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->workaround_bugs = h1->workaround_bugs;
587 h->low_delay = h1->low_delay;
588 h->droppable = h1->droppable;
590 // extradata/NAL handling
591 h->is_avc = h1->is_avc;
594 if ((ret = copy_parameter_set((void **)h->sps_buffers,
595 (void **)h1->sps_buffers,
596 MAX_SPS_COUNT, sizeof(SPS))) < 0)
599 if ((ret = copy_parameter_set((void **)h->pps_buffers,
600 (void **)h1->pps_buffers,
601 MAX_PPS_COUNT, sizeof(PPS))) < 0)
605 // Dequantization matrices
606 // FIXME these are big - can they be only copied when PPS changes?
607 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
609 for (i = 0; i < 6; i++)
610 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
611 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
613 for (i = 0; i < 6; i++)
614 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
615 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
617 h->dequant_coeff_pps = h1->dequant_coeff_pps;
620 copy_fields(h, h1, poc_lsb, default_ref_list);
623 copy_fields(h, h1, short_ref, current_slice);
625 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
626 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
627 copy_picture_range(h->delayed_pic, h1->delayed_pic,
628 MAX_DELAYED_PIC_COUNT + 2, h, h1);
630 h->frame_recovered = h1->frame_recovered;
632 if (context_reinitialized)
633 ff_h264_set_parameter_from_sps(h);
639 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
640 h->prev_poc_msb = h->poc_msb;
641 h->prev_poc_lsb = h->poc_lsb;
643 h->prev_frame_num_offset = h->frame_num_offset;
644 h->prev_frame_num = h->frame_num;
645 h->outputed_poc = h->next_outputed_poc;
647 h->recovery_frame = h1->recovery_frame;
652 static int h264_frame_start(H264Context *h)
656 const int pixel_shift = h->pixel_shift;
658 1<<(h->sps.bit_depth_luma-1),
659 1<<(h->sps.bit_depth_chroma-1),
660 1<<(h->sps.bit_depth_chroma-1),
664 if (!ff_thread_can_start_frame(h->avctx)) {
665 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
669 release_unused_pictures(h, 1);
670 h->cur_pic_ptr = NULL;
672 i = find_unused_picture(h);
674 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
679 pic->reference = h->droppable ? 0 : h->picture_structure;
680 pic->f.coded_picture_number = h->coded_picture_number++;
681 pic->field_picture = h->picture_structure != PICT_FRAME;
684 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
686 * See decode_nal_units().
688 pic->f.key_frame = 0;
691 pic->invalid_gap = 0;
692 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
694 if ((ret = alloc_picture(h, pic)) < 0)
696 if(!h->frame_recovered && !h->avctx->hwaccel &&
697 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
698 avpriv_color_frame(&pic->f, c);
700 h->cur_pic_ptr = pic;
701 ff_h264_unref_picture(h, &h->cur_pic);
702 if (CONFIG_ERROR_RESILIENCE) {
703 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
706 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
709 for (i = 0; i < h->nb_slice_ctx; i++) {
710 h->slice_ctx[i].linesize = h->cur_pic_ptr->f.linesize[0];
711 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f.linesize[1];
714 if (CONFIG_ERROR_RESILIENCE) {
715 ff_er_frame_start(&h->slice_ctx[0].er);
716 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
717 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
720 for (i = 0; i < 16; i++) {
721 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f.linesize[0] * ((scan8[i] - scan8[0]) >> 3);
722 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f.linesize[0] * ((scan8[i] - scan8[0]) >> 3);
724 for (i = 0; i < 16; i++) {
725 h->block_offset[16 + i] =
726 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f.linesize[1] * ((scan8[i] - scan8[0]) >> 3);
727 h->block_offset[48 + 16 + i] =
728 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f.linesize[1] * ((scan8[i] - scan8[0]) >> 3);
731 /* We mark the current picture as non-reference after allocating it, so
732 * that if we break out due to an error it can be released automatically
733 * in the next ff_mpv_frame_start().
735 h->cur_pic_ptr->reference = 0;
737 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
739 h->next_output_pic = NULL;
741 assert(h->cur_pic_ptr->long_ref == 0);
746 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
748 uint8_t *src_cb, uint8_t *src_cr,
749 int linesize, int uvlinesize,
754 const int pixel_shift = h->pixel_shift;
755 int chroma444 = CHROMA444(h);
756 int chroma422 = CHROMA422(h);
759 src_cb -= uvlinesize;
760 src_cr -= uvlinesize;
762 if (!simple && FRAME_MBAFF(h)) {
765 top_border = sl->top_borders[0][sl->mb_x];
766 AV_COPY128(top_border, src_y + 15 * linesize);
768 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
769 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
772 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
773 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
774 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
775 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
777 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
778 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
780 } else if (chroma422) {
782 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
783 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
785 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
786 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
790 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
791 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
793 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
794 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
799 } else if (MB_MBAFF(sl)) {
805 top_border = sl->top_borders[top_idx][sl->mb_x];
806 /* There are two lines saved, the line above the top macroblock
807 * of a pair, and the line above the bottom macroblock. */
808 AV_COPY128(top_border, src_y + 16 * linesize);
810 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
812 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
815 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
816 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
817 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
818 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
820 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
821 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
823 } else if (chroma422) {
825 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
826 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
828 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
829 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
833 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
834 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
836 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
837 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
844 * Initialize implicit_weight table.
845 * @param field 0/1 initialize the weight for interlaced MBAFF
846 * -1 initializes the rest
848 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
850 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
852 for (i = 0; i < 2; i++) {
853 sl->luma_weight_flag[i] = 0;
854 sl->chroma_weight_flag[i] = 0;
858 if (h->picture_structure == PICT_FRAME) {
859 cur_poc = h->cur_pic_ptr->poc;
861 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
863 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
864 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
866 sl->use_weight_chroma = 0;
870 ref_count0 = sl->ref_count[0];
871 ref_count1 = sl->ref_count[1];
873 cur_poc = h->cur_pic_ptr->field_poc[field];
875 ref_count0 = 16 + 2 * sl->ref_count[0];
876 ref_count1 = 16 + 2 * sl->ref_count[1];
880 sl->use_weight_chroma = 2;
881 sl->luma_log2_weight_denom = 5;
882 sl->chroma_log2_weight_denom = 5;
884 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
885 int poc0 = sl->ref_list[0][ref0].poc;
886 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
888 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
889 int poc1 = sl->ref_list[1][ref1].poc;
890 int td = av_clip_int8(poc1 - poc0);
892 int tb = av_clip_int8(cur_poc - poc0);
893 int tx = (16384 + (FFABS(td) >> 1)) / td;
894 int dist_scale_factor = (tb * tx + 32) >> 8;
895 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
896 w = 64 - dist_scale_factor;
900 sl->implicit_weight[ref0][ref1][0] =
901 sl->implicit_weight[ref0][ref1][1] = w;
903 sl->implicit_weight[ref0][ref1][field] = w;
910 * initialize scan tables
912 static void init_scan_tables(H264Context *h)
915 for (i = 0; i < 16; i++) {
916 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
917 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
918 h->field_scan[i] = TRANSPOSE(field_scan[i]);
921 for (i = 0; i < 64; i++) {
922 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
923 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
924 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
925 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
926 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
929 if (h->sps.transform_bypass) { // FIXME same ugly
930 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
931 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
932 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
933 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
934 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
935 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
937 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
938 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
939 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
940 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
941 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
942 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
946 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
948 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
949 CONFIG_H264_VAAPI_HWACCEL + \
950 (CONFIG_H264_VDA_HWACCEL * 2) + \
951 CONFIG_H264_VDPAU_HWACCEL)
952 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
953 const enum AVPixelFormat *choices = pix_fmts;
956 switch (h->sps.bit_depth_luma) {
959 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
960 *fmt++ = AV_PIX_FMT_GBRP9;
962 *fmt++ = AV_PIX_FMT_YUV444P9;
963 } else if (CHROMA422(h))
964 *fmt++ = AV_PIX_FMT_YUV422P9;
966 *fmt++ = AV_PIX_FMT_YUV420P9;
970 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
971 *fmt++ = AV_PIX_FMT_GBRP10;
973 *fmt++ = AV_PIX_FMT_YUV444P10;
974 } else if (CHROMA422(h))
975 *fmt++ = AV_PIX_FMT_YUV422P10;
977 *fmt++ = AV_PIX_FMT_YUV420P10;
981 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
982 *fmt++ = AV_PIX_FMT_GBRP12;
984 *fmt++ = AV_PIX_FMT_YUV444P12;
985 } else if (CHROMA422(h))
986 *fmt++ = AV_PIX_FMT_YUV422P12;
988 *fmt++ = AV_PIX_FMT_YUV420P12;
992 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
993 *fmt++ = AV_PIX_FMT_GBRP14;
995 *fmt++ = AV_PIX_FMT_YUV444P14;
996 } else if (CHROMA422(h))
997 *fmt++ = AV_PIX_FMT_YUV422P14;
999 *fmt++ = AV_PIX_FMT_YUV420P14;
1002 #if CONFIG_H264_VDPAU_HWACCEL
1003 *fmt++ = AV_PIX_FMT_VDPAU;
1006 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1007 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1008 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1009 *fmt++ = AV_PIX_FMT_GBRP;
1010 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1011 *fmt++ = AV_PIX_FMT_YUVJ444P;
1013 *fmt++ = AV_PIX_FMT_YUV444P;
1014 } else if (CHROMA422(h)) {
1015 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1016 *fmt++ = AV_PIX_FMT_YUVJ422P;
1018 *fmt++ = AV_PIX_FMT_YUV422P;
1020 #if CONFIG_H264_DXVA2_HWACCEL
1021 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1023 #if CONFIG_H264_VAAPI_HWACCEL
1024 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1026 #if CONFIG_H264_VDA_HWACCEL
1027 *fmt++ = AV_PIX_FMT_VDA_VLD;
1028 *fmt++ = AV_PIX_FMT_VDA;
1030 if (h->avctx->codec->pix_fmts)
1031 choices = h->avctx->codec->pix_fmts;
1032 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1033 *fmt++ = AV_PIX_FMT_YUVJ420P;
1035 *fmt++ = AV_PIX_FMT_YUV420P;
1039 av_log(h->avctx, AV_LOG_ERROR,
1040 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1041 return AVERROR_INVALIDDATA;
1044 *fmt = AV_PIX_FMT_NONE;
1046 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1047 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1049 return ff_thread_get_format(h->avctx, choices);
1052 /* export coded and cropped frame dimensions to AVCodecContext */
1053 static int init_dimensions(H264Context *h)
1055 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1056 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1057 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1058 h->sps.crop_right || h->sps.crop_bottom;
1059 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1060 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1062 /* handle container cropping */
1063 if (!crop_present &&
1064 FFALIGN(h->avctx->width, 16) == h->width &&
1065 FFALIGN(h->avctx->height, 16) == h->height) {
1066 width = h->avctx->width;
1067 height = h->avctx->height;
1070 if (width <= 0 || height <= 0) {
1071 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1073 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1074 return AVERROR_INVALIDDATA;
1076 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1077 h->sps.crop_bottom =
1087 h->avctx->coded_width = h->width;
1088 h->avctx->coded_height = h->height;
1089 h->avctx->width = width;
1090 h->avctx->height = height;
1095 static int h264_slice_header_init(H264Context *h, int reinit)
1097 int nb_slices = (HAVE_THREADS &&
1098 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1099 h->avctx->thread_count : 1;
1102 ff_set_sar(h->avctx, h->sps.sar);
1103 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1104 &h->chroma_x_shift, &h->chroma_y_shift);
1106 if (h->sps.timing_info_present_flag) {
1107 int64_t den = h->sps.time_scale;
1108 if (h->x264_build < 44U)
1110 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1111 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1115 ff_h264_free_tables(h, 0);
1117 h->prev_interlaced_frame = 1;
1119 init_scan_tables(h);
1120 ret = ff_h264_alloc_tables(h);
1122 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1126 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1129 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1131 max_slices = H264_MAX_THREADS;
1132 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1133 " reducing to %d\n", nb_slices, max_slices);
1134 nb_slices = max_slices;
1136 h->slice_context_count = nb_slices;
1138 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1139 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1141 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1145 for (i = 0; i < h->slice_context_count; i++) {
1146 H264SliceContext *sl = &h->slice_ctx[i];
1149 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1150 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1151 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1153 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1154 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1160 h->context_initialized = 1;
1164 ff_h264_free_tables(h, 0);
1165 h->context_initialized = 0;
1169 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1172 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1173 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1174 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1181 * Decode a slice header.
1182 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1184 * @param h h264context
1186 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1188 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1190 unsigned int first_mb_in_slice;
1191 unsigned int pps_id;
1193 unsigned int slice_type, tmp, i, j;
1194 int last_pic_structure, last_pic_droppable;
1196 int needs_reinit = 0;
1197 int field_pic_flag, bottom_field_flag;
1198 int first_slice = sl == h->slice_ctx && !h->current_slice;
1199 int frame_num, picture_structure, droppable;
1202 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1203 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1205 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1207 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1208 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1209 ff_h264_field_end(h, sl, 1);
1212 h->current_slice = 0;
1213 if (!h->first_field) {
1214 if (h->cur_pic_ptr && !h->droppable) {
1215 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1216 h->picture_structure == PICT_BOTTOM_FIELD);
1218 h->cur_pic_ptr = NULL;
1222 slice_type = get_ue_golomb_31(&sl->gb);
1223 if (slice_type > 9) {
1224 av_log(h->avctx, AV_LOG_ERROR,
1225 "slice type %d too large at %d\n",
1226 slice_type, first_mb_in_slice);
1227 return AVERROR_INVALIDDATA;
1229 if (slice_type > 4) {
1231 sl->slice_type_fixed = 1;
1233 sl->slice_type_fixed = 0;
1235 slice_type = golomb_to_pict_type[slice_type];
1237 sl->slice_type = slice_type;
1238 sl->slice_type_nos = slice_type & 3;
1240 if (h->nal_unit_type == NAL_IDR_SLICE &&
1241 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1242 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1243 return AVERROR_INVALIDDATA;
1247 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1248 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1249 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1250 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1251 h->avctx->skip_frame >= AVDISCARD_ALL) {
1252 return SLICE_SKIPED;
1255 // to make a few old functions happy, it's wrong though
1256 h->pict_type = sl->slice_type;
1258 pps_id = get_ue_golomb(&sl->gb);
1259 if (pps_id >= MAX_PPS_COUNT) {
1260 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1261 return AVERROR_INVALIDDATA;
1263 if (!h->pps_buffers[pps_id]) {
1264 av_log(h->avctx, AV_LOG_ERROR,
1265 "non-existing PPS %u referenced\n",
1267 return AVERROR_INVALIDDATA;
1269 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1270 av_log(h->avctx, AV_LOG_ERROR,
1271 "PPS change from %d to %d forbidden\n",
1272 h->au_pps_id, pps_id);
1273 return AVERROR_INVALIDDATA;
1276 pps = h->pps_buffers[pps_id];
1278 if (!h->sps_buffers[pps->sps_id]) {
1279 av_log(h->avctx, AV_LOG_ERROR,
1280 "non-existing SPS %u referenced\n",
1282 return AVERROR_INVALIDDATA;
1285 h->pps = *h->pps_buffers[pps_id];
1287 if (pps->sps_id != h->sps.sps_id ||
1288 pps->sps_id != h->current_sps_id ||
1289 h->sps_buffers[pps->sps_id]->new) {
1292 av_log(h->avctx, AV_LOG_ERROR,
1293 "SPS changed in the middle of the frame\n");
1294 return AVERROR_INVALIDDATA;
1297 h->sps = *h->sps_buffers[h->pps.sps_id];
1299 if (h->mb_width != h->sps.mb_width ||
1300 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1301 h->cur_bit_depth_luma != h->sps.bit_depth_luma ||
1302 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1306 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1307 h->chroma_format_idc != h->sps.chroma_format_idc) {
1308 h->bit_depth_luma = h->sps.bit_depth_luma;
1309 h->chroma_format_idc = h->sps.chroma_format_idc;
1312 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1316 h->avctx->profile = ff_h264_get_profile(&h->sps);
1317 h->avctx->level = h->sps.level_idc;
1318 h->avctx->refs = h->sps.ref_frame_count;
1320 must_reinit = (h->context_initialized &&
1321 ( 16*h->sps.mb_width != h->avctx->coded_width
1322 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1323 || h->cur_bit_depth_luma != h->sps.bit_depth_luma
1324 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1325 || h->mb_width != h->sps.mb_width
1326 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1328 if (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0)))
1331 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1334 h->mb_width = h->sps.mb_width;
1335 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1336 h->mb_num = h->mb_width * h->mb_height;
1337 h->mb_stride = h->mb_width + 1;
1339 h->b_stride = h->mb_width * 4;
1341 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1343 h->width = 16 * h->mb_width;
1344 h->height = 16 * h->mb_height;
1346 ret = init_dimensions(h);
1350 if (h->sps.video_signal_type_present_flag) {
1351 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1353 if (h->sps.colour_description_present_flag) {
1354 if (h->avctx->colorspace != h->sps.colorspace)
1356 h->avctx->color_primaries = h->sps.color_primaries;
1357 h->avctx->color_trc = h->sps.color_trc;
1358 h->avctx->colorspace = h->sps.colorspace;
1362 if (h->context_initialized &&
1363 (must_reinit || needs_reinit)) {
1364 if (sl != h->slice_ctx) {
1365 av_log(h->avctx, AV_LOG_ERROR,
1366 "changing width %d -> %d / height %d -> %d on "
1368 h->width, h->avctx->coded_width,
1369 h->height, h->avctx->coded_height,
1370 h->current_slice + 1);
1371 return AVERROR_INVALIDDATA;
1374 av_assert1(first_slice);
1376 ff_h264_flush_change(h);
1378 if ((ret = get_pixel_format(h, 1)) < 0)
1380 h->avctx->pix_fmt = ret;
1382 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1383 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1385 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1386 av_log(h->avctx, AV_LOG_ERROR,
1387 "h264_slice_header_init() failed\n");
1391 if (!h->context_initialized) {
1392 if (sl != h->slice_ctx) {
1393 av_log(h->avctx, AV_LOG_ERROR,
1394 "Cannot (re-)initialize context during parallel decoding.\n");
1395 return AVERROR_PATCHWELCOME;
1398 if ((ret = get_pixel_format(h, 1)) < 0)
1400 h->avctx->pix_fmt = ret;
1402 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1403 av_log(h->avctx, AV_LOG_ERROR,
1404 "h264_slice_header_init() failed\n");
1409 if (first_slice && h->dequant_coeff_pps != pps_id) {
1410 h->dequant_coeff_pps = pps_id;
1411 ff_h264_init_dequant_tables(h);
1414 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1416 if (h->frame_num != frame_num) {
1417 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1418 h->frame_num, frame_num);
1419 return AVERROR_INVALIDDATA;
1424 h->mb_aff_frame = 0;
1425 last_pic_structure = h->picture_structure;
1426 last_pic_droppable = h->droppable;
1427 droppable = h->nal_ref_idc == 0;
1428 if (h->sps.frame_mbs_only_flag) {
1429 picture_structure = PICT_FRAME;
1431 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1432 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1435 field_pic_flag = get_bits1(&sl->gb);
1437 if (field_pic_flag) {
1438 bottom_field_flag = get_bits1(&sl->gb);
1439 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1441 picture_structure = PICT_FRAME;
1442 h->mb_aff_frame = h->sps.mb_aff;
1445 if (h->current_slice) {
1446 if (last_pic_structure != picture_structure ||
1447 last_pic_droppable != droppable) {
1448 av_log(h->avctx, AV_LOG_ERROR,
1449 "Changing field mode (%d -> %d) between slices is not allowed\n",
1450 last_pic_structure, h->picture_structure);
1451 return AVERROR_INVALIDDATA;
1452 } else if (!h->cur_pic_ptr) {
1453 av_log(h->avctx, AV_LOG_ERROR,
1454 "unset cur_pic_ptr on slice %d\n",
1455 h->current_slice + 1);
1456 return AVERROR_INVALIDDATA;
1460 h->picture_structure = picture_structure;
1461 h->droppable = droppable;
1462 h->frame_num = frame_num;
1463 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1465 if (h->current_slice == 0) {
1466 /* Shorten frame num gaps so we don't have to allocate reference
1467 * frames just to throw them away */
1468 if (h->frame_num != h->prev_frame_num) {
1469 int unwrap_prev_frame_num = h->prev_frame_num;
1470 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1472 if (unwrap_prev_frame_num > h->frame_num)
1473 unwrap_prev_frame_num -= max_frame_num;
1475 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1476 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1477 if (unwrap_prev_frame_num < 0)
1478 unwrap_prev_frame_num += max_frame_num;
1480 h->prev_frame_num = unwrap_prev_frame_num;
1484 /* See if we have a decoded first field looking for a pair...
1485 * Here, we're using that to see if we should mark previously
1486 * decode frames as "finished".
1487 * We have to do that before the "dummy" in-between frame allocation,
1488 * since that can modify h->cur_pic_ptr. */
1489 if (h->first_field) {
1490 assert(h->cur_pic_ptr);
1491 assert(h->cur_pic_ptr->f.buf[0]);
1492 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1494 /* Mark old field/frame as completed */
1495 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1496 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1497 last_pic_structure == PICT_BOTTOM_FIELD);
1500 /* figure out if we have a complementary field pair */
1501 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1502 /* Previous field is unmatched. Don't display it, but let it
1503 * remain for reference if marked as such. */
1504 if (last_pic_structure != PICT_FRAME) {
1505 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1506 last_pic_structure == PICT_TOP_FIELD);
1509 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1510 /* This and previous field were reference, but had
1511 * different frame_nums. Consider this field first in
1512 * pair. Throw away previous field except for reference
1514 if (last_pic_structure != PICT_FRAME) {
1515 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1516 last_pic_structure == PICT_TOP_FIELD);
1519 /* Second field in complementary pair */
1520 if (!((last_pic_structure == PICT_TOP_FIELD &&
1521 h->picture_structure == PICT_BOTTOM_FIELD) ||
1522 (last_pic_structure == PICT_BOTTOM_FIELD &&
1523 h->picture_structure == PICT_TOP_FIELD))) {
1524 av_log(h->avctx, AV_LOG_ERROR,
1525 "Invalid field mode combination %d/%d\n",
1526 last_pic_structure, h->picture_structure);
1527 h->picture_structure = last_pic_structure;
1528 h->droppable = last_pic_droppable;
1529 return AVERROR_INVALIDDATA;
1530 } else if (last_pic_droppable != h->droppable) {
1531 avpriv_request_sample(h->avctx,
1532 "Found reference and non-reference fields in the same frame, which");
1533 h->picture_structure = last_pic_structure;
1534 h->droppable = last_pic_droppable;
1535 return AVERROR_PATCHWELCOME;
1541 while (h->frame_num != h->prev_frame_num && !h->first_field &&
1542 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1543 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1544 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1545 h->frame_num, h->prev_frame_num);
1546 if (!h->sps.gaps_in_frame_num_allowed_flag)
1547 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1548 h->last_pocs[i] = INT_MIN;
1549 ret = h264_frame_start(h);
1555 h->prev_frame_num++;
1556 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1557 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1558 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1559 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1560 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1561 ret = ff_generate_sliding_window_mmcos(h, 1);
1562 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1564 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1565 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1567 /* Error concealment: If a ref is missing, copy the previous ref
1569 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1570 * many assumptions about there being no actual duplicates.
1571 * FIXME: This does not copy padding for out-of-frame motion
1572 * vectors. Given we are concealing a lost frame, this probably
1573 * is not noticeable by comparison, but it should be fixed. */
1574 if (h->short_ref_count) {
1576 av_image_copy(h->short_ref[0]->f.data,
1577 h->short_ref[0]->f.linesize,
1578 (const uint8_t **)prev->f.data,
1583 h->short_ref[0]->poc = prev->poc + 2;
1585 h->short_ref[0]->frame_num = h->prev_frame_num;
1589 /* See if we have a decoded first field looking for a pair...
1590 * We're using that to see whether to continue decoding in that
1591 * frame, or to allocate a new one. */
1592 if (h->first_field) {
1593 assert(h->cur_pic_ptr);
1594 assert(h->cur_pic_ptr->f.buf[0]);
1595 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1597 /* figure out if we have a complementary field pair */
1598 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1599 /* Previous field is unmatched. Don't display it, but let it
1600 * remain for reference if marked as such. */
1601 h->missing_fields ++;
1602 h->cur_pic_ptr = NULL;
1603 h->first_field = FIELD_PICTURE(h);
1605 h->missing_fields = 0;
1606 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1607 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1608 h->picture_structure==PICT_BOTTOM_FIELD);
1609 /* This and the previous field had different frame_nums.
1610 * Consider this field first in pair. Throw away previous
1611 * one except for reference purposes. */
1613 h->cur_pic_ptr = NULL;
1615 /* Second field in complementary pair */
1620 /* Frame or first field in a potentially complementary pair */
1621 h->first_field = FIELD_PICTURE(h);
1624 if (!FIELD_PICTURE(h) || h->first_field) {
1625 if (h264_frame_start(h) < 0) {
1627 return AVERROR_INVALIDDATA;
1630 release_unused_pictures(h, 0);
1632 /* Some macroblocks can be accessed before they're available in case
1633 * of lost slices, MBAFF or threading. */
1634 if (FIELD_PICTURE(h)) {
1635 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1636 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1638 memset(h->slice_table, -1,
1639 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1641 h->last_slice_type = -1;
1645 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1647 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1648 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1649 first_mb_in_slice >= h->mb_num) {
1650 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1651 return AVERROR_INVALIDDATA;
1653 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1654 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1655 FIELD_OR_MBAFF_PICTURE(h);
1656 if (h->picture_structure == PICT_BOTTOM_FIELD)
1657 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1658 av_assert1(sl->mb_y < h->mb_height);
1660 if (h->picture_structure == PICT_FRAME) {
1661 h->curr_pic_num = h->frame_num;
1662 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1664 h->curr_pic_num = 2 * h->frame_num + 1;
1665 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1668 if (h->nal_unit_type == NAL_IDR_SLICE)
1669 get_ue_golomb(&sl->gb); /* idr_pic_id */
1671 if (h->sps.poc_type == 0) {
1672 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1674 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1675 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1678 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1679 h->delta_poc[0] = get_se_golomb(&sl->gb);
1681 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1682 h->delta_poc[1] = get_se_golomb(&sl->gb);
1685 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1687 if (h->pps.redundant_pic_cnt_present)
1688 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1690 ret = ff_set_ref_count(h, sl);
1694 if (slice_type != AV_PICTURE_TYPE_I &&
1695 (h->current_slice == 0 ||
1696 slice_type != h->last_slice_type ||
1697 memcmp(h->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1699 ff_h264_fill_default_ref_list(h, sl);
1702 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1703 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1705 sl->ref_count[1] = sl->ref_count[0] = 0;
1710 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1711 (h->pps.weighted_bipred_idc == 1 &&
1712 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1713 ff_pred_weight_table(h, sl);
1714 else if (h->pps.weighted_bipred_idc == 2 &&
1715 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1716 implicit_weight_table(h, sl, -1);
1719 for (i = 0; i < 2; i++) {
1720 sl->luma_weight_flag[i] = 0;
1721 sl->chroma_weight_flag[i] = 0;
1725 // If frame-mt is enabled, only update mmco tables for the first slice
1726 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1727 // or h->mmco, which will cause ref list mix-ups and decoding errors
1728 // further down the line. This may break decoding if the first slice is
1729 // corrupt, thus we only do this if frame-mt is enabled.
1730 if (h->nal_ref_idc) {
1731 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1732 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1733 h->current_slice == 0);
1734 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1735 return AVERROR_INVALIDDATA;
1738 if (FRAME_MBAFF(h)) {
1739 ff_h264_fill_mbaff_ref_list(h, sl);
1741 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1742 implicit_weight_table(h, sl, 0);
1743 implicit_weight_table(h, sl, 1);
1747 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1748 ff_h264_direct_dist_scale_factor(h, sl);
1749 ff_h264_direct_ref_list_init(h, sl);
1751 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1752 tmp = get_ue_golomb_31(&sl->gb);
1754 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1755 return AVERROR_INVALIDDATA;
1757 sl->cabac_init_idc = tmp;
1760 sl->last_qscale_diff = 0;
1761 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1762 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1763 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1764 return AVERROR_INVALIDDATA;
1767 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1768 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1769 // FIXME qscale / qp ... stuff
1770 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1771 get_bits1(&sl->gb); /* sp_for_switch_flag */
1772 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1773 sl->slice_type == AV_PICTURE_TYPE_SI)
1774 get_se_golomb(&sl->gb); /* slice_qs_delta */
1776 sl->deblocking_filter = 1;
1777 sl->slice_alpha_c0_offset = 0;
1778 sl->slice_beta_offset = 0;
1779 if (h->pps.deblocking_filter_parameters_present) {
1780 tmp = get_ue_golomb_31(&sl->gb);
1782 av_log(h->avctx, AV_LOG_ERROR,
1783 "deblocking_filter_idc %u out of range\n", tmp);
1784 return AVERROR_INVALIDDATA;
1786 sl->deblocking_filter = tmp;
1787 if (sl->deblocking_filter < 2)
1788 sl->deblocking_filter ^= 1; // 1<->0
1790 if (sl->deblocking_filter) {
1791 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1792 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1793 if (sl->slice_alpha_c0_offset > 12 ||
1794 sl->slice_alpha_c0_offset < -12 ||
1795 sl->slice_beta_offset > 12 ||
1796 sl->slice_beta_offset < -12) {
1797 av_log(h->avctx, AV_LOG_ERROR,
1798 "deblocking filter parameters %d %d out of range\n",
1799 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1800 return AVERROR_INVALIDDATA;
1805 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1806 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1807 h->nal_unit_type != NAL_IDR_SLICE) ||
1808 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1809 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1810 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1811 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1812 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1813 h->nal_ref_idc == 0))
1814 sl->deblocking_filter = 0;
1816 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1817 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1818 /* Cheat slightly for speed:
1819 * Do not bother to deblock across slices. */
1820 sl->deblocking_filter = 2;
1822 h->max_contexts = 1;
1823 if (!h->single_decode_warning) {
1824 av_log(h->avctx, AV_LOG_INFO,
1825 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1826 "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"
1827 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1828 "but will generate non-standard-compliant output.\n");
1829 h->single_decode_warning = 1;
1831 if (sl != h->slice_ctx) {
1832 av_log(h->avctx, AV_LOG_ERROR,
1833 "Deblocking switched inside frame.\n");
1834 return SLICE_SINGLETHREAD;
1838 sl->qp_thresh = 15 -
1839 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1841 h->pps.chroma_qp_index_offset[0],
1842 h->pps.chroma_qp_index_offset[1]) +
1843 6 * (h->sps.bit_depth_luma - 8);
1845 h->last_slice_type = slice_type;
1846 memcpy(h->last_ref_count, sl->ref_count, sizeof(h->last_ref_count));
1847 sl->slice_num = ++h->current_slice;
1850 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1851 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1852 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1853 && sl->slice_num >= MAX_SLICES) {
1854 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1855 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);
1858 for (j = 0; j < 2; j++) {
1860 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1861 for (i = 0; i < 16; i++) {
1863 if (j < sl->list_count && i < sl->ref_count[j] &&
1864 sl->ref_list[j][i].parent->f.buf[0]) {
1866 AVBuffer *buf = sl->ref_list[j][i].parent->f.buf[0]->buffer;
1867 for (k = 0; k < h->short_ref_count; k++)
1868 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1872 for (k = 0; k < h->long_ref_count; k++)
1873 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1874 id_list[i] = h->short_ref_count + k;
1882 for (i = 0; i < 16; i++)
1883 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1885 ref2frm[18 + 1] = -1;
1886 for (i = 16; i < 48; i++)
1887 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1888 (sl->ref_list[j][i].reference & 3);
1891 h->au_pps_id = pps_id;
1893 h->sps_buffers[h->pps.sps_id]->new = 0;
1894 h->current_sps_id = h->pps.sps_id;
1896 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1897 av_log(h->avctx, AV_LOG_DEBUG,
1898 "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",
1900 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1902 av_get_picture_type_char(sl->slice_type),
1903 sl->slice_type_fixed ? " fix" : "",
1904 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1905 pps_id, h->frame_num,
1906 h->cur_pic_ptr->field_poc[0],
1907 h->cur_pic_ptr->field_poc[1],
1908 sl->ref_count[0], sl->ref_count[1],
1910 sl->deblocking_filter,
1911 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1913 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1914 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1920 int ff_h264_get_slice_type(const H264SliceContext *sl)
1922 switch (sl->slice_type) {
1923 case AV_PICTURE_TYPE_P:
1925 case AV_PICTURE_TYPE_B:
1927 case AV_PICTURE_TYPE_I:
1929 case AV_PICTURE_TYPE_SP:
1931 case AV_PICTURE_TYPE_SI:
1934 return AVERROR_INVALIDDATA;
1938 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1939 H264SliceContext *sl,
1940 int mb_type, int top_xy,
1941 int left_xy[LEFT_MBS],
1943 int left_type[LEFT_MBS],
1944 int mb_xy, int list)
1946 int b_stride = h->b_stride;
1947 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1948 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1949 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1950 if (USES_LIST(top_type, list)) {
1951 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1952 const int b8_xy = 4 * top_xy + 2;
1953 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1954 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1955 ref_cache[0 - 1 * 8] =
1956 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1957 ref_cache[2 - 1 * 8] =
1958 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1960 AV_ZERO128(mv_dst - 1 * 8);
1961 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1964 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1965 if (USES_LIST(left_type[LTOP], list)) {
1966 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1967 const int b8_xy = 4 * left_xy[LTOP] + 1;
1968 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1969 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1970 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1971 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1972 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1974 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1975 ref_cache[-1 + 16] =
1976 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1978 AV_ZERO32(mv_dst - 1 + 0);
1979 AV_ZERO32(mv_dst - 1 + 8);
1980 AV_ZERO32(mv_dst - 1 + 16);
1981 AV_ZERO32(mv_dst - 1 + 24);
1984 ref_cache[-1 + 16] =
1985 ref_cache[-1 + 24] = LIST_NOT_USED;
1990 if (!USES_LIST(mb_type, list)) {
1991 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1992 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1993 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1994 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1995 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2000 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2001 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2002 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2003 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2004 AV_WN32A(&ref_cache[0 * 8], ref01);
2005 AV_WN32A(&ref_cache[1 * 8], ref01);
2006 AV_WN32A(&ref_cache[2 * 8], ref23);
2007 AV_WN32A(&ref_cache[3 * 8], ref23);
2011 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2012 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2013 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2014 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2015 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2021 * @return non zero if the loop filter can be skipped
2023 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2025 const int mb_xy = sl->mb_xy;
2026 int top_xy, left_xy[LEFT_MBS];
2027 int top_type, left_type[LEFT_MBS];
2031 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2033 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2034 * stuff, I can't imagine that these complex rules are worth it. */
2036 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2037 if (FRAME_MBAFF(h)) {
2038 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2039 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2041 if (left_mb_field_flag != curr_mb_field_flag)
2042 left_xy[LTOP] -= h->mb_stride;
2044 if (curr_mb_field_flag)
2045 top_xy += h->mb_stride &
2046 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2047 if (left_mb_field_flag != curr_mb_field_flag)
2048 left_xy[LBOT] += h->mb_stride;
2052 sl->top_mb_xy = top_xy;
2053 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2054 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2056 /* For sufficiently low qp, filtering wouldn't do anything.
2057 * This is a conservative estimate: could also check beta_offset
2058 * and more accurate chroma_qp. */
2059 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2060 int qp = h->cur_pic.qscale_table[mb_xy];
2061 if (qp <= qp_thresh &&
2062 (left_xy[LTOP] < 0 ||
2063 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2065 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2066 if (!FRAME_MBAFF(h))
2068 if ((left_xy[LTOP] < 0 ||
2069 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2070 (top_xy < h->mb_stride ||
2071 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2076 top_type = h->cur_pic.mb_type[top_xy];
2077 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2078 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2079 if (sl->deblocking_filter == 2) {
2080 if (h->slice_table[top_xy] != sl->slice_num)
2082 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2083 left_type[LTOP] = left_type[LBOT] = 0;
2085 if (h->slice_table[top_xy] == 0xFFFF)
2087 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2088 left_type[LTOP] = left_type[LBOT] = 0;
2090 sl->top_type = top_type;
2091 sl->left_type[LTOP] = left_type[LTOP];
2092 sl->left_type[LBOT] = left_type[LBOT];
2094 if (IS_INTRA(mb_type))
2097 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2098 top_type, left_type, mb_xy, 0);
2099 if (sl->list_count == 2)
2100 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2101 top_type, left_type, mb_xy, 1);
2103 nnz = h->non_zero_count[mb_xy];
2104 nnz_cache = sl->non_zero_count_cache;
2105 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2106 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2107 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2108 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2109 sl->cbp = h->cbp_table[mb_xy];
2112 nnz = h->non_zero_count[top_xy];
2113 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2116 if (left_type[LTOP]) {
2117 nnz = h->non_zero_count[left_xy[LTOP]];
2118 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2119 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2120 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2121 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2124 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2125 * from what the loop filter needs */
2126 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2127 if (IS_8x8DCT(top_type)) {
2128 nnz_cache[4 + 8 * 0] =
2129 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2130 nnz_cache[6 + 8 * 0] =
2131 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2133 if (IS_8x8DCT(left_type[LTOP])) {
2134 nnz_cache[3 + 8 * 1] =
2135 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2137 if (IS_8x8DCT(left_type[LBOT])) {
2138 nnz_cache[3 + 8 * 3] =
2139 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2142 if (IS_8x8DCT(mb_type)) {
2143 nnz_cache[scan8[0]] =
2144 nnz_cache[scan8[1]] =
2145 nnz_cache[scan8[2]] =
2146 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2148 nnz_cache[scan8[0 + 4]] =
2149 nnz_cache[scan8[1 + 4]] =
2150 nnz_cache[scan8[2 + 4]] =
2151 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2153 nnz_cache[scan8[0 + 8]] =
2154 nnz_cache[scan8[1 + 8]] =
2155 nnz_cache[scan8[2 + 8]] =
2156 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2158 nnz_cache[scan8[0 + 12]] =
2159 nnz_cache[scan8[1 + 12]] =
2160 nnz_cache[scan8[2 + 12]] =
2161 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2168 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2170 uint8_t *dest_y, *dest_cb, *dest_cr;
2171 int linesize, uvlinesize, mb_x, mb_y;
2172 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2173 const int old_slice_type = sl->slice_type;
2174 const int pixel_shift = h->pixel_shift;
2175 const int block_h = 16 >> h->chroma_y_shift;
2177 if (sl->deblocking_filter) {
2178 for (mb_x = start_x; mb_x < end_x; mb_x++)
2179 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2181 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2182 sl->slice_num = h->slice_table[mb_xy];
2183 mb_type = h->cur_pic.mb_type[mb_xy];
2184 sl->list_count = h->list_counts[mb_xy];
2188 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2192 dest_y = h->cur_pic.f.data[0] +
2193 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2194 dest_cb = h->cur_pic.f.data[1] +
2195 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2196 mb_y * sl->uvlinesize * block_h;
2197 dest_cr = h->cur_pic.f.data[2] +
2198 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2199 mb_y * sl->uvlinesize * block_h;
2200 // FIXME simplify above
2203 linesize = sl->mb_linesize = sl->linesize * 2;
2204 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2205 if (mb_y & 1) { // FIXME move out of this function?
2206 dest_y -= sl->linesize * 15;
2207 dest_cb -= sl->uvlinesize * (block_h - 1);
2208 dest_cr -= sl->uvlinesize * (block_h - 1);
2211 linesize = sl->mb_linesize = sl->linesize;
2212 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2214 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2216 if (fill_filter_caches(h, sl, mb_type))
2218 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2219 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2221 if (FRAME_MBAFF(h)) {
2222 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2223 linesize, uvlinesize);
2225 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2226 dest_cr, linesize, uvlinesize);
2230 sl->slice_type = old_slice_type;
2232 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2233 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2234 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2237 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2239 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2240 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2241 h->cur_pic.mb_type[mb_xy - 1] :
2242 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2243 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2244 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2248 * Draw edges and report progress for the last MB row.
2250 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2252 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2253 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2254 int height = 16 << FRAME_MBAFF(h);
2255 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2257 if (sl->deblocking_filter) {
2258 if ((top + height) >= pic_height)
2259 height += deblock_border;
2260 top -= deblock_border;
2263 if (top >= pic_height || (top + height) < 0)
2266 height = FFMIN(height, pic_height - top);
2268 height = top + height;
2272 ff_h264_draw_horiz_band(h, sl, top, height);
2274 if (h->droppable || sl->er.error_occurred)
2277 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2278 h->picture_structure == PICT_BOTTOM_FIELD);
2281 static void er_add_slice(H264SliceContext *sl,
2282 int startx, int starty,
2283 int endx, int endy, int status)
2285 if (CONFIG_ERROR_RESILIENCE) {
2286 ERContext *er = &sl->er;
2288 ff_er_add_slice(er, startx, starty, endx, endy, status);
2292 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2294 H264SliceContext *sl = arg;
2295 const H264Context *h = sl->h264;
2296 int lf_x_start = sl->mb_x;
2299 sl->linesize = h->cur_pic_ptr->f.linesize[0];
2300 sl->uvlinesize = h->cur_pic_ptr->f.linesize[1];
2302 ret = alloc_scratch_buffers(sl, sl->linesize);
2306 sl->mb_skip_run = -1;
2308 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2310 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2311 avctx->codec_id != AV_CODEC_ID_H264 ||
2312 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2314 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && sl->er.error_status_table) {
2315 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2317 int prev_status = sl->er.error_status_table[sl->er.mb_index2xy[start_i - 1]];
2318 prev_status &= ~ VP_START;
2319 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2320 sl->er.error_occurred = 1;
2326 align_get_bits(&sl->gb);
2329 ff_init_cabac_decoder(&sl->cabac,
2330 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2331 (get_bits_left(&sl->gb) + 7) / 8);
2333 ff_h264_init_cabac_states(h, sl);
2337 int ret = ff_h264_decode_mb_cabac(h, sl);
2339 // STOP_TIMER("decode_mb_cabac")
2342 ff_h264_hl_decode_mb(h, sl);
2344 // FIXME optimal? or let mb_decode decode 16x32 ?
2345 if (ret >= 0 && FRAME_MBAFF(h)) {
2348 ret = ff_h264_decode_mb_cabac(h, sl);
2351 ff_h264_hl_decode_mb(h, sl);
2354 eos = get_cabac_terminate(&sl->cabac);
2356 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2357 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2358 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2359 sl->mb_y, ER_MB_END);
2360 if (sl->mb_x >= lf_x_start)
2361 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2364 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2365 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2366 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2367 av_log(h->avctx, AV_LOG_ERROR,
2368 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2370 sl->cabac.bytestream_end - sl->cabac.bytestream);
2371 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2372 sl->mb_y, ER_MB_ERROR);
2373 return AVERROR_INVALIDDATA;
2376 if (++sl->mb_x >= h->mb_width) {
2377 loop_filter(h, sl, lf_x_start, sl->mb_x);
2378 sl->mb_x = lf_x_start = 0;
2379 decode_finish_row(h, sl);
2381 if (FIELD_OR_MBAFF_PICTURE(h)) {
2383 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2384 predict_field_decoding_flag(h, sl);
2388 if (eos || sl->mb_y >= h->mb_height) {
2389 tprintf(h->avctx, "slice end %d %d\n",
2390 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2391 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2392 sl->mb_y, ER_MB_END);
2393 if (sl->mb_x > lf_x_start)
2394 loop_filter(h, sl, lf_x_start, sl->mb_x);
2400 int ret = ff_h264_decode_mb_cavlc(h, sl);
2403 ff_h264_hl_decode_mb(h, sl);
2405 // FIXME optimal? or let mb_decode decode 16x32 ?
2406 if (ret >= 0 && FRAME_MBAFF(h)) {
2408 ret = ff_h264_decode_mb_cavlc(h, sl);
2411 ff_h264_hl_decode_mb(h, sl);
2416 av_log(h->avctx, AV_LOG_ERROR,
2417 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2418 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2419 sl->mb_y, ER_MB_ERROR);
2423 if (++sl->mb_x >= h->mb_width) {
2424 loop_filter(h, sl, lf_x_start, sl->mb_x);
2425 sl->mb_x = lf_x_start = 0;
2426 decode_finish_row(h, sl);
2428 if (FIELD_OR_MBAFF_PICTURE(h)) {
2430 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2431 predict_field_decoding_flag(h, sl);
2433 if (sl->mb_y >= h->mb_height) {
2434 tprintf(h->avctx, "slice end %d %d\n",
2435 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2437 if ( get_bits_left(&sl->gb) == 0
2438 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2439 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2440 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2444 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2445 sl->mb_x, sl->mb_y, ER_MB_END);
2447 return AVERROR_INVALIDDATA;
2452 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2453 tprintf(h->avctx, "slice end %d %d\n",
2454 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2456 if (get_bits_left(&sl->gb) == 0) {
2457 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2458 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2459 if (sl->mb_x > lf_x_start)
2460 loop_filter(h, sl, lf_x_start, sl->mb_x);
2464 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2465 sl->mb_y, ER_MB_ERROR);
2467 return AVERROR_INVALIDDATA;
2475 * Call decode_slice() for each context.
2477 * @param h h264 master context
2478 * @param context_count number of contexts to execute
2480 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2482 AVCodecContext *const avctx = h->avctx;
2483 H264SliceContext *sl;
2486 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2488 if (h->avctx->hwaccel ||
2489 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2491 if (context_count == 1) {
2492 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2493 h->mb_y = h->slice_ctx[0].mb_y;
2496 av_assert0(context_count > 0);
2497 for (i = 1; i < context_count; i++) {
2498 sl = &h->slice_ctx[i];
2499 if (CONFIG_ERROR_RESILIENCE) {
2500 sl->er.error_count = 0;
2504 avctx->execute(avctx, decode_slice, h->slice_ctx,
2505 NULL, context_count, sizeof(h->slice_ctx[0]));
2507 /* pull back stuff from slices to master context */
2508 sl = &h->slice_ctx[context_count - 1];
2510 if (CONFIG_ERROR_RESILIENCE) {
2511 for (i = 1; i < context_count; i++)
2512 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;