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 av_assert1((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_RGB)
1008 *fmt++ = AV_PIX_FMT_GBRP;
1009 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1010 *fmt++ = AV_PIX_FMT_YUVJ444P;
1012 *fmt++ = AV_PIX_FMT_YUV444P;
1013 } else if (CHROMA422(h)) {
1014 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1015 *fmt++ = AV_PIX_FMT_YUVJ422P;
1017 *fmt++ = AV_PIX_FMT_YUV422P;
1019 #if CONFIG_H264_DXVA2_HWACCEL
1020 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1022 #if CONFIG_H264_VAAPI_HWACCEL
1023 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1025 #if CONFIG_H264_VDA_HWACCEL
1026 *fmt++ = AV_PIX_FMT_VDA_VLD;
1027 *fmt++ = AV_PIX_FMT_VDA;
1029 if (h->avctx->codec->pix_fmts)
1030 choices = h->avctx->codec->pix_fmts;
1031 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1032 *fmt++ = AV_PIX_FMT_YUVJ420P;
1034 *fmt++ = AV_PIX_FMT_YUV420P;
1038 av_log(h->avctx, AV_LOG_ERROR,
1039 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1040 return AVERROR_INVALIDDATA;
1043 *fmt = AV_PIX_FMT_NONE;
1045 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1046 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1048 return ff_thread_get_format(h->avctx, choices);
1051 /* export coded and cropped frame dimensions to AVCodecContext */
1052 static int init_dimensions(H264Context *h)
1054 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1055 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1056 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1057 h->sps.crop_right || h->sps.crop_bottom;
1058 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1059 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1061 /* handle container cropping */
1062 if (!crop_present &&
1063 FFALIGN(h->avctx->width, 16) == h->width &&
1064 FFALIGN(h->avctx->height, 16) == h->height) {
1065 width = h->avctx->width;
1066 height = h->avctx->height;
1069 if (width <= 0 || height <= 0) {
1070 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1072 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1073 return AVERROR_INVALIDDATA;
1075 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1076 h->sps.crop_bottom =
1086 h->avctx->coded_width = h->width;
1087 h->avctx->coded_height = h->height;
1088 h->avctx->width = width;
1089 h->avctx->height = height;
1094 static int h264_slice_header_init(H264Context *h, int reinit)
1096 int nb_slices = (HAVE_THREADS &&
1097 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1098 h->avctx->thread_count : 1;
1101 ff_set_sar(h->avctx, h->sps.sar);
1102 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1103 &h->chroma_x_shift, &h->chroma_y_shift);
1105 if (h->sps.timing_info_present_flag) {
1106 int64_t den = h->sps.time_scale;
1107 if (h->x264_build < 44U)
1109 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1110 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1114 ff_h264_free_tables(h, 0);
1116 h->prev_interlaced_frame = 1;
1118 init_scan_tables(h);
1119 ret = ff_h264_alloc_tables(h);
1121 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1125 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1128 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1130 max_slices = H264_MAX_THREADS;
1131 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1132 " reducing to %d\n", nb_slices, max_slices);
1133 nb_slices = max_slices;
1135 h->slice_context_count = nb_slices;
1137 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1138 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1140 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1144 for (i = 0; i < h->slice_context_count; i++) {
1145 H264SliceContext *sl = &h->slice_ctx[i];
1148 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1149 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1150 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1152 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1153 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1159 h->context_initialized = 1;
1163 ff_h264_free_tables(h, 0);
1164 h->context_initialized = 0;
1168 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1171 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1172 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1173 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1180 * Decode a slice header.
1181 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1183 * @param h h264context
1185 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1187 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1189 unsigned int first_mb_in_slice;
1190 unsigned int pps_id;
1192 unsigned int slice_type, tmp, i, j;
1193 int last_pic_structure, last_pic_droppable;
1195 int needs_reinit = 0;
1196 int field_pic_flag, bottom_field_flag;
1197 int first_slice = sl == h->slice_ctx && !h->current_slice;
1198 int frame_num, picture_structure, droppable;
1199 int mb_aff_frame, last_mb_aff_frame;
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) {
1209 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1210 ff_h264_field_end(h, h->slice_ctx, 1);
1211 h->current_slice = 0;
1212 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == NAL_IDR_SLICE) {
1213 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1214 ff_h264_field_end(h, h->slice_ctx, 1);
1215 h->current_slice = 0;
1216 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1217 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1218 h->cur_pic_ptr = NULL;
1220 return AVERROR_INVALIDDATA;
1223 if (!h->first_field) {
1224 if (h->cur_pic_ptr && !h->droppable) {
1225 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1226 h->picture_structure == PICT_BOTTOM_FIELD);
1228 h->cur_pic_ptr = NULL;
1232 slice_type = get_ue_golomb_31(&sl->gb);
1233 if (slice_type > 9) {
1234 av_log(h->avctx, AV_LOG_ERROR,
1235 "slice type %d too large at %d\n",
1236 slice_type, first_mb_in_slice);
1237 return AVERROR_INVALIDDATA;
1239 if (slice_type > 4) {
1241 sl->slice_type_fixed = 1;
1243 sl->slice_type_fixed = 0;
1245 slice_type = golomb_to_pict_type[slice_type];
1247 sl->slice_type = slice_type;
1248 sl->slice_type_nos = slice_type & 3;
1250 if (h->nal_unit_type == NAL_IDR_SLICE &&
1251 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1252 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1253 return AVERROR_INVALIDDATA;
1257 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1258 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1259 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1260 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1261 h->avctx->skip_frame >= AVDISCARD_ALL) {
1262 return SLICE_SKIPED;
1265 // to make a few old functions happy, it's wrong though
1266 h->pict_type = sl->slice_type;
1268 pps_id = get_ue_golomb(&sl->gb);
1269 if (pps_id >= MAX_PPS_COUNT) {
1270 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1271 return AVERROR_INVALIDDATA;
1273 if (!h->pps_buffers[pps_id]) {
1274 av_log(h->avctx, AV_LOG_ERROR,
1275 "non-existing PPS %u referenced\n",
1277 return AVERROR_INVALIDDATA;
1279 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1280 av_log(h->avctx, AV_LOG_ERROR,
1281 "PPS change from %d to %d forbidden\n",
1282 h->au_pps_id, pps_id);
1283 return AVERROR_INVALIDDATA;
1286 pps = h->pps_buffers[pps_id];
1288 if (!h->sps_buffers[pps->sps_id]) {
1289 av_log(h->avctx, AV_LOG_ERROR,
1290 "non-existing SPS %u referenced\n",
1292 return AVERROR_INVALIDDATA;
1295 h->pps = *h->pps_buffers[pps_id];
1297 if (pps->sps_id != h->sps.sps_id ||
1298 pps->sps_id != h->current_sps_id ||
1299 h->sps_buffers[pps->sps_id]->new) {
1302 av_log(h->avctx, AV_LOG_ERROR,
1303 "SPS changed in the middle of the frame\n");
1304 return AVERROR_INVALIDDATA;
1307 h->sps = *h->sps_buffers[h->pps.sps_id];
1309 if (h->mb_width != h->sps.mb_width ||
1310 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1311 h->cur_bit_depth_luma != h->sps.bit_depth_luma ||
1312 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1316 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1317 h->chroma_format_idc != h->sps.chroma_format_idc) {
1318 h->bit_depth_luma = h->sps.bit_depth_luma;
1319 h->chroma_format_idc = h->sps.chroma_format_idc;
1322 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1326 h->avctx->profile = ff_h264_get_profile(&h->sps);
1327 h->avctx->level = h->sps.level_idc;
1328 h->avctx->refs = h->sps.ref_frame_count;
1330 must_reinit = (h->context_initialized &&
1331 ( 16*h->sps.mb_width != h->avctx->coded_width
1332 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1333 || h->cur_bit_depth_luma != h->sps.bit_depth_luma
1334 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1335 || h->mb_width != h->sps.mb_width
1336 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1338 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1339 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1342 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1345 h->mb_width = h->sps.mb_width;
1346 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1347 h->mb_num = h->mb_width * h->mb_height;
1348 h->mb_stride = h->mb_width + 1;
1350 h->b_stride = h->mb_width * 4;
1352 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1354 h->width = 16 * h->mb_width;
1355 h->height = 16 * h->mb_height;
1357 ret = init_dimensions(h);
1361 if (h->sps.video_signal_type_present_flag) {
1362 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1364 if (h->sps.colour_description_present_flag) {
1365 if (h->avctx->colorspace != h->sps.colorspace)
1367 h->avctx->color_primaries = h->sps.color_primaries;
1368 h->avctx->color_trc = h->sps.color_trc;
1369 h->avctx->colorspace = h->sps.colorspace;
1373 if (h->context_initialized &&
1374 (must_reinit || needs_reinit)) {
1375 if (sl != h->slice_ctx) {
1376 av_log(h->avctx, AV_LOG_ERROR,
1377 "changing width %d -> %d / height %d -> %d on "
1379 h->width, h->avctx->coded_width,
1380 h->height, h->avctx->coded_height,
1381 h->current_slice + 1);
1382 return AVERROR_INVALIDDATA;
1385 av_assert1(first_slice);
1387 ff_h264_flush_change(h);
1389 if ((ret = get_pixel_format(h, 1)) < 0)
1391 h->avctx->pix_fmt = ret;
1393 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1394 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1396 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1397 av_log(h->avctx, AV_LOG_ERROR,
1398 "h264_slice_header_init() failed\n");
1402 if (!h->context_initialized) {
1403 if (sl != h->slice_ctx) {
1404 av_log(h->avctx, AV_LOG_ERROR,
1405 "Cannot (re-)initialize context during parallel decoding.\n");
1406 return AVERROR_PATCHWELCOME;
1409 if ((ret = get_pixel_format(h, 1)) < 0)
1411 h->avctx->pix_fmt = ret;
1413 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1414 av_log(h->avctx, AV_LOG_ERROR,
1415 "h264_slice_header_init() failed\n");
1420 if (first_slice && h->dequant_coeff_pps != pps_id) {
1421 h->dequant_coeff_pps = pps_id;
1422 ff_h264_init_dequant_tables(h);
1425 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1427 if (h->frame_num != frame_num) {
1428 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1429 h->frame_num, frame_num);
1430 return AVERROR_INVALIDDATA;
1436 last_mb_aff_frame = h->mb_aff_frame;
1437 last_pic_structure = h->picture_structure;
1438 last_pic_droppable = h->droppable;
1439 droppable = h->nal_ref_idc == 0;
1440 if (h->sps.frame_mbs_only_flag) {
1441 picture_structure = PICT_FRAME;
1443 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1444 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1447 field_pic_flag = get_bits1(&sl->gb);
1449 if (field_pic_flag) {
1450 bottom_field_flag = get_bits1(&sl->gb);
1451 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1453 picture_structure = PICT_FRAME;
1454 mb_aff_frame = h->sps.mb_aff;
1457 if (h->current_slice) {
1458 if (last_pic_structure != picture_structure ||
1459 last_pic_droppable != droppable ||
1460 last_mb_aff_frame != mb_aff_frame) {
1461 av_log(h->avctx, AV_LOG_ERROR,
1462 "Changing field mode (%d -> %d) between slices is not allowed\n",
1463 last_pic_structure, h->picture_structure);
1464 return AVERROR_INVALIDDATA;
1465 } else if (!h->cur_pic_ptr) {
1466 av_log(h->avctx, AV_LOG_ERROR,
1467 "unset cur_pic_ptr on slice %d\n",
1468 h->current_slice + 1);
1469 return AVERROR_INVALIDDATA;
1473 h->picture_structure = picture_structure;
1474 h->droppable = droppable;
1475 h->frame_num = frame_num;
1476 h->mb_aff_frame = mb_aff_frame;
1477 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1479 if (h->current_slice == 0) {
1480 /* Shorten frame num gaps so we don't have to allocate reference
1481 * frames just to throw them away */
1482 if (h->frame_num != h->prev_frame_num) {
1483 int unwrap_prev_frame_num = h->prev_frame_num;
1484 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1486 if (unwrap_prev_frame_num > h->frame_num)
1487 unwrap_prev_frame_num -= max_frame_num;
1489 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1490 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1491 if (unwrap_prev_frame_num < 0)
1492 unwrap_prev_frame_num += max_frame_num;
1494 h->prev_frame_num = unwrap_prev_frame_num;
1498 /* See if we have a decoded first field looking for a pair...
1499 * Here, we're using that to see if we should mark previously
1500 * decode frames as "finished".
1501 * We have to do that before the "dummy" in-between frame allocation,
1502 * since that can modify h->cur_pic_ptr. */
1503 if (h->first_field) {
1504 av_assert0(h->cur_pic_ptr);
1505 av_assert0(h->cur_pic_ptr->f.buf[0]);
1506 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1508 /* Mark old field/frame as completed */
1509 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1510 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1511 last_pic_structure == PICT_BOTTOM_FIELD);
1514 /* figure out if we have a complementary field pair */
1515 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1516 /* Previous field is unmatched. Don't display it, but let it
1517 * remain for reference if marked as such. */
1518 if (last_pic_structure != PICT_FRAME) {
1519 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1520 last_pic_structure == PICT_TOP_FIELD);
1523 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1524 /* This and previous field were reference, but had
1525 * different frame_nums. Consider this field first in
1526 * pair. Throw away previous field except for reference
1528 if (last_pic_structure != PICT_FRAME) {
1529 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1530 last_pic_structure == PICT_TOP_FIELD);
1533 /* Second field in complementary pair */
1534 if (!((last_pic_structure == PICT_TOP_FIELD &&
1535 h->picture_structure == PICT_BOTTOM_FIELD) ||
1536 (last_pic_structure == PICT_BOTTOM_FIELD &&
1537 h->picture_structure == PICT_TOP_FIELD))) {
1538 av_log(h->avctx, AV_LOG_ERROR,
1539 "Invalid field mode combination %d/%d\n",
1540 last_pic_structure, h->picture_structure);
1541 h->picture_structure = last_pic_structure;
1542 h->droppable = last_pic_droppable;
1543 return AVERROR_INVALIDDATA;
1544 } else if (last_pic_droppable != h->droppable) {
1545 avpriv_request_sample(h->avctx,
1546 "Found reference and non-reference fields in the same frame, which");
1547 h->picture_structure = last_pic_structure;
1548 h->droppable = last_pic_droppable;
1549 return AVERROR_PATCHWELCOME;
1555 while (h->frame_num != h->prev_frame_num && !h->first_field &&
1556 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1557 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1558 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1559 h->frame_num, h->prev_frame_num);
1560 if (!h->sps.gaps_in_frame_num_allowed_flag)
1561 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1562 h->last_pocs[i] = INT_MIN;
1563 ret = h264_frame_start(h);
1569 h->prev_frame_num++;
1570 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1571 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1572 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1573 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1574 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1575 ret = ff_generate_sliding_window_mmcos(h, 1);
1576 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1578 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1579 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1581 /* Error concealment: If a ref is missing, copy the previous ref
1583 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1584 * many assumptions about there being no actual duplicates.
1585 * FIXME: This does not copy padding for out-of-frame motion
1586 * vectors. Given we are concealing a lost frame, this probably
1587 * is not noticeable by comparison, but it should be fixed. */
1588 if (h->short_ref_count) {
1590 av_image_copy(h->short_ref[0]->f.data,
1591 h->short_ref[0]->f.linesize,
1592 (const uint8_t **)prev->f.data,
1597 h->short_ref[0]->poc = prev->poc + 2;
1599 h->short_ref[0]->frame_num = h->prev_frame_num;
1603 /* See if we have a decoded first field looking for a pair...
1604 * We're using that to see whether to continue decoding in that
1605 * frame, or to allocate a new one. */
1606 if (h->first_field) {
1607 av_assert0(h->cur_pic_ptr);
1608 av_assert0(h->cur_pic_ptr->f.buf[0]);
1609 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1611 /* figure out if we have a complementary field pair */
1612 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1613 /* Previous field is unmatched. Don't display it, but let it
1614 * remain for reference if marked as such. */
1615 h->missing_fields ++;
1616 h->cur_pic_ptr = NULL;
1617 h->first_field = FIELD_PICTURE(h);
1619 h->missing_fields = 0;
1620 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1621 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1622 h->picture_structure==PICT_BOTTOM_FIELD);
1623 /* This and the previous field had different frame_nums.
1624 * Consider this field first in pair. Throw away previous
1625 * one except for reference purposes. */
1627 h->cur_pic_ptr = NULL;
1629 /* Second field in complementary pair */
1634 /* Frame or first field in a potentially complementary pair */
1635 h->first_field = FIELD_PICTURE(h);
1638 if (!FIELD_PICTURE(h) || h->first_field) {
1639 if (h264_frame_start(h) < 0) {
1641 return AVERROR_INVALIDDATA;
1644 release_unused_pictures(h, 0);
1646 /* Some macroblocks can be accessed before they're available in case
1647 * of lost slices, MBAFF or threading. */
1648 if (FIELD_PICTURE(h)) {
1649 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1650 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1652 memset(h->slice_table, -1,
1653 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1655 h->last_slice_type = -1;
1659 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1661 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1662 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1663 first_mb_in_slice >= h->mb_num) {
1664 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1665 return AVERROR_INVALIDDATA;
1667 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1668 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1669 FIELD_OR_MBAFF_PICTURE(h);
1670 if (h->picture_structure == PICT_BOTTOM_FIELD)
1671 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1672 av_assert1(sl->mb_y < h->mb_height);
1674 if (h->picture_structure == PICT_FRAME) {
1675 h->curr_pic_num = h->frame_num;
1676 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1678 h->curr_pic_num = 2 * h->frame_num + 1;
1679 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1682 if (h->nal_unit_type == NAL_IDR_SLICE)
1683 get_ue_golomb(&sl->gb); /* idr_pic_id */
1685 if (h->sps.poc_type == 0) {
1686 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1688 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1689 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1692 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1693 h->delta_poc[0] = get_se_golomb(&sl->gb);
1695 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1696 h->delta_poc[1] = get_se_golomb(&sl->gb);
1699 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1701 if (h->pps.redundant_pic_cnt_present)
1702 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1704 ret = ff_set_ref_count(h, sl);
1708 if (slice_type != AV_PICTURE_TYPE_I &&
1709 (h->current_slice == 0 ||
1710 slice_type != h->last_slice_type ||
1711 memcmp(h->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1713 ff_h264_fill_default_ref_list(h, sl);
1716 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1717 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1719 sl->ref_count[1] = sl->ref_count[0] = 0;
1724 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1725 (h->pps.weighted_bipred_idc == 1 &&
1726 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1727 ff_pred_weight_table(h, sl);
1728 else if (h->pps.weighted_bipred_idc == 2 &&
1729 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1730 implicit_weight_table(h, sl, -1);
1733 for (i = 0; i < 2; i++) {
1734 sl->luma_weight_flag[i] = 0;
1735 sl->chroma_weight_flag[i] = 0;
1739 // If frame-mt is enabled, only update mmco tables for the first slice
1740 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1741 // or h->mmco, which will cause ref list mix-ups and decoding errors
1742 // further down the line. This may break decoding if the first slice is
1743 // corrupt, thus we only do this if frame-mt is enabled.
1744 if (h->nal_ref_idc) {
1745 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1746 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1747 h->current_slice == 0);
1748 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1749 return AVERROR_INVALIDDATA;
1752 if (FRAME_MBAFF(h)) {
1753 ff_h264_fill_mbaff_ref_list(h, sl);
1755 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1756 implicit_weight_table(h, sl, 0);
1757 implicit_weight_table(h, sl, 1);
1761 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1762 ff_h264_direct_dist_scale_factor(h, sl);
1763 ff_h264_direct_ref_list_init(h, sl);
1765 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1766 tmp = get_ue_golomb_31(&sl->gb);
1768 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1769 return AVERROR_INVALIDDATA;
1771 sl->cabac_init_idc = tmp;
1774 sl->last_qscale_diff = 0;
1775 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1776 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1777 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1778 return AVERROR_INVALIDDATA;
1781 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1782 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1783 // FIXME qscale / qp ... stuff
1784 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1785 get_bits1(&sl->gb); /* sp_for_switch_flag */
1786 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1787 sl->slice_type == AV_PICTURE_TYPE_SI)
1788 get_se_golomb(&sl->gb); /* slice_qs_delta */
1790 sl->deblocking_filter = 1;
1791 sl->slice_alpha_c0_offset = 0;
1792 sl->slice_beta_offset = 0;
1793 if (h->pps.deblocking_filter_parameters_present) {
1794 tmp = get_ue_golomb_31(&sl->gb);
1796 av_log(h->avctx, AV_LOG_ERROR,
1797 "deblocking_filter_idc %u out of range\n", tmp);
1798 return AVERROR_INVALIDDATA;
1800 sl->deblocking_filter = tmp;
1801 if (sl->deblocking_filter < 2)
1802 sl->deblocking_filter ^= 1; // 1<->0
1804 if (sl->deblocking_filter) {
1805 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1806 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1807 if (sl->slice_alpha_c0_offset > 12 ||
1808 sl->slice_alpha_c0_offset < -12 ||
1809 sl->slice_beta_offset > 12 ||
1810 sl->slice_beta_offset < -12) {
1811 av_log(h->avctx, AV_LOG_ERROR,
1812 "deblocking filter parameters %d %d out of range\n",
1813 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1814 return AVERROR_INVALIDDATA;
1819 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1820 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1821 h->nal_unit_type != NAL_IDR_SLICE) ||
1822 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1823 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1824 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1825 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1826 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1827 h->nal_ref_idc == 0))
1828 sl->deblocking_filter = 0;
1830 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1831 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1832 /* Cheat slightly for speed:
1833 * Do not bother to deblock across slices. */
1834 sl->deblocking_filter = 2;
1836 h->max_contexts = 1;
1837 if (!h->single_decode_warning) {
1838 av_log(h->avctx, AV_LOG_INFO,
1839 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1840 "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"
1841 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1842 "but will generate non-standard-compliant output.\n");
1843 h->single_decode_warning = 1;
1845 if (sl != h->slice_ctx) {
1846 av_log(h->avctx, AV_LOG_ERROR,
1847 "Deblocking switched inside frame.\n");
1848 return SLICE_SINGLETHREAD;
1852 sl->qp_thresh = 15 -
1853 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1855 h->pps.chroma_qp_index_offset[0],
1856 h->pps.chroma_qp_index_offset[1]) +
1857 6 * (h->sps.bit_depth_luma - 8);
1859 h->last_slice_type = slice_type;
1860 memcpy(h->last_ref_count, sl->ref_count, sizeof(h->last_ref_count));
1861 sl->slice_num = ++h->current_slice;
1864 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1865 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1866 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1867 && sl->slice_num >= MAX_SLICES) {
1868 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1869 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);
1872 for (j = 0; j < 2; j++) {
1874 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1875 for (i = 0; i < 16; i++) {
1877 if (j < sl->list_count && i < sl->ref_count[j] &&
1878 sl->ref_list[j][i].parent->f.buf[0]) {
1880 AVBuffer *buf = sl->ref_list[j][i].parent->f.buf[0]->buffer;
1881 for (k = 0; k < h->short_ref_count; k++)
1882 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1886 for (k = 0; k < h->long_ref_count; k++)
1887 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1888 id_list[i] = h->short_ref_count + k;
1896 for (i = 0; i < 16; i++)
1897 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1899 ref2frm[18 + 1] = -1;
1900 for (i = 16; i < 48; i++)
1901 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1902 (sl->ref_list[j][i].reference & 3);
1905 h->au_pps_id = pps_id;
1907 h->sps_buffers[h->pps.sps_id]->new = 0;
1908 h->current_sps_id = h->pps.sps_id;
1910 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1911 av_log(h->avctx, AV_LOG_DEBUG,
1912 "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",
1914 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1916 av_get_picture_type_char(sl->slice_type),
1917 sl->slice_type_fixed ? " fix" : "",
1918 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1919 pps_id, h->frame_num,
1920 h->cur_pic_ptr->field_poc[0],
1921 h->cur_pic_ptr->field_poc[1],
1922 sl->ref_count[0], sl->ref_count[1],
1924 sl->deblocking_filter,
1925 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1927 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1928 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1934 int ff_h264_get_slice_type(const H264SliceContext *sl)
1936 switch (sl->slice_type) {
1937 case AV_PICTURE_TYPE_P:
1939 case AV_PICTURE_TYPE_B:
1941 case AV_PICTURE_TYPE_I:
1943 case AV_PICTURE_TYPE_SP:
1945 case AV_PICTURE_TYPE_SI:
1948 return AVERROR_INVALIDDATA;
1952 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1953 H264SliceContext *sl,
1954 int mb_type, int top_xy,
1955 int left_xy[LEFT_MBS],
1957 int left_type[LEFT_MBS],
1958 int mb_xy, int list)
1960 int b_stride = h->b_stride;
1961 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1962 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1963 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1964 if (USES_LIST(top_type, list)) {
1965 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1966 const int b8_xy = 4 * top_xy + 2;
1967 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1968 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1969 ref_cache[0 - 1 * 8] =
1970 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1971 ref_cache[2 - 1 * 8] =
1972 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1974 AV_ZERO128(mv_dst - 1 * 8);
1975 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1978 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1979 if (USES_LIST(left_type[LTOP], list)) {
1980 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1981 const int b8_xy = 4 * left_xy[LTOP] + 1;
1982 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1983 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1984 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1985 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1986 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1988 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1989 ref_cache[-1 + 16] =
1990 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1992 AV_ZERO32(mv_dst - 1 + 0);
1993 AV_ZERO32(mv_dst - 1 + 8);
1994 AV_ZERO32(mv_dst - 1 + 16);
1995 AV_ZERO32(mv_dst - 1 + 24);
1998 ref_cache[-1 + 16] =
1999 ref_cache[-1 + 24] = LIST_NOT_USED;
2004 if (!USES_LIST(mb_type, list)) {
2005 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2006 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2007 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2008 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2009 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2014 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2015 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2016 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2017 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2018 AV_WN32A(&ref_cache[0 * 8], ref01);
2019 AV_WN32A(&ref_cache[1 * 8], ref01);
2020 AV_WN32A(&ref_cache[2 * 8], ref23);
2021 AV_WN32A(&ref_cache[3 * 8], ref23);
2025 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2026 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2027 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2028 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2029 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2035 * @return non zero if the loop filter can be skipped
2037 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2039 const int mb_xy = sl->mb_xy;
2040 int top_xy, left_xy[LEFT_MBS];
2041 int top_type, left_type[LEFT_MBS];
2045 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2047 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2048 * stuff, I can't imagine that these complex rules are worth it. */
2050 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2051 if (FRAME_MBAFF(h)) {
2052 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2053 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2055 if (left_mb_field_flag != curr_mb_field_flag)
2056 left_xy[LTOP] -= h->mb_stride;
2058 if (curr_mb_field_flag)
2059 top_xy += h->mb_stride &
2060 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2061 if (left_mb_field_flag != curr_mb_field_flag)
2062 left_xy[LBOT] += h->mb_stride;
2066 sl->top_mb_xy = top_xy;
2067 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2068 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2070 /* For sufficiently low qp, filtering wouldn't do anything.
2071 * This is a conservative estimate: could also check beta_offset
2072 * and more accurate chroma_qp. */
2073 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2074 int qp = h->cur_pic.qscale_table[mb_xy];
2075 if (qp <= qp_thresh &&
2076 (left_xy[LTOP] < 0 ||
2077 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2079 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2080 if (!FRAME_MBAFF(h))
2082 if ((left_xy[LTOP] < 0 ||
2083 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2084 (top_xy < h->mb_stride ||
2085 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2090 top_type = h->cur_pic.mb_type[top_xy];
2091 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2092 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2093 if (sl->deblocking_filter == 2) {
2094 if (h->slice_table[top_xy] != sl->slice_num)
2096 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2097 left_type[LTOP] = left_type[LBOT] = 0;
2099 if (h->slice_table[top_xy] == 0xFFFF)
2101 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2102 left_type[LTOP] = left_type[LBOT] = 0;
2104 sl->top_type = top_type;
2105 sl->left_type[LTOP] = left_type[LTOP];
2106 sl->left_type[LBOT] = left_type[LBOT];
2108 if (IS_INTRA(mb_type))
2111 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2112 top_type, left_type, mb_xy, 0);
2113 if (sl->list_count == 2)
2114 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2115 top_type, left_type, mb_xy, 1);
2117 nnz = h->non_zero_count[mb_xy];
2118 nnz_cache = sl->non_zero_count_cache;
2119 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2120 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2121 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2122 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2123 sl->cbp = h->cbp_table[mb_xy];
2126 nnz = h->non_zero_count[top_xy];
2127 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2130 if (left_type[LTOP]) {
2131 nnz = h->non_zero_count[left_xy[LTOP]];
2132 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2133 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2134 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2135 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2138 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2139 * from what the loop filter needs */
2140 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2141 if (IS_8x8DCT(top_type)) {
2142 nnz_cache[4 + 8 * 0] =
2143 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2144 nnz_cache[6 + 8 * 0] =
2145 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2147 if (IS_8x8DCT(left_type[LTOP])) {
2148 nnz_cache[3 + 8 * 1] =
2149 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2151 if (IS_8x8DCT(left_type[LBOT])) {
2152 nnz_cache[3 + 8 * 3] =
2153 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2156 if (IS_8x8DCT(mb_type)) {
2157 nnz_cache[scan8[0]] =
2158 nnz_cache[scan8[1]] =
2159 nnz_cache[scan8[2]] =
2160 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2162 nnz_cache[scan8[0 + 4]] =
2163 nnz_cache[scan8[1 + 4]] =
2164 nnz_cache[scan8[2 + 4]] =
2165 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2167 nnz_cache[scan8[0 + 8]] =
2168 nnz_cache[scan8[1 + 8]] =
2169 nnz_cache[scan8[2 + 8]] =
2170 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2172 nnz_cache[scan8[0 + 12]] =
2173 nnz_cache[scan8[1 + 12]] =
2174 nnz_cache[scan8[2 + 12]] =
2175 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2182 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2184 uint8_t *dest_y, *dest_cb, *dest_cr;
2185 int linesize, uvlinesize, mb_x, mb_y;
2186 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2187 const int old_slice_type = sl->slice_type;
2188 const int pixel_shift = h->pixel_shift;
2189 const int block_h = 16 >> h->chroma_y_shift;
2191 if (sl->deblocking_filter) {
2192 for (mb_x = start_x; mb_x < end_x; mb_x++)
2193 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2195 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2196 sl->slice_num = h->slice_table[mb_xy];
2197 mb_type = h->cur_pic.mb_type[mb_xy];
2198 sl->list_count = h->list_counts[mb_xy];
2202 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2206 dest_y = h->cur_pic.f.data[0] +
2207 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2208 dest_cb = h->cur_pic.f.data[1] +
2209 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2210 mb_y * sl->uvlinesize * block_h;
2211 dest_cr = h->cur_pic.f.data[2] +
2212 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2213 mb_y * sl->uvlinesize * block_h;
2214 // FIXME simplify above
2217 linesize = sl->mb_linesize = sl->linesize * 2;
2218 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2219 if (mb_y & 1) { // FIXME move out of this function?
2220 dest_y -= sl->linesize * 15;
2221 dest_cb -= sl->uvlinesize * (block_h - 1);
2222 dest_cr -= sl->uvlinesize * (block_h - 1);
2225 linesize = sl->mb_linesize = sl->linesize;
2226 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2228 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2230 if (fill_filter_caches(h, sl, mb_type))
2232 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2233 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2235 if (FRAME_MBAFF(h)) {
2236 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2237 linesize, uvlinesize);
2239 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2240 dest_cr, linesize, uvlinesize);
2244 sl->slice_type = old_slice_type;
2246 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2247 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2248 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2251 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2253 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2254 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2255 h->cur_pic.mb_type[mb_xy - 1] :
2256 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2257 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2258 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2262 * Draw edges and report progress for the last MB row.
2264 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2266 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2267 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2268 int height = 16 << FRAME_MBAFF(h);
2269 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2271 if (sl->deblocking_filter) {
2272 if ((top + height) >= pic_height)
2273 height += deblock_border;
2274 top -= deblock_border;
2277 if (top >= pic_height || (top + height) < 0)
2280 height = FFMIN(height, pic_height - top);
2282 height = top + height;
2286 ff_h264_draw_horiz_band(h, sl, top, height);
2288 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2291 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2292 h->picture_structure == PICT_BOTTOM_FIELD);
2295 static void er_add_slice(H264SliceContext *sl,
2296 int startx, int starty,
2297 int endx, int endy, int status)
2299 if (!sl->h264->enable_er)
2302 if (CONFIG_ERROR_RESILIENCE) {
2303 ERContext *er = &sl->h264->slice_ctx[0].er;
2305 ff_er_add_slice(er, startx, starty, endx, endy, status);
2309 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2311 H264SliceContext *sl = arg;
2312 const H264Context *h = sl->h264;
2313 int lf_x_start = sl->mb_x;
2316 sl->linesize = h->cur_pic_ptr->f.linesize[0];
2317 sl->uvlinesize = h->cur_pic_ptr->f.linesize[1];
2319 ret = alloc_scratch_buffers(sl, sl->linesize);
2323 sl->mb_skip_run = -1;
2325 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2327 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2328 avctx->codec_id != AV_CODEC_ID_H264 ||
2329 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2331 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2332 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2334 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2335 prev_status &= ~ VP_START;
2336 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2337 h->slice_ctx[0].er.error_occurred = 1;
2343 align_get_bits(&sl->gb);
2346 ff_init_cabac_decoder(&sl->cabac,
2347 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2348 (get_bits_left(&sl->gb) + 7) / 8);
2350 ff_h264_init_cabac_states(h, sl);
2356 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->mb_index_end) {
2357 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps next at %d\n",
2359 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2360 sl->mb_y, ER_MB_ERROR);
2361 return AVERROR_INVALIDDATA;
2364 ret = ff_h264_decode_mb_cabac(h, sl);
2365 // STOP_TIMER("decode_mb_cabac")
2368 ff_h264_hl_decode_mb(h, sl);
2370 // FIXME optimal? or let mb_decode decode 16x32 ?
2371 if (ret >= 0 && FRAME_MBAFF(h)) {
2374 ret = ff_h264_decode_mb_cabac(h, sl);
2377 ff_h264_hl_decode_mb(h, sl);
2380 eos = get_cabac_terminate(&sl->cabac);
2382 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2383 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2384 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2385 sl->mb_y, ER_MB_END);
2386 if (sl->mb_x >= lf_x_start)
2387 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2390 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2391 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2392 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2393 av_log(h->avctx, AV_LOG_ERROR,
2394 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2396 sl->cabac.bytestream_end - sl->cabac.bytestream);
2397 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2398 sl->mb_y, ER_MB_ERROR);
2399 return AVERROR_INVALIDDATA;
2402 if (++sl->mb_x >= h->mb_width) {
2403 loop_filter(h, sl, lf_x_start, sl->mb_x);
2404 sl->mb_x = lf_x_start = 0;
2405 decode_finish_row(h, sl);
2407 if (FIELD_OR_MBAFF_PICTURE(h)) {
2409 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2410 predict_field_decoding_flag(h, sl);
2414 if (eos || sl->mb_y >= h->mb_height) {
2415 ff_tlog(h->avctx, "slice end %d %d\n",
2416 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2417 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2418 sl->mb_y, ER_MB_END);
2419 if (sl->mb_x > lf_x_start)
2420 loop_filter(h, sl, lf_x_start, sl->mb_x);
2428 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->mb_index_end) {
2429 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps next at %d\n",
2431 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2432 sl->mb_y, ER_MB_ERROR);
2433 return AVERROR_INVALIDDATA;
2436 ret = ff_h264_decode_mb_cavlc(h, sl);
2439 ff_h264_hl_decode_mb(h, sl);
2441 // FIXME optimal? or let mb_decode decode 16x32 ?
2442 if (ret >= 0 && FRAME_MBAFF(h)) {
2444 ret = ff_h264_decode_mb_cavlc(h, sl);
2447 ff_h264_hl_decode_mb(h, sl);
2452 av_log(h->avctx, AV_LOG_ERROR,
2453 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2454 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2455 sl->mb_y, ER_MB_ERROR);
2459 if (++sl->mb_x >= h->mb_width) {
2460 loop_filter(h, sl, lf_x_start, sl->mb_x);
2461 sl->mb_x = lf_x_start = 0;
2462 decode_finish_row(h, sl);
2464 if (FIELD_OR_MBAFF_PICTURE(h)) {
2466 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2467 predict_field_decoding_flag(h, sl);
2469 if (sl->mb_y >= h->mb_height) {
2470 ff_tlog(h->avctx, "slice end %d %d\n",
2471 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2473 if ( get_bits_left(&sl->gb) == 0
2474 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2475 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2476 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2480 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2481 sl->mb_x, sl->mb_y, ER_MB_END);
2483 return AVERROR_INVALIDDATA;
2488 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2489 ff_tlog(h->avctx, "slice end %d %d\n",
2490 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2492 if (get_bits_left(&sl->gb) == 0) {
2493 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2494 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2495 if (sl->mb_x > lf_x_start)
2496 loop_filter(h, sl, lf_x_start, sl->mb_x);
2500 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2501 sl->mb_y, ER_MB_ERROR);
2503 return AVERROR_INVALIDDATA;
2511 * Call decode_slice() for each context.
2513 * @param h h264 master context
2514 * @param context_count number of contexts to execute
2516 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2518 AVCodecContext *const avctx = h->avctx;
2519 H264SliceContext *sl;
2522 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2524 h->slice_ctx[0].mb_index_end = INT_MAX;
2526 if (h->avctx->hwaccel ||
2527 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2529 if (context_count == 1) {
2530 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2531 h->mb_y = h->slice_ctx[0].mb_y;
2535 av_assert0(context_count > 0);
2536 for (i = 0; i < context_count; i++) {
2537 int mb_index_end = h->mb_width * h->mb_height;
2538 sl = &h->slice_ctx[i];
2539 mb_index = sl->resync_mb_x + sl->resync_mb_y * h->mb_width;
2540 if (CONFIG_ERROR_RESILIENCE) {
2541 sl->er.error_count = 0;
2543 for (j = 0; j < context_count; j++) {
2544 H264SliceContext *sl2 = &h->slice_ctx[j];
2545 int mb_index2 = sl2->resync_mb_x + sl2->resync_mb_y * h->mb_width;
2547 if (i==j || mb_index > mb_index2)
2549 mb_index_end = FFMIN(mb_index_end, mb_index2);
2551 sl->mb_index_end = mb_index_end;
2554 avctx->execute(avctx, decode_slice, h->slice_ctx,
2555 NULL, context_count, sizeof(h->slice_ctx[0]));
2557 /* pull back stuff from slices to master context */
2558 sl = &h->slice_ctx[context_count - 1];
2560 if (CONFIG_ERROR_RESILIENCE) {
2561 for (i = 1; i < context_count; i++)
2562 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;