2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/avassert.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/timer.h"
33 #include "cabac_functions.h"
34 #include "error_resilience.h"
38 #include "h264chroma.h"
39 #include "h264_mvpred.h"
42 #include "mpegutils.h"
43 #include "rectangle.h"
47 static const uint8_t rem6[QP_MAX_NUM + 1] = {
48 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
49 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
50 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
51 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
55 static const uint8_t div6[QP_MAX_NUM + 1] = {
56 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
57 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
58 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
59 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
63 static const uint8_t field_scan[16+1] = {
64 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
65 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
66 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
67 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
70 static const uint8_t field_scan8x8[64+1] = {
71 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
72 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
73 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
74 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
75 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
76 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
77 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
78 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
79 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
80 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
81 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
82 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
83 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
84 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
85 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
86 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
89 static const uint8_t field_scan8x8_cavlc[64+1] = {
90 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
91 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
92 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
93 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
94 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
95 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
96 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
97 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
98 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
99 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
100 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
101 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
102 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
103 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
104 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
105 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
108 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
109 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
110 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
111 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
112 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
113 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
114 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
115 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
116 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
117 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
118 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
119 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
120 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
121 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
122 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
123 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
124 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
125 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
128 static const uint8_t dequant4_coeff_init[6][3] = {
137 static const uint8_t dequant8_coeff_init_scan[16] = {
138 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
141 static const uint8_t dequant8_coeff_init[6][6] = {
142 { 20, 18, 32, 19, 25, 24 },
143 { 22, 19, 35, 21, 28, 26 },
144 { 26, 23, 42, 24, 33, 31 },
145 { 28, 25, 45, 26, 35, 33 },
146 { 32, 28, 51, 30, 40, 38 },
147 { 36, 32, 58, 34, 46, 43 },
151 static void release_unused_pictures(H264Context *h, int remove_current)
155 /* release non reference frames */
156 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
157 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
158 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
159 ff_h264_unref_picture(h, &h->DPB[i]);
164 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
166 const H264Context *h = sl->h264;
167 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
169 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
170 // edge emu needs blocksize + filter length - 1
171 // (= 21x21 for h264)
172 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
174 av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
175 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
176 av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
177 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
179 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
180 !sl->top_borders[0] || !sl->top_borders[1]) {
181 av_freep(&sl->bipred_scratchpad);
182 av_freep(&sl->edge_emu_buffer);
183 av_freep(&sl->top_borders[0]);
184 av_freep(&sl->top_borders[1]);
186 sl->bipred_scratchpad_allocated = 0;
187 sl->edge_emu_buffer_allocated = 0;
188 sl->top_borders_allocated[0] = 0;
189 sl->top_borders_allocated[1] = 0;
190 return AVERROR(ENOMEM);
196 static int init_table_pools(H264Context *h)
198 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
199 const int mb_array_size = h->mb_stride * h->mb_height;
200 const int b4_stride = h->mb_width * 4 + 1;
201 const int b4_array_size = b4_stride * h->mb_height * 4;
203 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
205 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
206 sizeof(uint32_t), av_buffer_allocz);
207 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
208 sizeof(int16_t), av_buffer_allocz);
209 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
211 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
212 !h->ref_index_pool) {
213 av_buffer_pool_uninit(&h->qscale_table_pool);
214 av_buffer_pool_uninit(&h->mb_type_pool);
215 av_buffer_pool_uninit(&h->motion_val_pool);
216 av_buffer_pool_uninit(&h->ref_index_pool);
217 return AVERROR(ENOMEM);
223 static int alloc_picture(H264Context *h, H264Picture *pic)
227 av_assert0(!pic->f.data[0]);
230 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
231 AV_GET_BUFFER_FLAG_REF : 0);
235 h->linesize = pic->f.linesize[0];
236 h->uvlinesize = pic->f.linesize[1];
237 pic->crop = h->sps.crop;
238 pic->crop_top = h->sps.crop_top;
239 pic->crop_left= h->sps.crop_left;
241 if (h->avctx->hwaccel) {
242 const AVHWAccel *hwaccel = h->avctx->hwaccel;
243 av_assert0(!pic->hwaccel_picture_private);
244 if (hwaccel->frame_priv_data_size) {
245 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
246 if (!pic->hwaccel_priv_buf)
247 return AVERROR(ENOMEM);
248 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
251 if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
252 int h_chroma_shift, v_chroma_shift;
253 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
254 &h_chroma_shift, &v_chroma_shift);
256 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
257 memset(pic->f.data[1] + pic->f.linesize[1]*i,
258 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
259 memset(pic->f.data[2] + pic->f.linesize[2]*i,
260 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
264 if (!h->qscale_table_pool) {
265 ret = init_table_pools(h);
270 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
271 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
272 if (!pic->qscale_table_buf || !pic->mb_type_buf)
275 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
276 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
278 for (i = 0; i < 2; i++) {
279 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
280 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
281 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
284 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
285 pic->ref_index[i] = pic->ref_index_buf[i]->data;
290 ff_h264_unref_picture(h, pic);
291 return (ret < 0) ? ret : AVERROR(ENOMEM);
294 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
298 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
303 static int find_unused_picture(H264Context *h)
307 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
308 if (pic_is_unused(h, &h->DPB[i]))
311 if (i == H264_MAX_PICTURE_COUNT)
312 return AVERROR_INVALIDDATA;
314 if (h->DPB[i].needs_realloc) {
315 h->DPB[i].needs_realloc = 0;
316 ff_h264_unref_picture(h, &h->DPB[i]);
323 static void init_dequant8_coeff_table(H264Context *h)
326 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
328 for (i = 0; i < 6; i++) {
329 h->dequant8_coeff[i] = h->dequant8_buffer[i];
330 for (j = 0; j < i; j++)
331 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
332 64 * sizeof(uint8_t))) {
333 h->dequant8_coeff[i] = h->dequant8_buffer[j];
339 for (q = 0; q < max_qp + 1; q++) {
342 for (x = 0; x < 64; x++)
343 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
344 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
345 h->pps.scaling_matrix8[i][x]) << shift;
350 static void init_dequant4_coeff_table(H264Context *h)
353 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
354 for (i = 0; i < 6; i++) {
355 h->dequant4_coeff[i] = h->dequant4_buffer[i];
356 for (j = 0; j < i; j++)
357 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
358 16 * sizeof(uint8_t))) {
359 h->dequant4_coeff[i] = h->dequant4_buffer[j];
365 for (q = 0; q < max_qp + 1; q++) {
366 int shift = div6[q] + 2;
368 for (x = 0; x < 16; x++)
369 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
370 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
371 h->pps.scaling_matrix4[i][x]) << shift;
376 void ff_h264_init_dequant_tables(H264Context *h)
379 init_dequant4_coeff_table(h);
380 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
382 if (h->pps.transform_8x8_mode)
383 init_dequant8_coeff_table(h);
384 if (h->sps.transform_bypass) {
385 for (i = 0; i < 6; i++)
386 for (x = 0; x < 16; x++)
387 h->dequant4_coeff[i][0][x] = 1 << 6;
388 if (h->pps.transform_8x8_mode)
389 for (i = 0; i < 6; i++)
390 for (x = 0; x < 64; x++)
391 h->dequant8_coeff[i][0][x] = 1 << 6;
396 * Mimic alloc_tables(), but for every context thread.
398 static void clone_tables(H264Context *dst, H264SliceContext *sl,
399 H264Context *src, int i)
401 sl->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
402 sl->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
403 sl->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
405 dst->non_zero_count = src->non_zero_count;
406 dst->slice_table = src->slice_table;
407 dst->cbp_table = src->cbp_table;
408 dst->mb2b_xy = src->mb2b_xy;
409 dst->mb2br_xy = src->mb2br_xy;
410 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
411 dst->direct_table = src->direct_table;
412 dst->list_counts = src->list_counts;
414 dst->cur_pic_ptr = src->cur_pic_ptr;
415 dst->cur_pic = src->cur_pic;
416 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
417 src->sps.chroma_format_idc);
420 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
422 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
423 (((pic) && (pic) >= (old_ctx)->DPB && \
424 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
425 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
427 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
428 H264Context *new_base,
429 H264Context *old_base)
433 for (i = 0; i < count; i++) {
434 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
435 IN_RANGE(from[i], old_base->DPB,
436 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
438 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
442 static int copy_parameter_set(void **to, void **from, int count, int size)
446 for (i = 0; i < count; i++) {
447 if (to[i] && !from[i]) {
449 } else if (from[i] && !to[i]) {
450 to[i] = av_malloc(size);
452 return AVERROR(ENOMEM);
456 memcpy(to[i], from[i], size);
462 #define copy_fields(to, from, start_field, end_field) \
463 memcpy(&(to)->start_field, &(from)->start_field, \
464 (char *)&(to)->end_field - (char *)&(to)->start_field)
466 static int h264_slice_header_init(H264Context *h, int reinit);
468 int ff_h264_update_thread_context(AVCodecContext *dst,
469 const AVCodecContext *src)
471 H264Context *h = dst->priv_data, *h1 = src->priv_data;
472 int inited = h->context_initialized, err = 0;
473 int context_reinitialized = 0;
480 (h->width != h1->width ||
481 h->height != h1->height ||
482 h->mb_width != h1->mb_width ||
483 h->mb_height != h1->mb_height ||
484 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
485 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
486 h->sps.colorspace != h1->sps.colorspace)) {
488 /* set bits_per_raw_sample to the previous value. the check for changed
489 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
490 * the current value */
491 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
493 h->width = h1->width;
494 h->height = h1->height;
495 h->mb_height = h1->mb_height;
496 h->mb_width = h1->mb_width;
497 h->mb_num = h1->mb_num;
498 h->mb_stride = h1->mb_stride;
499 h->b_stride = h1->b_stride;
501 if ((ret = copy_parameter_set((void **)h->sps_buffers,
502 (void **)h1->sps_buffers,
503 MAX_SPS_COUNT, sizeof(SPS))) < 0)
506 if ((ret = copy_parameter_set((void **)h->pps_buffers,
507 (void **)h1->pps_buffers,
508 MAX_PPS_COUNT, sizeof(PPS))) < 0)
512 if ((err = h264_slice_header_init(h, 1)) < 0) {
513 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
516 context_reinitialized = 1;
519 h264_set_parameter_from_sps(h);
520 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
521 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
524 /* update linesize on resize for h264. The h264 decoder doesn't
525 * necessarily call ff_mpv_frame_start in the new thread */
526 h->linesize = h1->linesize;
527 h->uvlinesize = h1->uvlinesize;
529 /* copy block_offset since frame_start may not be called */
530 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
533 H264SliceContext *orig_slice_ctx = h->slice_ctx;
535 for (i = 0; i < MAX_SPS_COUNT; i++)
536 av_freep(h->sps_buffers + i);
538 for (i = 0; i < MAX_PPS_COUNT; i++)
539 av_freep(h->pps_buffers + i);
541 av_freep(&h->rbsp_buffer);
542 ff_h264_unref_picture(h, &h->last_pic_for_ec);
543 memcpy(h, h1, sizeof(H264Context));
545 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
546 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
548 memset(&h->er, 0, sizeof(h->er));
549 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
550 memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
552 h->slice_ctx = orig_slice_ctx;
554 memset(&h->slice_ctx[0].mb, 0, sizeof(h->slice_ctx[0].mb));
555 memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
556 memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
560 h->qscale_table_pool = NULL;
561 h->mb_type_pool = NULL;
562 h->ref_index_pool = NULL;
563 h->motion_val_pool = NULL;
564 h->intra4x4_pred_mode= NULL;
565 h->non_zero_count = NULL;
566 h->slice_table_base = NULL;
567 h->slice_table = NULL;
569 h->chroma_pred_mode_table = NULL;
570 memset(h->mvd_table, 0, sizeof(h->mvd_table));
571 h->direct_table = NULL;
572 h->list_counts = NULL;
575 h->rbsp_buffer = NULL;
576 h->rbsp_buffer_size = 0;
578 if (h1->context_initialized) {
579 h->context_initialized = 0;
581 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
582 av_frame_unref(&h->cur_pic.f);
583 h->cur_pic.tf.f = &h->cur_pic.f;
585 ret = ff_h264_alloc_tables(h);
587 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
590 ret = ff_h264_context_init(h);
592 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
598 h->thread_context[0] = h;
599 h->context_initialized = h1->context_initialized;
602 h->avctx->coded_height = h1->avctx->coded_height;
603 h->avctx->coded_width = h1->avctx->coded_width;
604 h->avctx->width = h1->avctx->width;
605 h->avctx->height = h1->avctx->height;
606 h->coded_picture_number = h1->coded_picture_number;
607 h->first_field = h1->first_field;
608 h->picture_structure = h1->picture_structure;
609 h->droppable = h1->droppable;
610 h->low_delay = h1->low_delay;
612 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
613 ff_h264_unref_picture(h, &h->DPB[i]);
614 if (h1->DPB && h1->DPB[i].f.buf[0] &&
615 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
619 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
620 ff_h264_unref_picture(h, &h->cur_pic);
621 if (h1->cur_pic.f.buf[0]) {
622 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
627 h->workaround_bugs = h1->workaround_bugs;
628 h->low_delay = h1->low_delay;
629 h->droppable = h1->droppable;
631 // extradata/NAL handling
632 h->is_avc = h1->is_avc;
635 if ((ret = copy_parameter_set((void **)h->sps_buffers,
636 (void **)h1->sps_buffers,
637 MAX_SPS_COUNT, sizeof(SPS))) < 0)
640 if ((ret = copy_parameter_set((void **)h->pps_buffers,
641 (void **)h1->pps_buffers,
642 MAX_PPS_COUNT, sizeof(PPS))) < 0)
646 // Dequantization matrices
647 // FIXME these are big - can they be only copied when PPS changes?
648 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
650 for (i = 0; i < 6; i++)
651 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
652 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
654 for (i = 0; i < 6; i++)
655 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
656 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
658 h->dequant_coeff_pps = h1->dequant_coeff_pps;
661 copy_fields(h, h1, poc_lsb, default_ref_list);
664 copy_fields(h, h1, short_ref, thread_context);
666 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
667 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
668 copy_picture_range(h->delayed_pic, h1->delayed_pic,
669 MAX_DELAYED_PIC_COUNT + 2, h, h1);
671 h->frame_recovered = h1->frame_recovered;
673 if (context_reinitialized)
674 ff_h264_set_parameter_from_sps(h);
680 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
681 h->prev_poc_msb = h->poc_msb;
682 h->prev_poc_lsb = h->poc_lsb;
684 h->prev_frame_num_offset = h->frame_num_offset;
685 h->prev_frame_num = h->frame_num;
686 h->outputed_poc = h->next_outputed_poc;
688 h->recovery_frame = h1->recovery_frame;
693 static int h264_frame_start(H264Context *h)
697 const int pixel_shift = h->pixel_shift;
699 1<<(h->sps.bit_depth_luma-1),
700 1<<(h->sps.bit_depth_chroma-1),
701 1<<(h->sps.bit_depth_chroma-1),
705 if (!ff_thread_can_start_frame(h->avctx)) {
706 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
710 release_unused_pictures(h, 1);
711 h->cur_pic_ptr = NULL;
713 i = find_unused_picture(h);
715 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
720 pic->reference = h->droppable ? 0 : h->picture_structure;
721 pic->f.coded_picture_number = h->coded_picture_number++;
722 pic->field_picture = h->picture_structure != PICT_FRAME;
725 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
727 * See decode_nal_units().
729 pic->f.key_frame = 0;
732 pic->invalid_gap = 0;
733 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
735 if ((ret = alloc_picture(h, pic)) < 0)
737 if(!h->frame_recovered && !h->avctx->hwaccel &&
738 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
739 avpriv_color_frame(&pic->f, c);
741 h->cur_pic_ptr = pic;
742 ff_h264_unref_picture(h, &h->cur_pic);
743 if (CONFIG_ERROR_RESILIENCE) {
744 ff_h264_set_erpic(&h->er.cur_pic, NULL);
747 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
750 if (CONFIG_ERROR_RESILIENCE) {
751 ff_er_frame_start(&h->er);
752 ff_h264_set_erpic(&h->er.last_pic, NULL);
753 ff_h264_set_erpic(&h->er.next_pic, NULL);
756 assert(h->linesize && h->uvlinesize);
758 for (i = 0; i < 16; i++) {
759 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
760 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
762 for (i = 0; i < 16; i++) {
763 h->block_offset[16 + i] =
764 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
765 h->block_offset[48 + 16 + i] =
766 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
769 /* We mark the current picture as non-reference after allocating it, so
770 * that if we break out due to an error it can be released automatically
771 * in the next ff_mpv_frame_start().
773 h->cur_pic_ptr->reference = 0;
775 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
777 h->next_output_pic = NULL;
779 assert(h->cur_pic_ptr->long_ref == 0);
784 static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *sl,
786 uint8_t *src_cb, uint8_t *src_cr,
787 int linesize, int uvlinesize,
792 const int pixel_shift = h->pixel_shift;
793 int chroma444 = CHROMA444(h);
794 int chroma422 = CHROMA422(h);
797 src_cb -= uvlinesize;
798 src_cr -= uvlinesize;
800 if (!simple && FRAME_MBAFF(h)) {
803 top_border = sl->top_borders[0][sl->mb_x];
804 AV_COPY128(top_border, src_y + 15 * linesize);
806 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
807 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
810 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
811 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
812 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
813 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
815 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
816 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
818 } else if (chroma422) {
820 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
821 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
823 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
824 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
828 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
829 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
831 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
832 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
837 } else if (MB_MBAFF(sl)) {
843 top_border = sl->top_borders[top_idx][sl->mb_x];
844 /* There are two lines saved, the line above the top macroblock
845 * of a pair, and the line above the bottom macroblock. */
846 AV_COPY128(top_border, src_y + 16 * linesize);
848 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
850 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
853 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
854 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
855 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
856 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
858 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
859 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
861 } else if (chroma422) {
863 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
864 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
866 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
867 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
871 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
872 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
874 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
875 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
882 * Initialize implicit_weight table.
883 * @param field 0/1 initialize the weight for interlaced MBAFF
884 * -1 initializes the rest
886 static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int field)
888 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
890 for (i = 0; i < 2; i++) {
891 sl->luma_weight_flag[i] = 0;
892 sl->chroma_weight_flag[i] = 0;
896 if (h->picture_structure == PICT_FRAME) {
897 cur_poc = h->cur_pic_ptr->poc;
899 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
901 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
902 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
904 sl->use_weight_chroma = 0;
908 ref_count0 = sl->ref_count[0];
909 ref_count1 = sl->ref_count[1];
911 cur_poc = h->cur_pic_ptr->field_poc[field];
913 ref_count0 = 16 + 2 * sl->ref_count[0];
914 ref_count1 = 16 + 2 * sl->ref_count[1];
918 sl->use_weight_chroma = 2;
919 sl->luma_log2_weight_denom = 5;
920 sl->chroma_log2_weight_denom = 5;
922 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
923 int poc0 = sl->ref_list[0][ref0].poc;
924 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
926 if (!sl->ref_list[0][ref0].long_ref && !sl->ref_list[1][ref1].long_ref) {
927 int poc1 = sl->ref_list[1][ref1].poc;
928 int td = av_clip_int8(poc1 - poc0);
930 int tb = av_clip_int8(cur_poc - poc0);
931 int tx = (16384 + (FFABS(td) >> 1)) / td;
932 int dist_scale_factor = (tb * tx + 32) >> 8;
933 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
934 w = 64 - dist_scale_factor;
938 sl->implicit_weight[ref0][ref1][0] =
939 sl->implicit_weight[ref0][ref1][1] = w;
941 sl->implicit_weight[ref0][ref1][field] = w;
948 * initialize scan tables
950 static void init_scan_tables(H264Context *h)
953 for (i = 0; i < 16; i++) {
954 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
955 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
956 h->field_scan[i] = TRANSPOSE(field_scan[i]);
959 for (i = 0; i < 64; i++) {
960 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
961 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
962 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
963 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
964 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
967 if (h->sps.transform_bypass) { // FIXME same ugly
968 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
969 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
970 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
971 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
972 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
973 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
975 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
976 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
977 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
978 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
979 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
980 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
985 * Replicate H264 "master" context to thread contexts.
987 static int clone_slice(H264Context *dst, H264Context *src)
989 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
990 dst->cur_pic_ptr = src->cur_pic_ptr;
991 dst->cur_pic = src->cur_pic;
992 dst->linesize = src->linesize;
993 dst->uvlinesize = src->uvlinesize;
994 dst->first_field = src->first_field;
996 dst->prev_poc_msb = src->prev_poc_msb;
997 dst->prev_poc_lsb = src->prev_poc_lsb;
998 dst->prev_frame_num_offset = src->prev_frame_num_offset;
999 dst->prev_frame_num = src->prev_frame_num;
1000 dst->short_ref_count = src->short_ref_count;
1002 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1003 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1004 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1006 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1007 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1012 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1014 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
1015 CONFIG_H264_VAAPI_HWACCEL + \
1016 (CONFIG_H264_VDA_HWACCEL * 2) + \
1017 CONFIG_H264_VDPAU_HWACCEL)
1018 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
1019 const enum AVPixelFormat *choices = pix_fmts;
1022 switch (h->sps.bit_depth_luma) {
1025 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1026 *fmt++ = AV_PIX_FMT_GBRP9;
1028 *fmt++ = AV_PIX_FMT_YUV444P9;
1029 } else if (CHROMA422(h))
1030 *fmt++ = AV_PIX_FMT_YUV422P9;
1032 *fmt++ = AV_PIX_FMT_YUV420P9;
1036 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1037 *fmt++ = AV_PIX_FMT_GBRP10;
1039 *fmt++ = AV_PIX_FMT_YUV444P10;
1040 } else if (CHROMA422(h))
1041 *fmt++ = AV_PIX_FMT_YUV422P10;
1043 *fmt++ = AV_PIX_FMT_YUV420P10;
1047 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1048 *fmt++ = AV_PIX_FMT_GBRP12;
1050 *fmt++ = AV_PIX_FMT_YUV444P12;
1051 } else if (CHROMA422(h))
1052 *fmt++ = AV_PIX_FMT_YUV422P12;
1054 *fmt++ = AV_PIX_FMT_YUV420P12;
1058 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1059 *fmt++ = AV_PIX_FMT_GBRP14;
1061 *fmt++ = AV_PIX_FMT_YUV444P14;
1062 } else if (CHROMA422(h))
1063 *fmt++ = AV_PIX_FMT_YUV422P14;
1065 *fmt++ = AV_PIX_FMT_YUV420P14;
1068 #if CONFIG_H264_VDPAU_HWACCEL
1069 *fmt++ = AV_PIX_FMT_VDPAU;
1072 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1073 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1074 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1075 *fmt++ = AV_PIX_FMT_GBRP;
1076 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1077 *fmt++ = AV_PIX_FMT_YUVJ444P;
1079 *fmt++ = AV_PIX_FMT_YUV444P;
1080 } else if (CHROMA422(h)) {
1081 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1082 *fmt++ = AV_PIX_FMT_YUVJ422P;
1084 *fmt++ = AV_PIX_FMT_YUV422P;
1086 #if CONFIG_H264_DXVA2_HWACCEL
1087 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1089 #if CONFIG_H264_VAAPI_HWACCEL
1090 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1092 #if CONFIG_H264_VDA_HWACCEL
1093 *fmt++ = AV_PIX_FMT_VDA_VLD;
1094 *fmt++ = AV_PIX_FMT_VDA;
1096 if (h->avctx->codec->pix_fmts)
1097 choices = h->avctx->codec->pix_fmts;
1098 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1099 *fmt++ = AV_PIX_FMT_YUVJ420P;
1101 *fmt++ = AV_PIX_FMT_YUV420P;
1105 av_log(h->avctx, AV_LOG_ERROR,
1106 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1107 return AVERROR_INVALIDDATA;
1110 *fmt = AV_PIX_FMT_NONE;
1112 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1113 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1115 return ff_thread_get_format(h->avctx, choices);
1118 /* export coded and cropped frame dimensions to AVCodecContext */
1119 static int init_dimensions(H264Context *h)
1121 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1122 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1123 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1124 h->sps.crop_right || h->sps.crop_bottom;
1125 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1126 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1128 /* handle container cropping */
1129 if (!crop_present &&
1130 FFALIGN(h->avctx->width, 16) == h->width &&
1131 FFALIGN(h->avctx->height, 16) == h->height) {
1132 width = h->avctx->width;
1133 height = h->avctx->height;
1136 if (width <= 0 || height <= 0) {
1137 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1139 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1140 return AVERROR_INVALIDDATA;
1142 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1143 h->sps.crop_bottom =
1153 h->avctx->coded_width = h->width;
1154 h->avctx->coded_height = h->height;
1155 h->avctx->width = width;
1156 h->avctx->height = height;
1161 static int h264_slice_header_init(H264Context *h, int reinit)
1163 int nb_slices = (HAVE_THREADS &&
1164 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1165 h->avctx->thread_count : 1;
1168 ff_set_sar(h->avctx, h->sps.sar);
1169 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1170 &h->chroma_x_shift, &h->chroma_y_shift);
1172 if (h->sps.timing_info_present_flag) {
1173 int64_t den = h->sps.time_scale;
1174 if (h->x264_build < 44U)
1176 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1177 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1181 ff_h264_free_tables(h, 0);
1183 h->prev_interlaced_frame = 1;
1185 init_scan_tables(h);
1186 ret = ff_h264_alloc_tables(h);
1188 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1192 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1195 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1197 max_slices = H264_MAX_THREADS;
1198 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1199 " reducing to %d\n", nb_slices, max_slices);
1200 nb_slices = max_slices;
1202 h->slice_context_count = nb_slices;
1204 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1205 ret = ff_h264_context_init(h);
1207 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1211 for (i = 1; i < h->slice_context_count; i++) {
1213 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1215 ret = AVERROR(ENOMEM);
1218 c->avctx = h->avctx;
1220 c->h264dsp = h->h264dsp;
1221 c->h264qpel = h->h264qpel;
1222 c->h264chroma = h->h264chroma;
1225 c->pixel_shift = h->pixel_shift;
1226 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1227 c->width = h->width;
1228 c->height = h->height;
1229 c->linesize = h->linesize;
1230 c->uvlinesize = h->uvlinesize;
1231 c->chroma_x_shift = h->chroma_x_shift;
1232 c->chroma_y_shift = h->chroma_y_shift;
1233 c->droppable = h->droppable;
1234 c->low_delay = h->low_delay;
1235 c->mb_width = h->mb_width;
1236 c->mb_height = h->mb_height;
1237 c->mb_stride = h->mb_stride;
1238 c->mb_num = h->mb_num;
1239 c->flags = h->flags;
1240 c->workaround_bugs = h->workaround_bugs;
1241 c->pict_type = h->pict_type;
1243 h->slice_ctx[i].h264 = c;
1245 init_scan_tables(c);
1246 clone_tables(c, &h->slice_ctx[i], h, i);
1247 c->context_initialized = 1;
1250 for (i = 0; i < h->slice_context_count; i++)
1251 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1252 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1257 h->context_initialized = 1;
1261 ff_h264_free_tables(h, 0);
1262 h->context_initialized = 0;
1266 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1269 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1270 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1271 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1278 * Decode a slice header.
1279 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1281 * @param h h264context
1282 * @param h0 h264 master context (differs from 'h' when doing sliced based
1283 * parallel decoding)
1285 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1287 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Context *h0)
1289 unsigned int first_mb_in_slice;
1290 unsigned int pps_id;
1292 unsigned int slice_type, tmp, i, j;
1293 int last_pic_structure, last_pic_droppable;
1295 int needs_reinit = 0;
1296 int field_pic_flag, bottom_field_flag;
1297 int first_slice = h == h0 && !h0->current_slice;
1298 int frame_num, picture_structure, droppable;
1301 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1302 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1304 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1306 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1307 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1308 ff_h264_field_end(h, sl, 1);
1311 h0->current_slice = 0;
1312 if (!h0->first_field) {
1313 if (h->cur_pic_ptr && !h->droppable) {
1314 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1315 h->picture_structure == PICT_BOTTOM_FIELD);
1317 h->cur_pic_ptr = NULL;
1321 slice_type = get_ue_golomb_31(&sl->gb);
1322 if (slice_type > 9) {
1323 av_log(h->avctx, AV_LOG_ERROR,
1324 "slice type %d too large at %d\n",
1325 slice_type, first_mb_in_slice);
1326 return AVERROR_INVALIDDATA;
1328 if (slice_type > 4) {
1330 sl->slice_type_fixed = 1;
1332 sl->slice_type_fixed = 0;
1334 slice_type = golomb_to_pict_type[slice_type];
1335 sl->slice_type = slice_type;
1336 sl->slice_type_nos = slice_type & 3;
1338 if (h->nal_unit_type == NAL_IDR_SLICE &&
1339 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1340 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1341 return AVERROR_INVALIDDATA;
1345 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1346 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1347 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1348 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1349 h->avctx->skip_frame >= AVDISCARD_ALL) {
1350 return SLICE_SKIPED;
1353 // to make a few old functions happy, it's wrong though
1354 h->pict_type = sl->slice_type;
1356 pps_id = get_ue_golomb(&sl->gb);
1357 if (pps_id >= MAX_PPS_COUNT) {
1358 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1359 return AVERROR_INVALIDDATA;
1361 if (!h0->pps_buffers[pps_id]) {
1362 av_log(h->avctx, AV_LOG_ERROR,
1363 "non-existing PPS %u referenced\n",
1365 return AVERROR_INVALIDDATA;
1367 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1368 av_log(h->avctx, AV_LOG_ERROR,
1369 "PPS change from %d to %d forbidden\n",
1370 h0->au_pps_id, pps_id);
1371 return AVERROR_INVALIDDATA;
1374 pps = h0->pps_buffers[pps_id];
1376 if (!h0->sps_buffers[pps->sps_id]) {
1377 av_log(h->avctx, AV_LOG_ERROR,
1378 "non-existing SPS %u referenced\n",
1380 return AVERROR_INVALIDDATA;
1383 h->pps = *h0->pps_buffers[pps_id];
1385 if (pps->sps_id != h->sps.sps_id ||
1386 pps->sps_id != h->current_sps_id ||
1387 h0->sps_buffers[pps->sps_id]->new) {
1390 av_log(h->avctx, AV_LOG_ERROR,
1391 "SPS changed in the middle of the frame\n");
1392 return AVERROR_INVALIDDATA;
1395 h->sps = *h0->sps_buffers[h->pps.sps_id];
1397 if (h->mb_width != h->sps.mb_width ||
1398 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1399 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1400 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1404 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1405 h->chroma_format_idc != h->sps.chroma_format_idc) {
1406 h->bit_depth_luma = h->sps.bit_depth_luma;
1407 h->chroma_format_idc = h->sps.chroma_format_idc;
1410 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1414 h->avctx->profile = ff_h264_get_profile(&h->sps);
1415 h->avctx->level = h->sps.level_idc;
1416 h->avctx->refs = h->sps.ref_frame_count;
1418 must_reinit = (h->context_initialized &&
1419 ( 16*h->sps.mb_width != h->avctx->coded_width
1420 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1421 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1422 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1423 || h->mb_width != h->sps.mb_width
1424 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1426 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1429 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1432 h->mb_width = h->sps.mb_width;
1433 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1434 h->mb_num = h->mb_width * h->mb_height;
1435 h->mb_stride = h->mb_width + 1;
1437 h->b_stride = h->mb_width * 4;
1439 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1441 h->width = 16 * h->mb_width;
1442 h->height = 16 * h->mb_height;
1444 ret = init_dimensions(h);
1448 if (h->sps.video_signal_type_present_flag) {
1449 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1451 if (h->sps.colour_description_present_flag) {
1452 if (h->avctx->colorspace != h->sps.colorspace)
1454 h->avctx->color_primaries = h->sps.color_primaries;
1455 h->avctx->color_trc = h->sps.color_trc;
1456 h->avctx->colorspace = h->sps.colorspace;
1460 if (h->context_initialized &&
1461 (must_reinit || needs_reinit)) {
1463 av_log(h->avctx, AV_LOG_ERROR,
1464 "changing width %d -> %d / height %d -> %d on "
1466 h->width, h->avctx->coded_width,
1467 h->height, h->avctx->coded_height,
1468 h0->current_slice + 1);
1469 return AVERROR_INVALIDDATA;
1472 av_assert1(first_slice);
1474 ff_h264_flush_change(h);
1476 if ((ret = get_pixel_format(h, 1)) < 0)
1478 h->avctx->pix_fmt = ret;
1480 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1481 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1483 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1484 av_log(h->avctx, AV_LOG_ERROR,
1485 "h264_slice_header_init() failed\n");
1489 if (!h->context_initialized) {
1491 av_log(h->avctx, AV_LOG_ERROR,
1492 "Cannot (re-)initialize context during parallel decoding.\n");
1493 return AVERROR_PATCHWELCOME;
1496 if ((ret = get_pixel_format(h, 1)) < 0)
1498 h->avctx->pix_fmt = ret;
1500 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1501 av_log(h->avctx, AV_LOG_ERROR,
1502 "h264_slice_header_init() failed\n");
1507 if (first_slice && h->dequant_coeff_pps != pps_id) {
1508 h->dequant_coeff_pps = pps_id;
1509 ff_h264_init_dequant_tables(h);
1512 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1514 if (h0->frame_num != frame_num) {
1515 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1516 h0->frame_num, frame_num);
1517 return AVERROR_INVALIDDATA;
1522 h->mb_aff_frame = 0;
1523 last_pic_structure = h0->picture_structure;
1524 last_pic_droppable = h0->droppable;
1525 droppable = h->nal_ref_idc == 0;
1526 if (h->sps.frame_mbs_only_flag) {
1527 picture_structure = PICT_FRAME;
1529 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1530 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1533 field_pic_flag = get_bits1(&sl->gb);
1535 if (field_pic_flag) {
1536 bottom_field_flag = get_bits1(&sl->gb);
1537 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1539 picture_structure = PICT_FRAME;
1540 h->mb_aff_frame = h->sps.mb_aff;
1543 if (h0->current_slice) {
1544 if (last_pic_structure != picture_structure ||
1545 last_pic_droppable != droppable) {
1546 av_log(h->avctx, AV_LOG_ERROR,
1547 "Changing field mode (%d -> %d) between slices is not allowed\n",
1548 last_pic_structure, h->picture_structure);
1549 return AVERROR_INVALIDDATA;
1550 } else if (!h0->cur_pic_ptr) {
1551 av_log(h->avctx, AV_LOG_ERROR,
1552 "unset cur_pic_ptr on slice %d\n",
1553 h0->current_slice + 1);
1554 return AVERROR_INVALIDDATA;
1558 h->picture_structure = picture_structure;
1559 h->droppable = droppable;
1560 h->frame_num = frame_num;
1561 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1563 if (h0->current_slice == 0) {
1564 /* Shorten frame num gaps so we don't have to allocate reference
1565 * frames just to throw them away */
1566 if (h->frame_num != h->prev_frame_num) {
1567 int unwrap_prev_frame_num = h->prev_frame_num;
1568 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1570 if (unwrap_prev_frame_num > h->frame_num)
1571 unwrap_prev_frame_num -= max_frame_num;
1573 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1574 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1575 if (unwrap_prev_frame_num < 0)
1576 unwrap_prev_frame_num += max_frame_num;
1578 h->prev_frame_num = unwrap_prev_frame_num;
1582 /* See if we have a decoded first field looking for a pair...
1583 * Here, we're using that to see if we should mark previously
1584 * decode frames as "finished".
1585 * We have to do that before the "dummy" in-between frame allocation,
1586 * since that can modify h->cur_pic_ptr. */
1587 if (h0->first_field) {
1588 assert(h0->cur_pic_ptr);
1589 assert(h0->cur_pic_ptr->f.buf[0]);
1590 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1592 /* Mark old field/frame as completed */
1593 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1594 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1595 last_pic_structure == PICT_BOTTOM_FIELD);
1598 /* figure out if we have a complementary field pair */
1599 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1600 /* Previous field is unmatched. Don't display it, but let it
1601 * remain for reference if marked as such. */
1602 if (last_pic_structure != PICT_FRAME) {
1603 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1604 last_pic_structure == PICT_TOP_FIELD);
1607 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1608 /* This and previous field were reference, but had
1609 * different frame_nums. Consider this field first in
1610 * pair. Throw away previous field except for reference
1612 if (last_pic_structure != PICT_FRAME) {
1613 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1614 last_pic_structure == PICT_TOP_FIELD);
1617 /* Second field in complementary pair */
1618 if (!((last_pic_structure == PICT_TOP_FIELD &&
1619 h->picture_structure == PICT_BOTTOM_FIELD) ||
1620 (last_pic_structure == PICT_BOTTOM_FIELD &&
1621 h->picture_structure == PICT_TOP_FIELD))) {
1622 av_log(h->avctx, AV_LOG_ERROR,
1623 "Invalid field mode combination %d/%d\n",
1624 last_pic_structure, h->picture_structure);
1625 h->picture_structure = last_pic_structure;
1626 h->droppable = last_pic_droppable;
1627 return AVERROR_INVALIDDATA;
1628 } else if (last_pic_droppable != h->droppable) {
1629 avpriv_request_sample(h->avctx,
1630 "Found reference and non-reference fields in the same frame, which");
1631 h->picture_structure = last_pic_structure;
1632 h->droppable = last_pic_droppable;
1633 return AVERROR_PATCHWELCOME;
1639 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1640 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1641 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1642 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1643 h->frame_num, h->prev_frame_num);
1644 if (!h->sps.gaps_in_frame_num_allowed_flag)
1645 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1646 h->last_pocs[i] = INT_MIN;
1647 ret = h264_frame_start(h);
1649 h0->first_field = 0;
1653 h->prev_frame_num++;
1654 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1655 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1656 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1657 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1658 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1659 ret = ff_generate_sliding_window_mmcos(h, 1);
1660 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1662 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1663 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1665 /* Error concealment: If a ref is missing, copy the previous ref
1667 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1668 * many assumptions about there being no actual duplicates.
1669 * FIXME: This does not copy padding for out-of-frame motion
1670 * vectors. Given we are concealing a lost frame, this probably
1671 * is not noticeable by comparison, but it should be fixed. */
1672 if (h->short_ref_count) {
1674 av_image_copy(h->short_ref[0]->f.data,
1675 h->short_ref[0]->f.linesize,
1676 (const uint8_t **)prev->f.data,
1681 h->short_ref[0]->poc = prev->poc + 2;
1683 h->short_ref[0]->frame_num = h->prev_frame_num;
1687 /* See if we have a decoded first field looking for a pair...
1688 * We're using that to see whether to continue decoding in that
1689 * frame, or to allocate a new one. */
1690 if (h0->first_field) {
1691 assert(h0->cur_pic_ptr);
1692 assert(h0->cur_pic_ptr->f.buf[0]);
1693 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1695 /* figure out if we have a complementary field pair */
1696 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1697 /* Previous field is unmatched. Don't display it, but let it
1698 * remain for reference if marked as such. */
1699 h0->missing_fields ++;
1700 h0->cur_pic_ptr = NULL;
1701 h0->first_field = FIELD_PICTURE(h);
1703 h0->missing_fields = 0;
1704 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1705 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1706 h0->picture_structure==PICT_BOTTOM_FIELD);
1707 /* This and the previous field had different frame_nums.
1708 * Consider this field first in pair. Throw away previous
1709 * one except for reference purposes. */
1710 h0->first_field = 1;
1711 h0->cur_pic_ptr = NULL;
1713 /* Second field in complementary pair */
1714 h0->first_field = 0;
1718 /* Frame or first field in a potentially complementary pair */
1719 h0->first_field = FIELD_PICTURE(h);
1722 if (!FIELD_PICTURE(h) || h0->first_field) {
1723 if (h264_frame_start(h) < 0) {
1724 h0->first_field = 0;
1725 return AVERROR_INVALIDDATA;
1728 release_unused_pictures(h, 0);
1730 /* Some macroblocks can be accessed before they're available in case
1731 * of lost slices, MBAFF or threading. */
1732 if (FIELD_PICTURE(h)) {
1733 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1734 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1736 memset(h->slice_table, -1,
1737 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1739 h0->last_slice_type = -1;
1741 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1745 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1747 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1748 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1749 first_mb_in_slice >= h->mb_num) {
1750 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1751 return AVERROR_INVALIDDATA;
1753 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1754 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1755 FIELD_OR_MBAFF_PICTURE(h);
1756 if (h->picture_structure == PICT_BOTTOM_FIELD)
1757 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1758 av_assert1(sl->mb_y < h->mb_height);
1760 if (h->picture_structure == PICT_FRAME) {
1761 h->curr_pic_num = h->frame_num;
1762 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1764 h->curr_pic_num = 2 * h->frame_num + 1;
1765 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1768 if (h->nal_unit_type == NAL_IDR_SLICE)
1769 get_ue_golomb(&sl->gb); /* idr_pic_id */
1771 if (h->sps.poc_type == 0) {
1772 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1774 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1775 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1778 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1779 h->delta_poc[0] = get_se_golomb(&sl->gb);
1781 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1782 h->delta_poc[1] = get_se_golomb(&sl->gb);
1785 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1787 if (h->pps.redundant_pic_cnt_present)
1788 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1790 ret = ff_set_ref_count(h, sl);
1794 if (slice_type != AV_PICTURE_TYPE_I &&
1795 (h0->current_slice == 0 ||
1796 slice_type != h0->last_slice_type ||
1797 memcmp(h0->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1799 ff_h264_fill_default_ref_list(h, sl);
1802 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1803 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1805 sl->ref_count[1] = sl->ref_count[0] = 0;
1810 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1811 (h->pps.weighted_bipred_idc == 1 &&
1812 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1813 ff_pred_weight_table(h, sl);
1814 else if (h->pps.weighted_bipred_idc == 2 &&
1815 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1816 implicit_weight_table(h, sl, -1);
1819 for (i = 0; i < 2; i++) {
1820 sl->luma_weight_flag[i] = 0;
1821 sl->chroma_weight_flag[i] = 0;
1825 // If frame-mt is enabled, only update mmco tables for the first slice
1826 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1827 // or h->mmco, which will cause ref list mix-ups and decoding errors
1828 // further down the line. This may break decoding if the first slice is
1829 // corrupt, thus we only do this if frame-mt is enabled.
1830 if (h->nal_ref_idc) {
1831 ret = ff_h264_decode_ref_pic_marking(h0, &sl->gb,
1832 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1833 h0->current_slice == 0);
1834 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1835 return AVERROR_INVALIDDATA;
1838 if (FRAME_MBAFF(h)) {
1839 ff_h264_fill_mbaff_ref_list(h, sl);
1841 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1842 implicit_weight_table(h, sl, 0);
1843 implicit_weight_table(h, sl, 1);
1847 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1848 ff_h264_direct_dist_scale_factor(h, sl);
1849 ff_h264_direct_ref_list_init(h, sl);
1851 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1852 tmp = get_ue_golomb_31(&sl->gb);
1854 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1855 return AVERROR_INVALIDDATA;
1857 sl->cabac_init_idc = tmp;
1860 sl->last_qscale_diff = 0;
1861 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1862 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1863 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1864 return AVERROR_INVALIDDATA;
1867 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1868 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1869 // FIXME qscale / qp ... stuff
1870 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1871 get_bits1(&sl->gb); /* sp_for_switch_flag */
1872 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1873 sl->slice_type == AV_PICTURE_TYPE_SI)
1874 get_se_golomb(&sl->gb); /* slice_qs_delta */
1876 sl->deblocking_filter = 1;
1877 sl->slice_alpha_c0_offset = 0;
1878 sl->slice_beta_offset = 0;
1879 if (h->pps.deblocking_filter_parameters_present) {
1880 tmp = get_ue_golomb_31(&sl->gb);
1882 av_log(h->avctx, AV_LOG_ERROR,
1883 "deblocking_filter_idc %u out of range\n", tmp);
1884 return AVERROR_INVALIDDATA;
1886 sl->deblocking_filter = tmp;
1887 if (sl->deblocking_filter < 2)
1888 sl->deblocking_filter ^= 1; // 1<->0
1890 if (sl->deblocking_filter) {
1891 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1892 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1893 if (sl->slice_alpha_c0_offset > 12 ||
1894 sl->slice_alpha_c0_offset < -12 ||
1895 sl->slice_beta_offset > 12 ||
1896 sl->slice_beta_offset < -12) {
1897 av_log(h->avctx, AV_LOG_ERROR,
1898 "deblocking filter parameters %d %d out of range\n",
1899 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1900 return AVERROR_INVALIDDATA;
1905 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1906 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1907 h->nal_unit_type != NAL_IDR_SLICE) ||
1908 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1909 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1910 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1911 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1912 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1913 h->nal_ref_idc == 0))
1914 sl->deblocking_filter = 0;
1916 if (sl->deblocking_filter == 1 && h0->max_contexts > 1) {
1917 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1918 /* Cheat slightly for speed:
1919 * Do not bother to deblock across slices. */
1920 sl->deblocking_filter = 2;
1922 h0->max_contexts = 1;
1923 if (!h0->single_decode_warning) {
1924 av_log(h->avctx, AV_LOG_INFO,
1925 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1926 "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"
1927 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1928 "but will generate non-standard-compliant output.\n");
1929 h0->single_decode_warning = 1;
1932 av_log(h->avctx, AV_LOG_ERROR,
1933 "Deblocking switched inside frame.\n");
1934 return SLICE_SINGLETHREAD;
1938 sl->qp_thresh = 15 -
1939 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1941 h->pps.chroma_qp_index_offset[0],
1942 h->pps.chroma_qp_index_offset[1]) +
1943 6 * (h->sps.bit_depth_luma - 8);
1945 h0->last_slice_type = slice_type;
1946 memcpy(h0->last_ref_count, sl->ref_count, sizeof(h0->last_ref_count));
1947 sl->slice_num = ++h0->current_slice;
1950 h0->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1951 if ( h0->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1952 && h0->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1953 && sl->slice_num >= MAX_SLICES) {
1954 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1955 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);
1958 for (j = 0; j < 2; j++) {
1960 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1961 for (i = 0; i < 16; i++) {
1963 if (j < sl->list_count && i < sl->ref_count[j] &&
1964 sl->ref_list[j][i].f.buf[0]) {
1966 AVBuffer *buf = sl->ref_list[j][i].f.buf[0]->buffer;
1967 for (k = 0; k < h->short_ref_count; k++)
1968 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1972 for (k = 0; k < h->long_ref_count; k++)
1973 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1974 id_list[i] = h->short_ref_count + k;
1982 for (i = 0; i < 16; i++)
1983 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1985 ref2frm[18 + 1] = -1;
1986 for (i = 16; i < 48; i++)
1987 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1988 (sl->ref_list[j][i].reference & 3);
1991 h0->au_pps_id = pps_id;
1993 h0->sps_buffers[h->pps.sps_id]->new = 0;
1994 h->current_sps_id = h->pps.sps_id;
1996 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1997 av_log(h->avctx, AV_LOG_DEBUG,
1998 "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",
2000 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
2002 av_get_picture_type_char(sl->slice_type),
2003 sl->slice_type_fixed ? " fix" : "",
2004 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2005 pps_id, h->frame_num,
2006 h->cur_pic_ptr->field_poc[0],
2007 h->cur_pic_ptr->field_poc[1],
2008 sl->ref_count[0], sl->ref_count[1],
2010 sl->deblocking_filter,
2011 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
2013 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
2014 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2020 int ff_h264_get_slice_type(const H264SliceContext *sl)
2022 switch (sl->slice_type) {
2023 case AV_PICTURE_TYPE_P:
2025 case AV_PICTURE_TYPE_B:
2027 case AV_PICTURE_TYPE_I:
2029 case AV_PICTURE_TYPE_SP:
2031 case AV_PICTURE_TYPE_SI:
2034 return AVERROR_INVALIDDATA;
2038 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2039 H264SliceContext *sl,
2040 int mb_type, int top_xy,
2041 int left_xy[LEFT_MBS],
2043 int left_type[LEFT_MBS],
2044 int mb_xy, int list)
2046 int b_stride = h->b_stride;
2047 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
2048 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
2049 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2050 if (USES_LIST(top_type, list)) {
2051 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2052 const int b8_xy = 4 * top_xy + 2;
2053 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2054 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2055 ref_cache[0 - 1 * 8] =
2056 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2057 ref_cache[2 - 1 * 8] =
2058 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2060 AV_ZERO128(mv_dst - 1 * 8);
2061 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2064 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2065 if (USES_LIST(left_type[LTOP], list)) {
2066 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2067 const int b8_xy = 4 * left_xy[LTOP] + 1;
2068 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2069 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2070 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2071 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2072 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2074 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2075 ref_cache[-1 + 16] =
2076 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2078 AV_ZERO32(mv_dst - 1 + 0);
2079 AV_ZERO32(mv_dst - 1 + 8);
2080 AV_ZERO32(mv_dst - 1 + 16);
2081 AV_ZERO32(mv_dst - 1 + 24);
2084 ref_cache[-1 + 16] =
2085 ref_cache[-1 + 24] = LIST_NOT_USED;
2090 if (!USES_LIST(mb_type, list)) {
2091 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2092 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2093 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2094 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2095 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2100 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2101 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
2102 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2103 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2104 AV_WN32A(&ref_cache[0 * 8], ref01);
2105 AV_WN32A(&ref_cache[1 * 8], ref01);
2106 AV_WN32A(&ref_cache[2 * 8], ref23);
2107 AV_WN32A(&ref_cache[3 * 8], ref23);
2111 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2112 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2113 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2114 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2115 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2121 * @return non zero if the loop filter can be skipped
2123 static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
2125 const int mb_xy = sl->mb_xy;
2126 int top_xy, left_xy[LEFT_MBS];
2127 int top_type, left_type[LEFT_MBS];
2131 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2133 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2134 * stuff, I can't imagine that these complex rules are worth it. */
2136 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2137 if (FRAME_MBAFF(h)) {
2138 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2139 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2141 if (left_mb_field_flag != curr_mb_field_flag)
2142 left_xy[LTOP] -= h->mb_stride;
2144 if (curr_mb_field_flag)
2145 top_xy += h->mb_stride &
2146 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2147 if (left_mb_field_flag != curr_mb_field_flag)
2148 left_xy[LBOT] += h->mb_stride;
2152 sl->top_mb_xy = top_xy;
2153 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2154 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2156 /* For sufficiently low qp, filtering wouldn't do anything.
2157 * This is a conservative estimate: could also check beta_offset
2158 * and more accurate chroma_qp. */
2159 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2160 int qp = h->cur_pic.qscale_table[mb_xy];
2161 if (qp <= qp_thresh &&
2162 (left_xy[LTOP] < 0 ||
2163 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2165 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2166 if (!FRAME_MBAFF(h))
2168 if ((left_xy[LTOP] < 0 ||
2169 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2170 (top_xy < h->mb_stride ||
2171 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2176 top_type = h->cur_pic.mb_type[top_xy];
2177 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2178 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2179 if (sl->deblocking_filter == 2) {
2180 if (h->slice_table[top_xy] != sl->slice_num)
2182 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2183 left_type[LTOP] = left_type[LBOT] = 0;
2185 if (h->slice_table[top_xy] == 0xFFFF)
2187 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2188 left_type[LTOP] = left_type[LBOT] = 0;
2190 sl->top_type = top_type;
2191 sl->left_type[LTOP] = left_type[LTOP];
2192 sl->left_type[LBOT] = left_type[LBOT];
2194 if (IS_INTRA(mb_type))
2197 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2198 top_type, left_type, mb_xy, 0);
2199 if (sl->list_count == 2)
2200 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2201 top_type, left_type, mb_xy, 1);
2203 nnz = h->non_zero_count[mb_xy];
2204 nnz_cache = sl->non_zero_count_cache;
2205 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2206 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2207 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2208 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2209 sl->cbp = h->cbp_table[mb_xy];
2212 nnz = h->non_zero_count[top_xy];
2213 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2216 if (left_type[LTOP]) {
2217 nnz = h->non_zero_count[left_xy[LTOP]];
2218 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2219 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2220 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2221 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2224 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2225 * from what the loop filter needs */
2226 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2227 if (IS_8x8DCT(top_type)) {
2228 nnz_cache[4 + 8 * 0] =
2229 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2230 nnz_cache[6 + 8 * 0] =
2231 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2233 if (IS_8x8DCT(left_type[LTOP])) {
2234 nnz_cache[3 + 8 * 1] =
2235 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2237 if (IS_8x8DCT(left_type[LBOT])) {
2238 nnz_cache[3 + 8 * 3] =
2239 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2242 if (IS_8x8DCT(mb_type)) {
2243 nnz_cache[scan8[0]] =
2244 nnz_cache[scan8[1]] =
2245 nnz_cache[scan8[2]] =
2246 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2248 nnz_cache[scan8[0 + 4]] =
2249 nnz_cache[scan8[1 + 4]] =
2250 nnz_cache[scan8[2 + 4]] =
2251 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2253 nnz_cache[scan8[0 + 8]] =
2254 nnz_cache[scan8[1 + 8]] =
2255 nnz_cache[scan8[2 + 8]] =
2256 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2258 nnz_cache[scan8[0 + 12]] =
2259 nnz_cache[scan8[1 + 12]] =
2260 nnz_cache[scan8[2 + 12]] =
2261 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2268 static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2270 uint8_t *dest_y, *dest_cb, *dest_cr;
2271 int linesize, uvlinesize, mb_x, mb_y;
2272 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2273 const int old_slice_type = sl->slice_type;
2274 const int pixel_shift = h->pixel_shift;
2275 const int block_h = 16 >> h->chroma_y_shift;
2277 if (sl->deblocking_filter) {
2278 for (mb_x = start_x; mb_x < end_x; mb_x++)
2279 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2281 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2282 sl->slice_num = h->slice_table[mb_xy];
2283 mb_type = h->cur_pic.mb_type[mb_xy];
2284 sl->list_count = h->list_counts[mb_xy];
2288 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2292 dest_y = h->cur_pic.f.data[0] +
2293 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2294 dest_cb = h->cur_pic.f.data[1] +
2295 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2296 mb_y * h->uvlinesize * block_h;
2297 dest_cr = h->cur_pic.f.data[2] +
2298 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2299 mb_y * h->uvlinesize * block_h;
2300 // FIXME simplify above
2303 linesize = sl->mb_linesize = h->linesize * 2;
2304 uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
2305 if (mb_y & 1) { // FIXME move out of this function?
2306 dest_y -= h->linesize * 15;
2307 dest_cb -= h->uvlinesize * (block_h - 1);
2308 dest_cr -= h->uvlinesize * (block_h - 1);
2311 linesize = sl->mb_linesize = h->linesize;
2312 uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
2314 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2316 if (fill_filter_caches(h, sl, mb_type))
2318 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2319 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2321 if (FRAME_MBAFF(h)) {
2322 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2323 linesize, uvlinesize);
2325 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2326 dest_cr, linesize, uvlinesize);
2330 sl->slice_type = old_slice_type;
2332 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2333 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2334 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2337 static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
2339 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2340 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2341 h->cur_pic.mb_type[mb_xy - 1] :
2342 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2343 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2344 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2348 * Draw edges and report progress for the last MB row.
2350 static void decode_finish_row(H264Context *h, H264SliceContext *sl)
2352 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2353 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2354 int height = 16 << FRAME_MBAFF(h);
2355 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2357 if (sl->deblocking_filter) {
2358 if ((top + height) >= pic_height)
2359 height += deblock_border;
2360 top -= deblock_border;
2363 if (top >= pic_height || (top + height) < 0)
2366 height = FFMIN(height, pic_height - top);
2368 height = top + height;
2372 ff_h264_draw_horiz_band(h, sl, top, height);
2374 if (h->droppable || h->er.error_occurred)
2377 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2378 h->picture_structure == PICT_BOTTOM_FIELD);
2381 static void er_add_slice(H264Context *h, H264SliceContext *sl,
2382 int startx, int starty,
2383 int endx, int endy, int status)
2385 if (CONFIG_ERROR_RESILIENCE) {
2386 ERContext *er = &h->er;
2388 ff_er_add_slice(er, startx, starty, endx, endy, status);
2392 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2394 H264SliceContext *sl = arg;
2395 H264Context *h = sl->h264;
2396 int lf_x_start = sl->mb_x;
2399 ret = alloc_scratch_buffers(sl, h->linesize);
2403 sl->mb_skip_run = -1;
2405 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2407 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2408 avctx->codec_id != AV_CODEC_ID_H264 ||
2409 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2411 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2412 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2414 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2415 prev_status &= ~ VP_START;
2416 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2417 h->er.error_occurred = 1;
2423 align_get_bits(&sl->gb);
2426 ff_init_cabac_decoder(&sl->cabac,
2427 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2428 (get_bits_left(&sl->gb) + 7) / 8);
2430 ff_h264_init_cabac_states(h, sl);
2434 int ret = ff_h264_decode_mb_cabac(h, sl);
2436 // STOP_TIMER("decode_mb_cabac")
2439 ff_h264_hl_decode_mb(h, sl);
2441 // FIXME optimal? or let mb_decode decode 16x32 ?
2442 if (ret >= 0 && FRAME_MBAFF(h)) {
2445 ret = ff_h264_decode_mb_cabac(h, sl);
2448 ff_h264_hl_decode_mb(h, sl);
2451 eos = get_cabac_terminate(&sl->cabac);
2453 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2454 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2455 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2456 sl->mb_y, ER_MB_END);
2457 if (sl->mb_x >= lf_x_start)
2458 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2461 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2462 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2463 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2464 av_log(h->avctx, AV_LOG_ERROR,
2465 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2467 sl->cabac.bytestream_end - sl->cabac.bytestream);
2468 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2469 sl->mb_y, ER_MB_ERROR);
2470 return AVERROR_INVALIDDATA;
2473 if (++sl->mb_x >= h->mb_width) {
2474 loop_filter(h, sl, lf_x_start, sl->mb_x);
2475 sl->mb_x = lf_x_start = 0;
2476 decode_finish_row(h, sl);
2478 if (FIELD_OR_MBAFF_PICTURE(h)) {
2480 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2481 predict_field_decoding_flag(h, sl);
2485 if (eos || sl->mb_y >= h->mb_height) {
2486 tprintf(h->avctx, "slice end %d %d\n",
2487 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2488 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2489 sl->mb_y, ER_MB_END);
2490 if (sl->mb_x > lf_x_start)
2491 loop_filter(h, sl, lf_x_start, sl->mb_x);
2497 int ret = ff_h264_decode_mb_cavlc(h, sl);
2500 ff_h264_hl_decode_mb(h, sl);
2502 // FIXME optimal? or let mb_decode decode 16x32 ?
2503 if (ret >= 0 && FRAME_MBAFF(h)) {
2505 ret = ff_h264_decode_mb_cavlc(h, sl);
2508 ff_h264_hl_decode_mb(h, sl);
2513 av_log(h->avctx, AV_LOG_ERROR,
2514 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2515 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2516 sl->mb_y, ER_MB_ERROR);
2520 if (++sl->mb_x >= h->mb_width) {
2521 loop_filter(h, sl, lf_x_start, sl->mb_x);
2522 sl->mb_x = lf_x_start = 0;
2523 decode_finish_row(h, sl);
2525 if (FIELD_OR_MBAFF_PICTURE(h)) {
2527 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2528 predict_field_decoding_flag(h, sl);
2530 if (sl->mb_y >= h->mb_height) {
2531 tprintf(h->avctx, "slice end %d %d\n",
2532 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2534 if ( get_bits_left(&sl->gb) == 0
2535 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2536 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
2537 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2541 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
2542 sl->mb_x, sl->mb_y, ER_MB_END);
2544 return AVERROR_INVALIDDATA;
2549 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2550 tprintf(h->avctx, "slice end %d %d\n",
2551 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2553 if (get_bits_left(&sl->gb) == 0) {
2554 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
2555 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2556 if (sl->mb_x > lf_x_start)
2557 loop_filter(h, sl, lf_x_start, sl->mb_x);
2561 er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2562 sl->mb_y, ER_MB_ERROR);
2564 return AVERROR_INVALIDDATA;
2572 * Call decode_slice() for each context.
2574 * @param h h264 master context
2575 * @param context_count number of contexts to execute
2577 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2579 AVCodecContext *const avctx = h->avctx;
2581 H264SliceContext *sl;
2584 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2586 if (h->avctx->hwaccel ||
2587 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2589 if (context_count == 1) {
2590 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2591 h->mb_y = h->slice_ctx[0].mb_y;
2594 av_assert0(context_count > 0);
2595 for (i = 1; i < context_count; i++) {
2596 hx = h->thread_context[i];
2597 if (CONFIG_ERROR_RESILIENCE) {
2598 hx->er.error_count = 0;
2600 hx->x264_build = h->x264_build;
2603 avctx->execute(avctx, decode_slice, h->slice_ctx,
2604 NULL, context_count, sizeof(h->slice_ctx[0]));
2606 /* pull back stuff from slices to master context */
2607 hx = h->thread_context[context_count - 1];
2608 sl = &h->slice_ctx[context_count - 1];
2610 h->droppable = hx->droppable;
2611 h->picture_structure = hx->picture_structure;
2612 if (CONFIG_ERROR_RESILIENCE) {
2613 for (i = 1; i < context_count; i++)
2614 h->er.error_count += h->thread_context[i]->er.error_count;