avctx, AV_LOG_ERROR,
1407 "Cannot (re-)initialize context during parallel decoding.\n");
1408 return AVERROR_PATCHWELCOME;
1411 if ((ret = get_pixel_format(h, 1)) < 0)
1413 h->avctx->pix_fmt = ret;
1415 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1416 av_log(h->avctx, AV_LOG_ERROR,
1417 "h264_slice_header_init() failed\n");
1422 if (first_slice && h->dequant_coeff_pps != pps_id) {
1423 h->dequant_coeff_pps = pps_id;
1424 ff_h264_init_dequant_tables(h);
1427 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1429 if (h->frame_num != frame_num) {
1430 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1431 h->frame_num, frame_num);
1432 return AVERROR_INVALIDDATA;
1438 last_mb_aff_frame = h->mb_aff_frame;
1439 last_pic_structure = h->picture_structure;
1440 last_pic_droppable = h->droppable;
1441 droppable = h->nal_ref_idc == 0;
1442 if (h->sps.frame_mbs_only_flag) {
1443 picture_structure = PICT_FRAME;
1445 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1446 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1449 field_pic_flag = get_bits1(&sl->gb);
1451 if (field_pic_flag) {
1452 bottom_field_flag = get_bits1(&sl->gb);
1453 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1455 picture_structure = PICT_FRAME;
1456 mb_aff_frame = h->sps.mb_aff;
1459 if (h->current_slice) {
1460 if (last_pic_structure != picture_structure ||
1461 last_pic_droppable != droppable ||
1462 last_mb_aff_frame != mb_aff_frame) {
1463 av_log(h->avctx, AV_LOG_ERROR,
1464 "Changing field mode (%d -> %d) between slices is not allowed\n",
1465 last_pic_structure, h->picture_structure);
1466 return AVERROR_INVALIDDATA;
1467 } else if (!h->cur_pic_ptr) {
1468 av_log(h->avctx, AV_LOG_ERROR,
1469 "unset cur_pic_ptr on slice %d\n",
1470 h->current_slice + 1);
1471 return AVERROR_INVALIDDATA;
1475 h->picture_structure = picture_structure;
1476 h->droppable = droppable;
1477 h->frame_num = frame_num;
1478 h->mb_aff_frame = mb_aff_frame;
1479 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1481 if (h->current_slice == 0) {
1482 /* Shorten frame num gaps so we don't have to allocate reference
1483 * frames just to throw them away */
1484 if (h->frame_num != h->prev_frame_num) {
1485 int unwrap_prev_frame_num = h->prev_frame_num;
1486 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1488 if (unwrap_prev_frame_num > h->frame_num)
1489 unwrap_prev_frame_num -= max_frame_num;
1491 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1492 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1493 if (unwrap_prev_frame_num < 0)
1494 unwrap_prev_frame_num += max_frame_num;
1496 h->prev_frame_num = unwrap_prev_frame_num;
1500 /* See if we have a decoded first field looking for a pair...
1501 * Here, we're using that to see if we should mark previously
1502 * decode frames as "finished".
1503 * We have to do that before the "dummy" in-between frame allocation,
1504 * since that can modify h->cur_pic_ptr. */
1505 if (h->first_field) {
1506 assert(h->cur_pic_ptr);
1507 assert(h->cur_pic_ptr->f.buf[0]);
1508 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1510 /* Mark old field/frame as completed */
1511 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1512 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1513 last_pic_structure == PICT_BOTTOM_FIELD);
1516 /* figure out if we have a complementary field pair */
1517 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1518 /* Previous field is unmatched. Don't display it, but let it
1519 * remain for reference if marked as such. */
1520 if (last_pic_structure != PICT_FRAME) {
1521 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1522 last_pic_structure == PICT_TOP_FIELD);
1525 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1526 /* This and previous field were reference, but had
1527 * different frame_nums. Consider this field first in
1528 * pair. Throw away previous field except for reference
1530 if (last_pic_structure != PICT_FRAME) {
1531 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1532 last_pic_structure == PICT_TOP_FIELD);
1535 /* Second field in complementary pair */
1536 if (!((last_pic_structure == PICT_TOP_FIELD &&
1537 h->picture_structure == PICT_BOTTOM_FIELD) ||
1538 (last_pic_structure == PICT_BOTTOM_FIELD &&
1539 h->picture_structure == PICT_TOP_FIELD))) {
1540 av_log(h->avctx, AV_LOG_ERROR,
1541 "Invalid field mode combination %d/%d\n",
1542 last_pic_structure, h->picture_structure);
1543 h->picture_structure = last_pic_structure;
1544 h->droppable = last_pic_droppable;
1545 return AVERROR_INVALIDDATA;
1546 } else if (last_pic_droppable != h->droppable) {
1547 avpriv_request_sample(h->avctx,
1548 "Found reference and non-reference fields in the same frame, which");
1549 h->picture_structure = last_pic_structure;
1550 h->droppable = last_pic_droppable;
1551 return AVERROR_PATCHWELCOME;
1557 while (h->frame_num != h->prev_frame_num && !h->first_field &&
1558 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1559 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1560 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1561 h->frame_num, h->prev_frame_num);
1562 if (!h->sps.gaps_in_frame_num_allowed_flag)
1563 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1564 h->last_pocs[i] = INT_MIN;
1565 ret = h264_frame_start(h);
1571 h->prev_frame_num++;
1572 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1573 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1574 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1575 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1576 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1577 ret = ff_generate_sliding_window_mmcos(h, 1);
1578 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1580 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1581 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1583 /* Error concealment: If a ref is missing, copy the previous ref
1585 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1586 * many assumptions about there being no actual duplicates.
1587 * FIXME: This does not copy padding for out-of-frame motion
1588 * vectors. Given we are concealing a lost frame, this probably
1589 * is not noticeable by comparison, but it should be fixed. */
1590 if (h->short_ref_count) {
1592 av_image_copy(h->short_ref[0]->f.data,
1593 h->short_ref[0]->f.linesize,
1594 (const uint8_t **)prev->f.data,
1599 h->short_ref[0]->poc = prev->poc + 2;
1601 h->short_ref[0]->frame_num = h->prev_frame_num;
1605 /* See if we have a decoded first field looking for a pair...
1606 * We're using that to see whether to continue decoding in that
1607 * frame, or to allocate a new one. */
1608 if (h->first_field) {
1609 assert(h->cur_pic_ptr);
1610 assert(h->cur_pic_ptr->f.buf[0]);
1611 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1613 /* figure out if we have a complementary field pair */
1614 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1615 /* Previous field is unmatched. Don't display it, but let it
1616 * remain for reference if marked as such. */
1617 h->missing_fields ++;
1618 h->cur_pic_ptr = NULL;
1619 h->first_field = FIELD_PICTURE(h);
1621 h->missing_fields = 0;
1622 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1623 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1624 h->picture_structure==PICT_BOTTOM_FIELD);
1625 /* This and the previous field had different frame_nums.
1626 * Consider this field first in pair. Throw away previous
1627 * one except for reference purposes. */
1629 h->cur_pic_ptr = NULL;
1631 /* Second field in complementary pair */
1636 /* Frame or first field in a potentially complementary pair */
1637 h->first_field = FIELD_PICTURE(h);
1640 if (!FIELD_PICTURE(h) || h->first_field) {
1641 if (h264_frame_start(h) < 0) {
1643 return AVERROR_INVALIDDATA;
1646 release_unused_pictures(h, 0);
1648 /* Some macroblocks can be accessed before they're available in case
1649 * of lost slices, MBAFF or threading. */
1650 if (FIELD_PICTURE(h)) {
1651 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1652 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1654 memset(h->slice_table, -1,
1655 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1657 h->last_slice_type = -1;
1661 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1663 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1664 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1665 first_mb_in_slice >= h->mb_num) {
1666 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1667 return AVERROR_INVALIDDATA;
1669 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1670 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1671 FIELD_OR_MBAFF_PICTURE(h);
1672 if (h->picture_structure == PICT_BOTTOM_FIELD)
1673 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1674 av_assert1(sl->mb_y < h->mb_height);
1676 if (h->picture_structure == PICT_FRAME) {
1677 h->curr_pic_num = h->frame_num;
1678 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1680 h->curr_pic_num = 2 * h->frame_num + 1;
1681 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1684 if (h->nal_unit_type == NAL_IDR_SLICE)
1685 get_ue_golomb(&sl->gb); /* idr_pic_id */
1687 if (h->sps.poc_type == 0) {
1688 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1690 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1691 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1694 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1695 h->delta_poc[0] = get_se_golomb(&sl->gb);
1697 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1698 h->delta_poc[1] = get_se_golomb(&sl->gb);
1701 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1703 if (h->pps.redundant_pic_cnt_present)
1704 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1706 ret = ff_set_ref_count(h, sl);
1710 if (slice_type != AV_PICTURE_TYPE_I &&
1711 (h->current_slice == 0 ||
1712 slice_type != h->last_slice_type ||
1713 memcmp(h->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1715 ff_h264_fill_default_ref_list(h, sl);
1718 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1719 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1721 sl->ref_count[1] = sl->ref_count[0] = 0;
1726 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1727 (h->pps.weighted_bipred_idc == 1 &&
1728 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1729 ff_pred_weight_table(h, sl);
1730 else if (h->pps.weighted_bipred_idc == 2 &&
1731 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1732 implicit_weight_table(h, sl, -1);
1735 for (i = 0; i < 2; i++) {
1736 sl->luma_weight_flag[i] = 0;
1737 sl->chroma_weight_flag[i] = 0;
1741 // If frame-mt is enabled, only update mmco tables for the first slice
1742 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1743 // or h->mmco, which will cause ref list mix-ups and decoding errors
1744 // further down the line. This may break decoding if the first slice is
1745 // corrupt, thus we only do this if frame-mt is enabled.
1746 if (h->nal_ref_idc) {
1747 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1748 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1749 h->current_slice == 0);
1750 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1751 return AVERROR_INVALIDDATA;
1754 if (FRAME_MBAFF(h)) {
1755 ff_h264_fill_mbaff_ref_list(h, sl);
1757 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1758 implicit_weight_table(h, sl, 0);
1759 implicit_weight_table(h, sl, 1);
1763 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1764 ff_h264_direct_dist_scale_factor(h, sl);
1765 ff_h264_direct_ref_list_init(h, sl);
1767 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1768 tmp = get_ue_golomb_31(&sl->gb);
1770 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1771 return AVERROR_INVALIDDATA;
1773 sl->cabac_init_idc = tmp;
1776 sl->last_qscale_diff = 0;
1777 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1778 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1779 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1780 return AVERROR_INVALIDDATA;
1783 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1784 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1785 // FIXME qscale / qp ... stuff
1786 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1787 get_bits1(&sl->gb); /* sp_for_switch_flag */
1788 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1789 sl->slice_type == AV_PICTURE_TYPE_SI)
1790 get_se_golomb(&sl->gb); /* slice_qs_delta */
1792 sl->deblocking_filter = 1;
1793 sl->slice_alpha_c0_offset = 0;
1794 sl->slice_beta_offset = 0;
1795 if (h->pps.deblocking_filter_parameters_present) {
1796 tmp = get_ue_golomb_31(&sl->gb);
1798 av_log(h->avctx, AV_LOG_ERROR,
1799 "deblocking_filter_idc %u out of range\n", tmp);
1800 return AVERROR_INVALIDDATA;
1802 sl->deblocking_filter = tmp;
1803 if (sl->deblocking_filter < 2)
1804 sl->deblocking_filter ^= 1; // 1<->0
1806 if (sl->deblocking_filter) {
1807 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1808 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1809 if (sl->slice_alpha_c0_offset > 12 ||
1810 sl->slice_alpha_c0_offset < -12 ||
1811 sl->slice_beta_offset > 12 ||
1812 sl->slice_beta_offset < -12) {
1813 av_log(h->avctx, AV_LOG_ERROR,
1814 "deblocking filter parameters %d %d out of range\n",
1815 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1816 return AVERROR_INVALIDDATA;
1821 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1822 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1823 h->nal_unit_type != NAL_IDR_SLICE) ||
1824 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1825 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1826 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1827 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1828 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1829 h->nal_ref_idc == 0))
1830 sl->deblocking_filter = 0;
1832 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1833 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1834 /* Cheat slightly for speed:
1835 * Do not bother to deblock across slices. */
1836 sl->deblocking_filter = 2;
1838 h->max_contexts = 1;
1839 if (!h->single_decode_warning) {
1840 av_log(h->avctx, AV_LOG_INFO,
1841 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1842 "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"
1843 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1844 "but will generate non-standard-compliant output.\n");
1845 h->single_decode_warning = 1;
1847 if (sl != h->slice_ctx) {
1848 av_log(h->avctx, AV_LOG_ERROR,
1849 "Deblocking switched inside frame.\n");
1850 return SLICE_SINGLETHREAD;
1854 sl->qp_thresh = 15 -
1855 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1857 h->pps.chroma_qp_index_offset[0],
1858 h->pps.chroma_qp_index_offset[1]) +
1859 6 * (h->sps.bit_depth_luma - 8);
1861 h->last_slice_type = slice_type;
1862 memcpy(h->last_ref_count, sl->ref_count, sizeof(h->last_ref_count));
1863 sl->slice_num = ++h->current_slice;
1866 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1867 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1868 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1869 && sl->slice_num >= MAX_SLICES) {
1870 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1871 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);
1874 for (j = 0; j < 2; j++) {
1876 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1877 for (i = 0; i < 16; i++) {
1879 if (j < sl->list_count && i < sl->ref_count[j] &&
1880 sl->ref_list[j][i].parent->f.buf[0]) {
1882 AVBuffer *buf = sl->ref_list[j][i].parent->f.buf[0]->buffer;
1883 for (k = 0; k < h->short_ref_count; k++)
1884 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1888 for (k = 0; k < h->long_ref_count; k++)
1889 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1890 id_list[i] = h->short_ref_count + k;
1898 for (i = 0; i < 16; i++)
1899 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1901 ref2frm[18 + 1] = -1;
1902 for (i = 16; i < 48; i++)
1903 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1904 (sl->ref_list[j][i].reference & 3);
1907 h->au_pps_id = pps_id;
1909 h->sps_buffers[h->pps.sps_id]->new = 0;
1910 h->current_sps_id = h->pps.sps_id;
1912 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1913 av_log(h->avctx, AV_LOG_DEBUG,
1914 "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",
1916 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1918 av_get_picture_type_char(sl->slice_type),
1919 sl->slice_type_fixed ? " fix" : "",
1920 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1921 pps_id, h->frame_num,
1922 h->cur_pic_ptr->field_poc[0],
1923 h->cur_pic_ptr->field_poc[1],
1924 sl->ref_count[0], sl->ref_count[1],
1926 sl->deblocking_filter,
1927 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1929 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1930 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1936 int ff_h264_get_slice_type(const H264SliceContext *sl)
1938 switch (sl->slice_type) {
1939 case AV_PICTURE_TYPE_P:
1941 case AV_PICTURE_TYPE_B:
1943 case AV_PICTURE_TYPE_I:
1945 case AV_PICTURE_TYPE_SP:
1947 case AV_PICTURE_TYPE_SI:
1950 return AVERROR_INVALIDDATA;
1954 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1955 H264SliceContext *sl,
1956 int mb_type, int top_xy,
1957 int left_xy[LEFT_MBS],
1959 int left_type[LEFT_MBS],
1960 int mb_xy, int list)
1962 int b_stride = h->b_stride;
1963 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1964 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1965 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1966 if (USES_LIST(top_type, list)) {
1967 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1968 const int b8_xy = 4 * top_xy + 2;
1969 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1970 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1971 ref_cache[0 - 1 * 8] =
1972 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1973 ref_cache[2 - 1 * 8] =
1974 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1976 AV_ZERO128(mv_dst - 1 * 8);
1977 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1980 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1981 if (USES_LIST(left_type[LTOP], list)) {
1982 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1983 const int b8_xy = 4 * left_xy[LTOP] + 1;
1984 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1985 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1986 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1987 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1988 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1990 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1991 ref_cache[-1 + 16] =
1992 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1994 AV_ZERO32(mv_dst - 1 + 0);
1995 AV_ZERO32(mv_dst - 1 + 8);
1996 AV_ZERO32(mv_dst - 1 + 16);
1997 AV_ZERO32(mv_dst - 1 + 24);
2000 ref_cache[-1 + 16] =
2001 ref_cache[-1 + 24] = LIST_NOT_USED;
2006 if (!USES_LIST(mb_type, list)) {
2007 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2008 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2009 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2010 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2011 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2016 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2017 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2018 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2019 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2020 AV_WN32A(&ref_cache[0 * 8], ref01);
2021 AV_WN32A(&ref_cache[1 * 8], ref01);
2022 AV_WN32A(&ref_cache[2 * 8], ref23);
2023 AV_WN32A(&ref_cache[3 * 8], ref23);
2027 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2028 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2029 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2030 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2031 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2037 * @return non zero if the loop filter can be skipped
2039 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2041 const int mb_xy = sl->mb_xy;
2042 int top_xy, left_xy[LEFT_MBS];
2043 int top_type, left_type[LEFT_MBS];
2047 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2049 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2050 * stuff, I can't imagine that these complex rules are worth it. */
2052 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2053 if (FRAME_MBAFF(h)) {
2054 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2055 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2057 if (left_mb_field_flag != curr_mb_field_flag)
2058 left_xy[LTOP] -= h->mb_stride;
2060 if (curr_mb_field_flag)
2061 top_xy += h->mb_stride &
2062 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2063 if (left_mb_field_flag != curr_mb_field_flag)
2064 left_xy[LBOT] += h->mb_stride;
2068 sl->top_mb_xy = top_xy;
2069 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2070 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2072 /* For sufficiently low qp, filtering wouldn't do anything.
2073 * This is a conservative estimate: could also check beta_offset
2074 * and more accurate chroma_qp. */
2075 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2076 int qp = h->cur_pic.qscale_table[mb_xy];
2077 if (qp <= qp_thresh &&
2078 (left_xy[LTOP] < 0 ||
2079 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2081 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2082 if (!FRAME_MBAFF(h))
2084 if ((left_xy[LTOP] < 0 ||
2085 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2086 (top_xy < h->mb_stride ||
2087 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2092 top_type = h->cur_pic.mb_type[top_xy];
2093 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2094 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2095 if (sl->deblocking_filter == 2) {
2096 if (h->slice_table[top_xy] != sl->slice_num)
2098 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2099 left_type[LTOP] = left_type[LBOT] = 0;
2101 if (h->slice_table[top_xy] == 0xFFFF)
2103 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2104 left_type[LTOP] = left_type[LBOT] = 0;
2106 sl->top_type = top_type;
2107 sl->left_type[LTOP] = left_type[LTOP];
2108 sl->left_type[LBOT] = left_type[LBOT];
2110 if (IS_INTRA(mb_type))
2113 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2114 top_type, left_type, mb_xy, 0);
2115 if (sl->list_count == 2)
2116 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2117 top_type, left_type, mb_xy, 1);
2119 nnz = h->non_zero_count[mb_xy];
2120 nnz_cache = sl->non_zero_count_cache;
2121 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2122 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2123 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2124 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2125 sl->cbp = h->cbp_table[mb_xy];
2128 nnz = h->non_zero_count[top_xy];
2129 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2132 if (left_type[LTOP]) {
2133 nnz = h->non_zero_count[left_xy[LTOP]];
2134 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2135 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2136 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2137 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2140 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2141 * from what the loop filter needs */
2142 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2143 if (IS_8x8DCT(top_type)) {
2144 nnz_cache[4 + 8 * 0] =
2145 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2146 nnz_cache[6 + 8 * 0] =
2147 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2149 if (IS_8x8DCT(left_type[LTOP])) {
2150 nnz_cache[3 + 8 * 1] =
2151 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2153 if (IS_8x8DCT(left_type[LBOT])) {
2154 nnz_cache[3 + 8 * 3] =
2155 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2158 if (IS_8x8DCT(mb_type)) {
2159 nnz_cache[scan8[0]] =
2160 nnz_cache[scan8[1]] =
2161 nnz_cache[scan8[2]] =
2162 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2164 nnz_cache[scan8[0 + 4]] =
2165 nnz_cache[scan8[1 + 4]] =
2166 nnz_cache[scan8[2 + 4]] =
2167 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2169 nnz_cache[scan8[0 + 8]] =
2170 nnz_cache[scan8[1 + 8]] =
2171 nnz_cache[scan8[2 + 8]] =
2172 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2174 nnz_cache[scan8[0 + 12]] =
2175 nnz_cache[scan8[1 + 12]] =
2176 nnz_cache[scan8[2 + 12]] =
2177 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2184 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2186 uint8_t *dest_y, *dest_cb, *dest_cr;
2187 int linesize, uvlinesize, mb_x, mb_y;
2188 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2189 const int old_slice_type = sl->slice_type;
2190 const int pixel_shift = h->pixel_shift;
2191 const int block_h = 16 >> h->chroma_y_shift;
2193 if (sl->deblocking_filter) {
2194 for (mb_x = start_x; mb_x < end_x; mb_x++)
2195 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2197 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2198 sl->slice_num = h->slice_table[mb_xy];
2199 mb_type = h->cur_pic.mb_type[mb_xy];
2200 sl->list_count = h->list_counts[mb_xy];
2204 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2208 dest_y = h->cur_pic.f.data[0] +
2209 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2210 dest_cb = h->cur_pic.f.data[1] +
2211 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2212 mb_y * sl->uvlinesize * block_h;
2213 dest_cr = h->cur_pic.f.data[2] +
2214 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2215 mb_y * sl->uvlinesize * block_h;
2216 // FIXME simplify above
2219 linesize = sl->mb_linesize = sl->linesize * 2;
2220 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2221 if (mb_y & 1) { // FIXME move out of this function?
2222 dest_y -= sl->linesize * 15;
2223 dest_cb -= sl->uvlinesize * (block_h - 1);
2224 dest_cr -= sl->uvlinesize * (block_h - 1);
2227 linesize = sl->mb_linesize = sl->linesize;
2228 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2230 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2232 if (fill_filter_caches(h, sl, mb_type))
2234 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2235 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2237 if (FRAME_MBAFF(h)) {
2238 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2239 linesize, uvlinesize);
2241 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2242 dest_cr, linesize, uvlinesize);
2246 sl->slice_type = old_slice_type;
2248 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2249 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2250 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2253 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2255 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2256 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2257 h->cur_pic.mb_type[mb_xy - 1] :
2258 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2259 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2260 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2264 * Draw edges and report progress for the last MB row.
2266 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2268 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2269 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2270 int height = 16 << FRAME_MBAFF(h);
2271 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2273 if (sl->deblocking_filter) {
2274 if ((top + height) >= pic_height)
2275 height += deblock_border;
2276 top -= deblock_border;
2279 if (top >= pic_height || (top + height) < 0)
2282 height = FFMIN(height, pic_height - top);
2284 height = top + height;
2288 ff_h264_draw_horiz_band(h, sl, top, height);
2290 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2293 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2294 h->picture_structure == PICT_BOTTOM_FIELD);
2297 static void er_add_slice(H264SliceContext *sl,
2298 int startx, int starty,
2299 int endx, int endy, int status)
2301 if (!sl->h264->enable_er)
2304 if (CONFIG_ERROR_RESILIENCE) {
2305 ERContext *er = &sl->h264->slice_ctx[0].er;
2307 ff_er_add_slice(er, startx, starty, endx, endy, status);
2311 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2313 H264SliceContext *sl = arg;
2314 const H264Context *h = sl->h264;
2315 int lf_x_start = sl->mb_x;
2318 sl->linesize = h->cur_pic_ptr->f.linesize[0];
2319 sl->uvlinesize = h->cur_pic_ptr->f.linesize[1];
2321 ret = alloc_scratch_buffers(sl, sl->linesize);
2325 sl->mb_skip_run = -1;
2327 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2329 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2330 avctx->codec_id != AV_CODEC_ID_H264 ||
2331 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2333 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2334 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2336 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2337 prev_status &= ~ VP_START;
2338 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2339 h->slice_ctx[0].er.error_occurred = 1;
2345 align_get_bits(&sl->gb);
2348 ff_init_cabac_decoder(&sl->cabac,
2349 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2350 (get_bits_left(&sl->gb) + 7) / 8);
2352 ff_h264_init_cabac_states(h, sl);
2358 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->mb_index_end) {
2359 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps next at %d\n",
2361 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2362 sl->mb_y, ER_MB_ERROR);
2363 return AVERROR_INVALIDDATA;
2366 ret = ff_h264_decode_mb_cabac(h, sl);
2367 // STOP_TIMER("decode_mb_cabac")
2370 ff_h264_hl_decode_mb(h, sl);
2372 // FIXME optimal? or let mb_decode decode 16x32 ?
2373 if (ret >= 0 && FRAME_MBAFF(h)) {
2376 ret = ff_h264_decode_mb_cabac(h, sl);
2379 ff_h264_hl_decode_mb(h, sl);
2382 eos = get_cabac_terminate(&sl->cabac);
2384 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2385 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2386 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2387 sl->mb_y, ER_MB_END);
2388 if (sl->mb_x >= lf_x_start)
2389 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2392 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2393 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2394 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2395 av_log(h->avctx, AV_LOG_ERROR,
2396 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2398 sl->cabac.bytestream_end - sl->cabac.bytestream);
2399 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2400 sl->mb_y, ER_MB_ERROR);
2401 return AVERROR_INVALIDDATA;
2404 if (++sl->mb_x >= h->mb_width) {
2405 loop_filter(h, sl, lf_x_start, sl->mb_x);
2406 sl->mb_x = lf_x_start = 0;
2407 decode_finish_row(h, sl);
2409 if (FIELD_OR_MBAFF_PICTURE(h)) {
2411 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2412 predict_field_decoding_flag(h, sl);
2416 if (eos || sl->mb_y >= h->mb_height) {
2417 tprintf(h->avctx, "slice end %d %d\n",
2418 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2419 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2420 sl->mb_y, ER_MB_END);
2421 if (sl->mb_x > lf_x_start)
2422 loop_filter(h, sl, lf_x_start, sl->mb_x);
2430 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->mb_index_end) {
2431 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps next at %d\n",
2433 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2434 sl->mb_y, ER_MB_ERROR);
2435 return AVERROR_INVALIDDATA;
2438 ret = ff_h264_decode_mb_cavlc(h, sl);
2441 ff_h264_hl_decode_mb(h, sl);
2443 // FIXME optimal? or let mb_decode decode 16x32 ?
2444 if (ret >= 0 && FRAME_MBAFF(h)) {
2446 ret = ff_h264_decode_mb_cavlc(h, sl);
2449 ff_h264_hl_decode_mb(h, sl);
2454 av_log(h->avctx, AV_LOG_ERROR,
2455 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2456 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2457 sl->mb_y, ER_MB_ERROR);
2461 if (++sl->mb_x >= h->mb_width) {
2462 loop_filter(h, sl, lf_x_start, sl->mb_x);
2463 sl->mb_x = lf_x_start = 0;
2464 decode_finish_row(h, sl);
2466 if (FIELD_OR_MBAFF_PICTURE(h)) {
2468 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2469 predict_field_decoding_flag(h, sl);
2471 if (sl->mb_y >= h->mb_height) {
2472 tprintf(h->avctx, "slice end %d %d\n",
2473 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2475 if ( get_bits_left(&sl->gb) == 0
2476 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2477 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2478 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2482 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2483 sl->mb_x, sl->mb_y, ER_MB_END);
2485 return AVERROR_INVALIDDATA;
2490 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2491 tprintf(h->avctx, "slice end %d %d\n",
2492 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2494 if (get_bits_left(&sl->gb) == 0) {
2495 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2496 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2497 if (sl->mb_x > lf_x_start)
2498 loop_filter(h, sl, lf_x_start, sl->mb_x);
2502 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2503 sl->mb_y, ER_MB_ERROR);
2505 return AVERROR_INVALIDDATA;
2513 * Call decode_slice() for each context.
2515 * @param h h264 master context
2516 * @param context_count number of contexts to execute
2518 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2520 AVCodecContext *const avctx = h->avctx;
2521 H264SliceContext *sl;
2524 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2526 h->slice_ctx[0].mb_index_end = INT_MAX;
2528 if (h->avctx->hwaccel ||
2529 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2531 if (context_count == 1) {
2532 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2533 h->mb_y = h->slice_ctx[0].mb_y;
2537 av_assert0(context_count > 0);
2538 for (i = 0; i < context_count; i++) {
2539 int mb_index_end = h->mb_width * h->mb_height;
2540 sl = &h->slice_ctx[i];
2541 mb_index = sl->resync_mb_x + sl->resync_mb_y * h->mb_width;
2542 if (CONFIG_ERROR_RESILIENCE) {
2543 sl->er.error_count = 0;
2545 for (j = 0; j < context_count; j++) {
2546 H264SliceContext *sl2 = &h->slice_ctx[j];
2547 int mb_index2 = sl2->resync_mb_x + sl2->resync_mb_y * h->mb_width;
2549 if (i==j || mb_index > mb_index2)
2551 mb_index_end = FFMIN(mb_index_end, mb_index2);
2553 sl->mb_index_end = mb_index_end;
2556 avctx->execute(avctx, decode_slice, h->slice_ctx,
2557 NULL, context_count, sizeof(h->slice_ctx[0]));
2559 /* pull back stuff from slices to master context */
2560 sl = &h->slice_ctx[context_count - 1];
2562 if (CONFIG_ERROR_RESILIENCE) {
2563 for (i = 1; i < context_count; i++)
2564 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;