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 },
150 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
151 #if CONFIG_H264_DXVA2_HWACCEL
152 AV_PIX_FMT_DXVA2_VLD,
154 #if CONFIG_H264_VAAPI_HWACCEL
155 AV_PIX_FMT_VAAPI_VLD,
157 #if CONFIG_H264_VDA_HWACCEL
161 #if CONFIG_H264_VDPAU_HWACCEL
168 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
169 #if CONFIG_H264_DXVA2_HWACCEL
170 AV_PIX_FMT_DXVA2_VLD,
172 #if CONFIG_H264_VAAPI_HWACCEL
173 AV_PIX_FMT_VAAPI_VLD,
175 #if CONFIG_H264_VDA_HWACCEL
179 #if CONFIG_H264_VDPAU_HWACCEL
187 static void release_unused_pictures(H264Context *h, int remove_current)
191 /* release non reference frames */
192 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
193 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
194 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
195 ff_h264_unref_picture(h, &h->DPB[i]);
200 static int alloc_scratch_buffers(H264Context *h, int linesize)
202 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
204 if (h->bipred_scratchpad)
207 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
208 // edge emu needs blocksize + filter length - 1
209 // (= 21x21 for h264)
210 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
212 if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
213 av_freep(&h->bipred_scratchpad);
214 av_freep(&h->edge_emu_buffer);
215 return AVERROR(ENOMEM);
221 static int init_table_pools(H264Context *h)
223 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
224 const int mb_array_size = h->mb_stride * h->mb_height;
225 const int b4_stride = h->mb_width * 4 + 1;
226 const int b4_array_size = b4_stride * h->mb_height * 4;
228 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
230 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
231 sizeof(uint32_t), av_buffer_allocz);
232 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
233 sizeof(int16_t), av_buffer_allocz);
234 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
236 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
237 !h->ref_index_pool) {
238 av_buffer_pool_uninit(&h->qscale_table_pool);
239 av_buffer_pool_uninit(&h->mb_type_pool);
240 av_buffer_pool_uninit(&h->motion_val_pool);
241 av_buffer_pool_uninit(&h->ref_index_pool);
242 return AVERROR(ENOMEM);
248 static int alloc_picture(H264Context *h, H264Picture *pic)
252 av_assert0(!pic->f.data[0]);
255 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
256 AV_GET_BUFFER_FLAG_REF : 0);
260 h->linesize = pic->f.linesize[0];
261 h->uvlinesize = pic->f.linesize[1];
262 pic->crop = h->sps.crop;
263 pic->crop_top = h->sps.crop_top;
264 pic->crop_left= h->sps.crop_left;
266 if (h->avctx->hwaccel) {
267 const AVHWAccel *hwaccel = h->avctx->hwaccel;
268 av_assert0(!pic->hwaccel_picture_private);
269 if (hwaccel->frame_priv_data_size) {
270 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
271 if (!pic->hwaccel_priv_buf)
272 return AVERROR(ENOMEM);
273 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
276 if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
277 int h_chroma_shift, v_chroma_shift;
278 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
279 &h_chroma_shift, &v_chroma_shift);
281 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
282 memset(pic->f.data[1] + pic->f.linesize[1]*i,
283 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
284 memset(pic->f.data[2] + pic->f.linesize[2]*i,
285 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
289 if (!h->qscale_table_pool) {
290 ret = init_table_pools(h);
295 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
296 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
297 if (!pic->qscale_table_buf || !pic->mb_type_buf)
300 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
301 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
303 for (i = 0; i < 2; i++) {
304 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
305 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
306 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
309 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
310 pic->ref_index[i] = pic->ref_index_buf[i]->data;
315 ff_h264_unref_picture(h, pic);
316 return (ret < 0) ? ret : AVERROR(ENOMEM);
319 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
323 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
328 static int find_unused_picture(H264Context *h)
332 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
333 if (pic_is_unused(h, &h->DPB[i]))
336 if (i == H264_MAX_PICTURE_COUNT)
337 return AVERROR_INVALIDDATA;
339 if (h->DPB[i].needs_realloc) {
340 h->DPB[i].needs_realloc = 0;
341 ff_h264_unref_picture(h, &h->DPB[i]);
348 static void init_dequant8_coeff_table(H264Context *h)
351 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
353 for (i = 0; i < 6; i++) {
354 h->dequant8_coeff[i] = h->dequant8_buffer[i];
355 for (j = 0; j < i; j++)
356 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
357 64 * sizeof(uint8_t))) {
358 h->dequant8_coeff[i] = h->dequant8_buffer[j];
364 for (q = 0; q < max_qp + 1; q++) {
367 for (x = 0; x < 64; x++)
368 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
369 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
370 h->pps.scaling_matrix8[i][x]) << shift;
375 static void init_dequant4_coeff_table(H264Context *h)
378 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
379 for (i = 0; i < 6; i++) {
380 h->dequant4_coeff[i] = h->dequant4_buffer[i];
381 for (j = 0; j < i; j++)
382 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
383 16 * sizeof(uint8_t))) {
384 h->dequant4_coeff[i] = h->dequant4_buffer[j];
390 for (q = 0; q < max_qp + 1; q++) {
391 int shift = div6[q] + 2;
393 for (x = 0; x < 16; x++)
394 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
395 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
396 h->pps.scaling_matrix4[i][x]) << shift;
401 void h264_init_dequant_tables(H264Context *h)
404 init_dequant4_coeff_table(h);
405 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
407 if (h->pps.transform_8x8_mode)
408 init_dequant8_coeff_table(h);
409 if (h->sps.transform_bypass) {
410 for (i = 0; i < 6; i++)
411 for (x = 0; x < 16; x++)
412 h->dequant4_coeff[i][0][x] = 1 << 6;
413 if (h->pps.transform_8x8_mode)
414 for (i = 0; i < 6; i++)
415 for (x = 0; x < 64; x++)
416 h->dequant8_coeff[i][0][x] = 1 << 6;
421 * Mimic alloc_tables(), but for every context thread.
423 static void clone_tables(H264Context *dst, H264Context *src, int i)
425 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
426 dst->non_zero_count = src->non_zero_count;
427 dst->slice_table = src->slice_table;
428 dst->cbp_table = src->cbp_table;
429 dst->mb2b_xy = src->mb2b_xy;
430 dst->mb2br_xy = src->mb2br_xy;
431 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
432 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
433 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
434 dst->direct_table = src->direct_table;
435 dst->list_counts = src->list_counts;
437 dst->cur_pic_ptr = src->cur_pic_ptr;
438 dst->cur_pic = src->cur_pic;
439 dst->bipred_scratchpad = NULL;
440 dst->edge_emu_buffer = NULL;
441 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
442 src->sps.chroma_format_idc);
445 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
446 #undef REBASE_PICTURE
447 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
448 (((pic) && (pic) >= (old_ctx)->DPB && \
449 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
450 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
452 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
453 H264Context *new_base,
454 H264Context *old_base)
458 for (i = 0; i < count; i++) {
459 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
460 IN_RANGE(from[i], old_base->DPB,
461 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
463 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
467 static int copy_parameter_set(void **to, void **from, int count, int size)
471 for (i = 0; i < count; i++) {
472 if (to[i] && !from[i]) {
474 } else if (from[i] && !to[i]) {
475 to[i] = av_malloc(size);
477 return AVERROR(ENOMEM);
481 memcpy(to[i], from[i], size);
487 #define copy_fields(to, from, start_field, end_field) \
488 memcpy(&(to)->start_field, &(from)->start_field, \
489 (char *)&(to)->end_field - (char *)&(to)->start_field)
491 static int h264_slice_header_init(H264Context *h, int reinit);
493 int ff_h264_update_thread_context(AVCodecContext *dst,
494 const AVCodecContext *src)
496 H264Context *h = dst->priv_data, *h1 = src->priv_data;
497 int inited = h->context_initialized, err = 0;
498 int context_reinitialized = 0;
505 (h->width != h1->width ||
506 h->height != h1->height ||
507 h->mb_width != h1->mb_width ||
508 h->mb_height != h1->mb_height ||
509 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
510 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
511 h->sps.colorspace != h1->sps.colorspace)) {
513 /* set bits_per_raw_sample to the previous value. the check for changed
514 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
515 * the current value */
516 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
518 av_freep(&h->bipred_scratchpad);
520 h->width = h1->width;
521 h->height = h1->height;
522 h->mb_height = h1->mb_height;
523 h->mb_width = h1->mb_width;
524 h->mb_num = h1->mb_num;
525 h->mb_stride = h1->mb_stride;
526 h->b_stride = h1->b_stride;
528 if ((ret = copy_parameter_set((void **)h->sps_buffers,
529 (void **)h1->sps_buffers,
530 MAX_SPS_COUNT, sizeof(SPS))) < 0)
533 if ((ret = copy_parameter_set((void **)h->pps_buffers,
534 (void **)h1->pps_buffers,
535 MAX_PPS_COUNT, sizeof(PPS))) < 0)
539 if ((err = h264_slice_header_init(h, 1)) < 0) {
540 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
543 context_reinitialized = 1;
546 h264_set_parameter_from_sps(h);
547 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
548 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
551 /* update linesize on resize for h264. The h264 decoder doesn't
552 * necessarily call ff_mpv_frame_start in the new thread */
553 h->linesize = h1->linesize;
554 h->uvlinesize = h1->uvlinesize;
556 /* copy block_offset since frame_start may not be called */
557 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
560 for (i = 0; i < MAX_SPS_COUNT; i++)
561 av_freep(h->sps_buffers + i);
563 for (i = 0; i < MAX_PPS_COUNT; i++)
564 av_freep(h->pps_buffers + i);
566 av_freep(&h->rbsp_buffer[0]);
567 av_freep(&h->rbsp_buffer[1]);
568 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
569 memcpy(&h->cabac, &h1->cabac,
570 sizeof(H264Context) - offsetof(H264Context, cabac));
571 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
573 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
574 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
576 memset(&h->er, 0, sizeof(h->er));
577 memset(&h->mb, 0, sizeof(h->mb));
578 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
579 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
580 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
584 h->qscale_table_pool = NULL;
585 h->mb_type_pool = NULL;
586 h->ref_index_pool = NULL;
587 h->motion_val_pool = NULL;
588 for (i = 0; i < 2; i++) {
589 h->rbsp_buffer[i] = NULL;
590 h->rbsp_buffer_size[i] = 0;
593 if (h1->context_initialized) {
594 h->context_initialized = 0;
596 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
597 av_frame_unref(&h->cur_pic.f);
598 h->cur_pic.tf.f = &h->cur_pic.f;
600 ret = ff_h264_alloc_tables(h);
602 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
605 ret = ff_h264_context_init(h);
607 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
612 h->bipred_scratchpad = NULL;
613 h->edge_emu_buffer = NULL;
615 h->thread_context[0] = h;
616 h->context_initialized = h1->context_initialized;
619 h->avctx->coded_height = h1->avctx->coded_height;
620 h->avctx->coded_width = h1->avctx->coded_width;
621 h->avctx->width = h1->avctx->width;
622 h->avctx->height = h1->avctx->height;
623 h->coded_picture_number = h1->coded_picture_number;
624 h->first_field = h1->first_field;
625 h->picture_structure = h1->picture_structure;
626 h->qscale = h1->qscale;
627 h->droppable = h1->droppable;
628 h->low_delay = h1->low_delay;
630 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
631 ff_h264_unref_picture(h, &h->DPB[i]);
632 if (h1->DPB && h1->DPB[i].f.buf[0] &&
633 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
637 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
638 ff_h264_unref_picture(h, &h->cur_pic);
639 if (h1->cur_pic.f.buf[0] && (ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
642 h->workaround_bugs = h1->workaround_bugs;
643 h->low_delay = h1->low_delay;
644 h->droppable = h1->droppable;
646 // extradata/NAL handling
647 h->is_avc = h1->is_avc;
650 if ((ret = copy_parameter_set((void **)h->sps_buffers,
651 (void **)h1->sps_buffers,
652 MAX_SPS_COUNT, sizeof(SPS))) < 0)
655 if ((ret = copy_parameter_set((void **)h->pps_buffers,
656 (void **)h1->pps_buffers,
657 MAX_PPS_COUNT, sizeof(PPS))) < 0)
661 // Dequantization matrices
662 // FIXME these are big - can they be only copied when PPS changes?
663 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
665 for (i = 0; i < 6; i++)
666 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
667 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
669 for (i = 0; i < 6; i++)
670 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
671 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
673 h->dequant_coeff_pps = h1->dequant_coeff_pps;
676 copy_fields(h, h1, poc_lsb, redundant_pic_count);
679 copy_fields(h, h1, short_ref, cabac_init_idc);
681 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
682 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
683 copy_picture_range(h->delayed_pic, h1->delayed_pic,
684 MAX_DELAYED_PIC_COUNT + 2, h, h1);
686 h->frame_recovered = h1->frame_recovered;
688 if (context_reinitialized)
689 ff_h264_set_parameter_from_sps(h);
695 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
696 h->prev_poc_msb = h->poc_msb;
697 h->prev_poc_lsb = h->poc_lsb;
699 h->prev_frame_num_offset = h->frame_num_offset;
700 h->prev_frame_num = h->frame_num;
701 h->outputed_poc = h->next_outputed_poc;
703 h->recovery_frame = h1->recovery_frame;
708 static int h264_frame_start(H264Context *h)
712 const int pixel_shift = h->pixel_shift;
714 1<<(h->sps.bit_depth_luma-1),
715 1<<(h->sps.bit_depth_chroma-1),
716 1<<(h->sps.bit_depth_chroma-1),
720 if (!ff_thread_can_start_frame(h->avctx)) {
721 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
725 release_unused_pictures(h, 1);
726 h->cur_pic_ptr = NULL;
728 i = find_unused_picture(h);
730 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
735 pic->reference = h->droppable ? 0 : h->picture_structure;
736 pic->f.coded_picture_number = h->coded_picture_number++;
737 pic->field_picture = h->picture_structure != PICT_FRAME;
740 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
742 * See decode_nal_units().
744 pic->f.key_frame = 0;
747 pic->invalid_gap = 0;
748 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
750 if ((ret = alloc_picture(h, pic)) < 0)
752 if(!h->frame_recovered && !h->avctx->hwaccel &&
753 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
754 avpriv_color_frame(&pic->f, c);
756 h->cur_pic_ptr = pic;
757 ff_h264_unref_picture(h, &h->cur_pic);
758 if (CONFIG_ERROR_RESILIENCE) {
759 ff_h264_set_erpic(&h->er.cur_pic, NULL);
762 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
765 if (CONFIG_ERROR_RESILIENCE) {
766 ff_er_frame_start(&h->er);
767 ff_h264_set_erpic(&h->er.last_pic, NULL);
768 ff_h264_set_erpic(&h->er.next_pic, NULL);
771 assert(h->linesize && h->uvlinesize);
773 for (i = 0; i < 16; i++) {
774 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
775 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
777 for (i = 0; i < 16; i++) {
778 h->block_offset[16 + i] =
779 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
780 h->block_offset[48 + 16 + i] =
781 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
784 /* We mark the current picture as non-reference after allocating it, so
785 * that if we break out due to an error it can be released automatically
786 * in the next ff_mpv_frame_start().
788 h->cur_pic_ptr->reference = 0;
790 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
792 h->next_output_pic = NULL;
794 assert(h->cur_pic_ptr->long_ref == 0);
799 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
800 uint8_t *src_cb, uint8_t *src_cr,
801 int linesize, int uvlinesize,
806 const int pixel_shift = h->pixel_shift;
807 int chroma444 = CHROMA444(h);
808 int chroma422 = CHROMA422(h);
811 src_cb -= uvlinesize;
812 src_cr -= uvlinesize;
814 if (!simple && FRAME_MBAFF(h)) {
817 top_border = h->top_borders[0][h->mb_x];
818 AV_COPY128(top_border, src_y + 15 * linesize);
820 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
821 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
824 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
825 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
826 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
827 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
829 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
830 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
832 } else if (chroma422) {
834 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
835 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
837 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
838 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
842 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
843 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
845 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
846 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
851 } else if (MB_MBAFF(h)) {
857 top_border = h->top_borders[top_idx][h->mb_x];
858 /* There are two lines saved, the line above the top macroblock
859 * of a pair, and the line above the bottom macroblock. */
860 AV_COPY128(top_border, src_y + 16 * linesize);
862 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
864 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
867 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
868 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
869 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
870 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
872 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
873 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
875 } else if (chroma422) {
877 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
878 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
880 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
881 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
885 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
886 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
888 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
889 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
896 * Initialize implicit_weight table.
897 * @param field 0/1 initialize the weight for interlaced MBAFF
898 * -1 initializes the rest
900 static void implicit_weight_table(H264Context *h, int field)
902 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
904 for (i = 0; i < 2; i++) {
905 h->luma_weight_flag[i] = 0;
906 h->chroma_weight_flag[i] = 0;
910 if (h->picture_structure == PICT_FRAME) {
911 cur_poc = h->cur_pic_ptr->poc;
913 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
915 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
916 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
918 h->use_weight_chroma = 0;
922 ref_count0 = h->ref_count[0];
923 ref_count1 = h->ref_count[1];
925 cur_poc = h->cur_pic_ptr->field_poc[field];
927 ref_count0 = 16 + 2 * h->ref_count[0];
928 ref_count1 = 16 + 2 * h->ref_count[1];
932 h->use_weight_chroma = 2;
933 h->luma_log2_weight_denom = 5;
934 h->chroma_log2_weight_denom = 5;
936 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
937 int poc0 = h->ref_list[0][ref0].poc;
938 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
940 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
941 int poc1 = h->ref_list[1][ref1].poc;
942 int td = av_clip(poc1 - poc0, -128, 127);
944 int tb = av_clip(cur_poc - poc0, -128, 127);
945 int tx = (16384 + (FFABS(td) >> 1)) / td;
946 int dist_scale_factor = (tb * tx + 32) >> 8;
947 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
948 w = 64 - dist_scale_factor;
952 h->implicit_weight[ref0][ref1][0] =
953 h->implicit_weight[ref0][ref1][1] = w;
955 h->implicit_weight[ref0][ref1][field] = w;
962 * initialize scan tables
964 static void init_scan_tables(H264Context *h)
967 for (i = 0; i < 16; i++) {
968 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
969 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
970 h->field_scan[i] = TRANSPOSE(field_scan[i]);
973 for (i = 0; i < 64; i++) {
974 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
975 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
976 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
977 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
978 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
981 if (h->sps.transform_bypass) { // FIXME same ugly
982 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
983 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
984 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
985 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
986 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
987 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
989 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
990 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
991 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
992 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
993 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
994 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
999 * Replicate H264 "master" context to thread contexts.
1001 static int clone_slice(H264Context *dst, H264Context *src)
1003 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
1004 dst->cur_pic_ptr = src->cur_pic_ptr;
1005 dst->cur_pic = src->cur_pic;
1006 dst->linesize = src->linesize;
1007 dst->uvlinesize = src->uvlinesize;
1008 dst->first_field = src->first_field;
1010 dst->prev_poc_msb = src->prev_poc_msb;
1011 dst->prev_poc_lsb = src->prev_poc_lsb;
1012 dst->prev_frame_num_offset = src->prev_frame_num_offset;
1013 dst->prev_frame_num = src->prev_frame_num;
1014 dst->short_ref_count = src->short_ref_count;
1016 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1017 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1018 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1020 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1021 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1026 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1028 enum AVPixelFormat pix_fmts[2];
1029 const enum AVPixelFormat *choices = pix_fmts;
1032 pix_fmts[1] = AV_PIX_FMT_NONE;
1034 switch (h->sps.bit_depth_luma) {
1037 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1038 pix_fmts[0] = AV_PIX_FMT_GBRP9;
1040 pix_fmts[0] = AV_PIX_FMT_YUV444P9;
1041 } else if (CHROMA422(h))
1042 pix_fmts[0] = AV_PIX_FMT_YUV422P9;
1044 pix_fmts[0] = AV_PIX_FMT_YUV420P9;
1048 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1049 pix_fmts[0] = AV_PIX_FMT_GBRP10;
1051 pix_fmts[0] = AV_PIX_FMT_YUV444P10;
1052 } else if (CHROMA422(h))
1053 pix_fmts[0] = AV_PIX_FMT_YUV422P10;
1055 pix_fmts[0] = AV_PIX_FMT_YUV420P10;
1059 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1060 pix_fmts[0] = AV_PIX_FMT_GBRP12;
1062 pix_fmts[0] = AV_PIX_FMT_YUV444P12;
1063 } else if (CHROMA422(h))
1064 pix_fmts[0] = AV_PIX_FMT_YUV422P12;
1066 pix_fmts[0] = AV_PIX_FMT_YUV420P12;
1070 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1071 pix_fmts[0] = AV_PIX_FMT_GBRP14;
1073 pix_fmts[0] = AV_PIX_FMT_YUV444P14;
1074 } else if (CHROMA422(h))
1075 pix_fmts[0] = AV_PIX_FMT_YUV422P14;
1077 pix_fmts[0] = AV_PIX_FMT_YUV420P14;
1081 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1082 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1083 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1084 pix_fmts[0] = AV_PIX_FMT_GBRP;
1085 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1086 pix_fmts[0] = AV_PIX_FMT_YUVJ444P;
1088 pix_fmts[0] = AV_PIX_FMT_YUV444P;
1089 } else if (CHROMA422(h)) {
1090 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1091 pix_fmts[0] = AV_PIX_FMT_YUVJ422P;
1093 pix_fmts[0] = AV_PIX_FMT_YUV422P;
1095 if (h->avctx->codec->pix_fmts)
1096 choices = h->avctx->codec->pix_fmts;
1097 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1098 choices = h264_hwaccel_pixfmt_list_jpeg_420;
1100 choices = h264_hwaccel_pixfmt_list_420;
1104 av_log(h->avctx, AV_LOG_ERROR,
1105 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1106 return AVERROR_INVALIDDATA;
1109 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1110 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1112 return ff_thread_get_format(h->avctx, choices);
1115 /* export coded and cropped frame dimensions to AVCodecContext */
1116 static int init_dimensions(H264Context *h)
1118 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1119 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1120 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1121 h->sps.crop_right || h->sps.crop_bottom;
1122 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1123 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1125 /* handle container cropping */
1126 if (!crop_present &&
1127 FFALIGN(h->avctx->width, 16) == h->width &&
1128 FFALIGN(h->avctx->height, 16) == h->height) {
1129 width = h->avctx->width;
1130 height = h->avctx->height;
1133 if (width <= 0 || height <= 0) {
1134 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1136 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1137 return AVERROR_INVALIDDATA;
1139 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1140 h->sps.crop_bottom =
1150 h->avctx->coded_width = h->width;
1151 h->avctx->coded_height = h->height;
1152 h->avctx->width = width;
1153 h->avctx->height = height;
1158 static int h264_slice_header_init(H264Context *h, int reinit)
1160 int nb_slices = (HAVE_THREADS &&
1161 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1162 h->avctx->thread_count : 1;
1165 ff_set_sar(h->avctx, h->sps.sar);
1166 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1167 &h->chroma_x_shift, &h->chroma_y_shift);
1169 if (h->sps.timing_info_present_flag) {
1170 int64_t den = h->sps.time_scale;
1171 if (h->x264_build < 44U)
1173 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1174 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1178 ff_h264_free_tables(h, 0);
1180 h->prev_interlaced_frame = 1;
1182 init_scan_tables(h);
1183 ret = ff_h264_alloc_tables(h);
1185 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1189 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1192 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1194 max_slices = H264_MAX_THREADS;
1195 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1196 " reducing to %d\n", nb_slices, max_slices);
1197 nb_slices = max_slices;
1199 h->slice_context_count = nb_slices;
1201 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1202 ret = ff_h264_context_init(h);
1204 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1208 for (i = 1; i < h->slice_context_count; i++) {
1210 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1212 ret = AVERROR(ENOMEM);
1215 c->avctx = h->avctx;
1216 if (CONFIG_ERROR_RESILIENCE) {
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->qscale = h->qscale;
1234 c->droppable = h->droppable;
1235 c->data_partitioning = h->data_partitioning;
1236 c->low_delay = h->low_delay;
1237 c->mb_width = h->mb_width;
1238 c->mb_height = h->mb_height;
1239 c->mb_stride = h->mb_stride;
1240 c->mb_num = h->mb_num;
1241 c->flags = h->flags;
1242 c->workaround_bugs = h->workaround_bugs;
1243 c->pict_type = h->pict_type;
1245 init_scan_tables(c);
1246 clone_tables(c, 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, 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;
1298 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1299 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1301 first_mb_in_slice = get_ue_golomb_long(&h->gb);
1303 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1304 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1305 ff_h264_field_end(h, 1);
1308 h0->current_slice = 0;
1309 if (!h0->first_field) {
1310 if (h->cur_pic_ptr && !h->droppable) {
1311 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1312 h->picture_structure == PICT_BOTTOM_FIELD);
1314 h->cur_pic_ptr = NULL;
1318 slice_type = get_ue_golomb_31(&h->gb);
1319 if (slice_type > 9) {
1320 av_log(h->avctx, AV_LOG_ERROR,
1321 "slice type %d too large at %d %d\n",
1322 slice_type, h->mb_x, h->mb_y);
1323 return AVERROR_INVALIDDATA;
1325 if (slice_type > 4) {
1327 h->slice_type_fixed = 1;
1329 h->slice_type_fixed = 0;
1331 slice_type = golomb_to_pict_type[slice_type];
1332 h->slice_type = slice_type;
1333 h->slice_type_nos = slice_type & 3;
1335 if (h->nal_unit_type == NAL_IDR_SLICE &&
1336 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1337 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1338 return AVERROR_INVALIDDATA;
1342 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1343 (h->avctx->skip_frame >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1344 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1345 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1346 h->avctx->skip_frame >= AVDISCARD_ALL) {
1347 return SLICE_SKIPED;
1350 // to make a few old functions happy, it's wrong though
1351 h->pict_type = h->slice_type;
1353 pps_id = get_ue_golomb(&h->gb);
1354 if (pps_id >= MAX_PPS_COUNT) {
1355 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1356 return AVERROR_INVALIDDATA;
1358 if (!h0->pps_buffers[pps_id]) {
1359 av_log(h->avctx, AV_LOG_ERROR,
1360 "non-existing PPS %u referenced\n",
1362 return AVERROR_INVALIDDATA;
1364 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1365 av_log(h->avctx, AV_LOG_ERROR,
1366 "PPS change from %d to %d forbidden\n",
1367 h0->au_pps_id, pps_id);
1368 return AVERROR_INVALIDDATA;
1370 h->pps = *h0->pps_buffers[pps_id];
1372 if (!h0->sps_buffers[h->pps.sps_id]) {
1373 av_log(h->avctx, AV_LOG_ERROR,
1374 "non-existing SPS %u referenced\n",
1376 return AVERROR_INVALIDDATA;
1379 if (h->pps.sps_id != h->sps.sps_id ||
1380 h->pps.sps_id != h->current_sps_id ||
1381 h0->sps_buffers[h->pps.sps_id]->new) {
1383 h->sps = *h0->sps_buffers[h->pps.sps_id];
1385 if (h->mb_width != h->sps.mb_width ||
1386 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1387 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1388 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1392 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1393 h->chroma_format_idc != h->sps.chroma_format_idc) {
1394 h->bit_depth_luma = h->sps.bit_depth_luma;
1395 h->chroma_format_idc = h->sps.chroma_format_idc;
1398 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1402 h->avctx->profile = ff_h264_get_profile(&h->sps);
1403 h->avctx->level = h->sps.level_idc;
1404 h->avctx->refs = h->sps.ref_frame_count;
1406 must_reinit = (h->context_initialized &&
1407 ( 16*h->sps.mb_width != h->avctx->coded_width
1408 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1409 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1410 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1411 || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
1412 || h->mb_width != h->sps.mb_width
1413 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1415 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1418 h->mb_width = h->sps.mb_width;
1419 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1420 h->mb_num = h->mb_width * h->mb_height;
1421 h->mb_stride = h->mb_width + 1;
1423 h->b_stride = h->mb_width * 4;
1425 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1427 h->width = 16 * h->mb_width;
1428 h->height = 16 * h->mb_height;
1430 ret = init_dimensions(h);
1434 if (h->sps.video_signal_type_present_flag) {
1435 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1437 if (h->sps.colour_description_present_flag) {
1438 if (h->avctx->colorspace != h->sps.colorspace)
1440 h->avctx->color_primaries = h->sps.color_primaries;
1441 h->avctx->color_trc = h->sps.color_trc;
1442 h->avctx->colorspace = h->sps.colorspace;
1446 if (h->context_initialized &&
1447 (must_reinit || needs_reinit)) {
1449 av_log(h->avctx, AV_LOG_ERROR,
1450 "changing width %d -> %d / height %d -> %d on "
1452 h->width, h->avctx->coded_width,
1453 h->height, h->avctx->coded_height,
1454 h0->current_slice + 1);
1455 return AVERROR_INVALIDDATA;
1458 ff_h264_flush_change(h);
1460 if ((ret = get_pixel_format(h, 1)) < 0)
1462 h->avctx->pix_fmt = ret;
1464 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1465 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1467 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1468 av_log(h->avctx, AV_LOG_ERROR,
1469 "h264_slice_header_init() failed\n");
1473 if (!h->context_initialized) {
1475 av_log(h->avctx, AV_LOG_ERROR,
1476 "Cannot (re-)initialize context during parallel decoding.\n");
1477 return AVERROR_PATCHWELCOME;
1480 if ((ret = get_pixel_format(h, 1)) < 0)
1482 h->avctx->pix_fmt = ret;
1484 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1485 av_log(h->avctx, AV_LOG_ERROR,
1486 "h264_slice_header_init() failed\n");
1491 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1492 h->dequant_coeff_pps = pps_id;
1493 h264_init_dequant_tables(h);
1496 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1499 h->mb_aff_frame = 0;
1500 last_pic_structure = h0->picture_structure;
1501 last_pic_droppable = h0->droppable;
1502 h->droppable = h->nal_ref_idc == 0;
1503 if (h->sps.frame_mbs_only_flag) {
1504 h->picture_structure = PICT_FRAME;
1506 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1507 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1510 field_pic_flag = get_bits1(&h->gb);
1511 if (field_pic_flag) {
1512 bottom_field_flag = get_bits1(&h->gb);
1513 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1515 h->picture_structure = PICT_FRAME;
1516 h->mb_aff_frame = h->sps.mb_aff;
1519 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1521 if (h0->current_slice != 0) {
1522 if (last_pic_structure != h->picture_structure ||
1523 last_pic_droppable != h->droppable) {
1524 av_log(h->avctx, AV_LOG_ERROR,
1525 "Changing field mode (%d -> %d) between slices is not allowed\n",
1526 last_pic_structure, h->picture_structure);
1527 h->picture_structure = last_pic_structure;
1528 h->droppable = last_pic_droppable;
1529 return AVERROR_INVALIDDATA;
1530 } else if (!h0->cur_pic_ptr) {
1531 av_log(h->avctx, AV_LOG_ERROR,
1532 "unset cur_pic_ptr on slice %d\n",
1533 h0->current_slice + 1);
1534 return AVERROR_INVALIDDATA;
1537 /* Shorten frame num gaps so we don't have to allocate reference
1538 * frames just to throw them away */
1539 if (h->frame_num != h->prev_frame_num) {
1540 int unwrap_prev_frame_num = h->prev_frame_num;
1541 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1543 if (unwrap_prev_frame_num > h->frame_num)
1544 unwrap_prev_frame_num -= max_frame_num;
1546 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1547 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1548 if (unwrap_prev_frame_num < 0)
1549 unwrap_prev_frame_num += max_frame_num;
1551 h->prev_frame_num = unwrap_prev_frame_num;
1555 /* See if we have a decoded first field looking for a pair...
1556 * Here, we're using that to see if we should mark previously
1557 * decode frames as "finished".
1558 * We have to do that before the "dummy" in-between frame allocation,
1559 * since that can modify h->cur_pic_ptr. */
1560 if (h0->first_field) {
1561 assert(h0->cur_pic_ptr);
1562 assert(h0->cur_pic_ptr->f.buf[0]);
1563 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1565 /* Mark old field/frame as completed */
1566 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1567 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1568 last_pic_structure == PICT_BOTTOM_FIELD);
1571 /* figure out if we have a complementary field pair */
1572 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1573 /* Previous field is unmatched. Don't display it, but let it
1574 * remain for reference if marked as such. */
1575 if (last_pic_structure != PICT_FRAME) {
1576 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1577 last_pic_structure == PICT_TOP_FIELD);
1580 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1581 /* This and previous field were reference, but had
1582 * different frame_nums. Consider this field first in
1583 * pair. Throw away previous field except for reference
1585 if (last_pic_structure != PICT_FRAME) {
1586 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1587 last_pic_structure == PICT_TOP_FIELD);
1590 /* Second field in complementary pair */
1591 if (!((last_pic_structure == PICT_TOP_FIELD &&
1592 h->picture_structure == PICT_BOTTOM_FIELD) ||
1593 (last_pic_structure == PICT_BOTTOM_FIELD &&
1594 h->picture_structure == PICT_TOP_FIELD))) {
1595 av_log(h->avctx, AV_LOG_ERROR,
1596 "Invalid field mode combination %d/%d\n",
1597 last_pic_structure, h->picture_structure);
1598 h->picture_structure = last_pic_structure;
1599 h->droppable = last_pic_droppable;
1600 return AVERROR_INVALIDDATA;
1601 } else if (last_pic_droppable != h->droppable) {
1602 avpriv_request_sample(h->avctx,
1603 "Found reference and non-reference fields in the same frame, which");
1604 h->picture_structure = last_pic_structure;
1605 h->droppable = last_pic_droppable;
1606 return AVERROR_PATCHWELCOME;
1612 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1613 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1614 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1615 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1616 h->frame_num, h->prev_frame_num);
1617 if (!h->sps.gaps_in_frame_num_allowed_flag)
1618 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1619 h->last_pocs[i] = INT_MIN;
1620 ret = h264_frame_start(h);
1622 h0->first_field = 0;
1626 h->prev_frame_num++;
1627 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1628 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1629 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1630 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1631 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1632 ret = ff_generate_sliding_window_mmcos(h, 1);
1633 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1635 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1636 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1638 /* Error concealment: If a ref is missing, copy the previous ref
1640 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1641 * many assumptions about there being no actual duplicates.
1642 * FIXME: This does not copy padding for out-of-frame motion
1643 * vectors. Given we are concealing a lost frame, this probably
1644 * is not noticeable by comparison, but it should be fixed. */
1645 if (h->short_ref_count) {
1647 av_image_copy(h->short_ref[0]->f.data,
1648 h->short_ref[0]->f.linesize,
1649 (const uint8_t **)prev->f.data,
1654 h->short_ref[0]->poc = prev->poc + 2;
1656 h->short_ref[0]->frame_num = h->prev_frame_num;
1660 /* See if we have a decoded first field looking for a pair...
1661 * We're using that to see whether to continue decoding in that
1662 * frame, or to allocate a new one. */
1663 if (h0->first_field) {
1664 assert(h0->cur_pic_ptr);
1665 assert(h0->cur_pic_ptr->f.buf[0]);
1666 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1668 /* figure out if we have a complementary field pair */
1669 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1670 /* Previous field is unmatched. Don't display it, but let it
1671 * remain for reference if marked as such. */
1672 h0->cur_pic_ptr = NULL;
1673 h0->first_field = FIELD_PICTURE(h);
1675 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1676 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1677 h0->picture_structure==PICT_BOTTOM_FIELD);
1678 /* This and the previous field had different frame_nums.
1679 * Consider this field first in pair. Throw away previous
1680 * one except for reference purposes. */
1681 h0->first_field = 1;
1682 h0->cur_pic_ptr = NULL;
1684 /* Second field in complementary pair */
1685 h0->first_field = 0;
1689 /* Frame or first field in a potentially complementary pair */
1690 h0->first_field = FIELD_PICTURE(h);
1693 if (!FIELD_PICTURE(h) || h0->first_field) {
1694 if (h264_frame_start(h) < 0) {
1695 h0->first_field = 0;
1696 return AVERROR_INVALIDDATA;
1699 release_unused_pictures(h, 0);
1701 /* Some macroblocks can be accessed before they're available in case
1702 * of lost slices, MBAFF or threading. */
1703 if (FIELD_PICTURE(h)) {
1704 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1705 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1707 memset(h->slice_table, -1,
1708 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1710 h0->last_slice_type = -1;
1712 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1715 /* can't be in alloc_tables because linesize isn't known there.
1716 * FIXME: redo bipred weight to not require extra buffer? */
1717 for (i = 0; i < h->slice_context_count; i++)
1718 if (h->thread_context[i]) {
1719 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1724 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1726 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1727 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1728 first_mb_in_slice >= h->mb_num) {
1729 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1730 return AVERROR_INVALIDDATA;
1732 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1733 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1734 FIELD_OR_MBAFF_PICTURE(h);
1735 if (h->picture_structure == PICT_BOTTOM_FIELD)
1736 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1737 av_assert1(h->mb_y < h->mb_height);
1739 if (h->picture_structure == PICT_FRAME) {
1740 h->curr_pic_num = h->frame_num;
1741 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1743 h->curr_pic_num = 2 * h->frame_num + 1;
1744 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1747 if (h->nal_unit_type == NAL_IDR_SLICE)
1748 get_ue_golomb(&h->gb); /* idr_pic_id */
1750 if (h->sps.poc_type == 0) {
1751 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1753 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1754 h->delta_poc_bottom = get_se_golomb(&h->gb);
1757 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1758 h->delta_poc[0] = get_se_golomb(&h->gb);
1760 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1761 h->delta_poc[1] = get_se_golomb(&h->gb);
1764 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1766 if (h->pps.redundant_pic_cnt_present)
1767 h->redundant_pic_count = get_ue_golomb(&h->gb);
1769 ret = ff_set_ref_count(h);
1773 if (slice_type != AV_PICTURE_TYPE_I &&
1774 (h0->current_slice == 0 ||
1775 slice_type != h0->last_slice_type ||
1776 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
1778 ff_h264_fill_default_ref_list(h);
1781 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1782 ret = ff_h264_decode_ref_pic_list_reordering(h);
1784 h->ref_count[1] = h->ref_count[0] = 0;
1789 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1790 (h->pps.weighted_bipred_idc == 1 &&
1791 h->slice_type_nos == AV_PICTURE_TYPE_B))
1792 ff_pred_weight_table(h);
1793 else if (h->pps.weighted_bipred_idc == 2 &&
1794 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1795 implicit_weight_table(h, -1);
1798 for (i = 0; i < 2; i++) {
1799 h->luma_weight_flag[i] = 0;
1800 h->chroma_weight_flag[i] = 0;
1804 // If frame-mt is enabled, only update mmco tables for the first slice
1805 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1806 // or h->mmco, which will cause ref list mix-ups and decoding errors
1807 // further down the line. This may break decoding if the first slice is
1808 // corrupt, thus we only do this if frame-mt is enabled.
1809 if (h->nal_ref_idc) {
1810 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1811 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1812 h0->current_slice == 0);
1813 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1814 return AVERROR_INVALIDDATA;
1817 if (FRAME_MBAFF(h)) {
1818 ff_h264_fill_mbaff_ref_list(h);
1820 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1821 implicit_weight_table(h, 0);
1822 implicit_weight_table(h, 1);
1826 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1827 ff_h264_direct_dist_scale_factor(h);
1828 ff_h264_direct_ref_list_init(h);
1830 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1831 tmp = get_ue_golomb_31(&h->gb);
1833 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1834 return AVERROR_INVALIDDATA;
1836 h->cabac_init_idc = tmp;
1839 h->last_qscale_diff = 0;
1840 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1841 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1842 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1843 return AVERROR_INVALIDDATA;
1846 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1847 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1848 // FIXME qscale / qp ... stuff
1849 if (h->slice_type == AV_PICTURE_TYPE_SP)
1850 get_bits1(&h->gb); /* sp_for_switch_flag */
1851 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1852 h->slice_type == AV_PICTURE_TYPE_SI)
1853 get_se_golomb(&h->gb); /* slice_qs_delta */
1855 h->deblocking_filter = 1;
1856 h->slice_alpha_c0_offset = 0;
1857 h->slice_beta_offset = 0;
1858 if (h->pps.deblocking_filter_parameters_present) {
1859 tmp = get_ue_golomb_31(&h->gb);
1861 av_log(h->avctx, AV_LOG_ERROR,
1862 "deblocking_filter_idc %u out of range\n", tmp);
1863 return AVERROR_INVALIDDATA;
1865 h->deblocking_filter = tmp;
1866 if (h->deblocking_filter < 2)
1867 h->deblocking_filter ^= 1; // 1<->0
1869 if (h->deblocking_filter) {
1870 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1871 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1872 if (h->slice_alpha_c0_offset > 12 ||
1873 h->slice_alpha_c0_offset < -12 ||
1874 h->slice_beta_offset > 12 ||
1875 h->slice_beta_offset < -12) {
1876 av_log(h->avctx, AV_LOG_ERROR,
1877 "deblocking filter parameters %d %d out of range\n",
1878 h->slice_alpha_c0_offset, h->slice_beta_offset);
1879 return AVERROR_INVALIDDATA;
1884 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1885 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1886 h->nal_unit_type != NAL_IDR_SLICE) ||
1887 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1888 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1889 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1890 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1891 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1892 h->nal_ref_idc == 0))
1893 h->deblocking_filter = 0;
1895 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1896 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1897 /* Cheat slightly for speed:
1898 * Do not bother to deblock across slices. */
1899 h->deblocking_filter = 2;
1901 h0->max_contexts = 1;
1902 if (!h0->single_decode_warning) {
1903 av_log(h->avctx, AV_LOG_INFO,
1904 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1905 "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"
1906 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1907 "but will generate non-standard-compliant output.\n");
1908 h0->single_decode_warning = 1;
1911 av_log(h->avctx, AV_LOG_ERROR,
1912 "Deblocking switched inside frame.\n");
1913 return SLICE_SINGLETHREAD;
1918 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1920 h->pps.chroma_qp_index_offset[0],
1921 h->pps.chroma_qp_index_offset[1]) +
1922 6 * (h->sps.bit_depth_luma - 8);
1924 h0->last_slice_type = slice_type;
1925 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
1926 h->slice_num = ++h0->current_slice;
1929 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
1930 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
1931 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
1932 && h->slice_num >= MAX_SLICES) {
1933 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1934 av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
1937 for (j = 0; j < 2; j++) {
1939 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1940 for (i = 0; i < 16; i++) {
1942 if (j < h->list_count && i < h->ref_count[j] &&
1943 h->ref_list[j][i].f.buf[0]) {
1945 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1946 for (k = 0; k < h->short_ref_count; k++)
1947 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1951 for (k = 0; k < h->long_ref_count; k++)
1952 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1953 id_list[i] = h->short_ref_count + k;
1961 for (i = 0; i < 16; i++)
1962 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1964 ref2frm[18 + 1] = -1;
1965 for (i = 16; i < 48; i++)
1966 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1967 (h->ref_list[j][i].reference & 3);
1970 if (h->ref_count[0]) ff_h264_set_erpic(&h->er.last_pic, &h->ref_list[0][0]);
1971 if (h->ref_count[1]) ff_h264_set_erpic(&h->er.next_pic, &h->ref_list[1][0]);
1973 h->er.ref_count = h->ref_count[0];
1974 h0->au_pps_id = pps_id;
1976 h0->sps_buffers[h->pps.sps_id]->new = 0;
1977 h->current_sps_id = h->pps.sps_id;
1979 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1980 av_log(h->avctx, AV_LOG_DEBUG,
1981 "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",
1983 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1985 av_get_picture_type_char(h->slice_type),
1986 h->slice_type_fixed ? " fix" : "",
1987 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1988 pps_id, h->frame_num,
1989 h->cur_pic_ptr->field_poc[0],
1990 h->cur_pic_ptr->field_poc[1],
1991 h->ref_count[0], h->ref_count[1],
1993 h->deblocking_filter,
1994 h->slice_alpha_c0_offset, h->slice_beta_offset,
1996 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1997 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2003 int ff_h264_get_slice_type(const H264Context *h)
2005 switch (h->slice_type) {
2006 case AV_PICTURE_TYPE_P:
2008 case AV_PICTURE_TYPE_B:
2010 case AV_PICTURE_TYPE_I:
2012 case AV_PICTURE_TYPE_SP:
2014 case AV_PICTURE_TYPE_SI:
2017 return AVERROR_INVALIDDATA;
2021 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2022 int mb_type, int top_xy,
2023 int left_xy[LEFT_MBS],
2025 int left_type[LEFT_MBS],
2026 int mb_xy, int list)
2028 int b_stride = h->b_stride;
2029 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
2030 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
2031 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2032 if (USES_LIST(top_type, list)) {
2033 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2034 const int b8_xy = 4 * top_xy + 2;
2035 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2036 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2037 ref_cache[0 - 1 * 8] =
2038 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2039 ref_cache[2 - 1 * 8] =
2040 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2042 AV_ZERO128(mv_dst - 1 * 8);
2043 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2046 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2047 if (USES_LIST(left_type[LTOP], list)) {
2048 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2049 const int b8_xy = 4 * left_xy[LTOP] + 1;
2050 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2051 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2052 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2053 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2054 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2056 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2057 ref_cache[-1 + 16] =
2058 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2060 AV_ZERO32(mv_dst - 1 + 0);
2061 AV_ZERO32(mv_dst - 1 + 8);
2062 AV_ZERO32(mv_dst - 1 + 16);
2063 AV_ZERO32(mv_dst - 1 + 24);
2066 ref_cache[-1 + 16] =
2067 ref_cache[-1 + 24] = LIST_NOT_USED;
2072 if (!USES_LIST(mb_type, list)) {
2073 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2074 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2075 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2076 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2077 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2082 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2083 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2084 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2085 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2086 AV_WN32A(&ref_cache[0 * 8], ref01);
2087 AV_WN32A(&ref_cache[1 * 8], ref01);
2088 AV_WN32A(&ref_cache[2 * 8], ref23);
2089 AV_WN32A(&ref_cache[3 * 8], ref23);
2093 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
2094 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2095 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2096 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2097 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2103 * @return non zero if the loop filter can be skipped
2105 static int fill_filter_caches(H264Context *h, int mb_type)
2107 const int mb_xy = h->mb_xy;
2108 int top_xy, left_xy[LEFT_MBS];
2109 int top_type, left_type[LEFT_MBS];
2113 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
2115 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2116 * stuff, I can't imagine that these complex rules are worth it. */
2118 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2119 if (FRAME_MBAFF(h)) {
2120 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2121 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2123 if (left_mb_field_flag != curr_mb_field_flag)
2124 left_xy[LTOP] -= h->mb_stride;
2126 if (curr_mb_field_flag)
2127 top_xy += h->mb_stride &
2128 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2129 if (left_mb_field_flag != curr_mb_field_flag)
2130 left_xy[LBOT] += h->mb_stride;
2134 h->top_mb_xy = top_xy;
2135 h->left_mb_xy[LTOP] = left_xy[LTOP];
2136 h->left_mb_xy[LBOT] = left_xy[LBOT];
2138 /* For sufficiently low qp, filtering wouldn't do anything.
2139 * This is a conservative estimate: could also check beta_offset
2140 * and more accurate chroma_qp. */
2141 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2142 int qp = h->cur_pic.qscale_table[mb_xy];
2143 if (qp <= qp_thresh &&
2144 (left_xy[LTOP] < 0 ||
2145 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2147 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2148 if (!FRAME_MBAFF(h))
2150 if ((left_xy[LTOP] < 0 ||
2151 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2152 (top_xy < h->mb_stride ||
2153 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2158 top_type = h->cur_pic.mb_type[top_xy];
2159 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2160 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2161 if (h->deblocking_filter == 2) {
2162 if (h->slice_table[top_xy] != h->slice_num)
2164 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
2165 left_type[LTOP] = left_type[LBOT] = 0;
2167 if (h->slice_table[top_xy] == 0xFFFF)
2169 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2170 left_type[LTOP] = left_type[LBOT] = 0;
2172 h->top_type = top_type;
2173 h->left_type[LTOP] = left_type[LTOP];
2174 h->left_type[LBOT] = left_type[LBOT];
2176 if (IS_INTRA(mb_type))
2179 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2180 top_type, left_type, mb_xy, 0);
2181 if (h->list_count == 2)
2182 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2183 top_type, left_type, mb_xy, 1);
2185 nnz = h->non_zero_count[mb_xy];
2186 nnz_cache = h->non_zero_count_cache;
2187 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2188 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2189 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2190 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2191 h->cbp = h->cbp_table[mb_xy];
2194 nnz = h->non_zero_count[top_xy];
2195 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2198 if (left_type[LTOP]) {
2199 nnz = h->non_zero_count[left_xy[LTOP]];
2200 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2201 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2202 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2203 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2206 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2207 * from what the loop filter needs */
2208 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2209 if (IS_8x8DCT(top_type)) {
2210 nnz_cache[4 + 8 * 0] =
2211 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2212 nnz_cache[6 + 8 * 0] =
2213 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2215 if (IS_8x8DCT(left_type[LTOP])) {
2216 nnz_cache[3 + 8 * 1] =
2217 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2219 if (IS_8x8DCT(left_type[LBOT])) {
2220 nnz_cache[3 + 8 * 3] =
2221 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2224 if (IS_8x8DCT(mb_type)) {
2225 nnz_cache[scan8[0]] =
2226 nnz_cache[scan8[1]] =
2227 nnz_cache[scan8[2]] =
2228 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2230 nnz_cache[scan8[0 + 4]] =
2231 nnz_cache[scan8[1 + 4]] =
2232 nnz_cache[scan8[2 + 4]] =
2233 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2235 nnz_cache[scan8[0 + 8]] =
2236 nnz_cache[scan8[1 + 8]] =
2237 nnz_cache[scan8[2 + 8]] =
2238 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2240 nnz_cache[scan8[0 + 12]] =
2241 nnz_cache[scan8[1 + 12]] =
2242 nnz_cache[scan8[2 + 12]] =
2243 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2250 static void loop_filter(H264Context *h, int start_x, int end_x)
2252 uint8_t *dest_y, *dest_cb, *dest_cr;
2253 int linesize, uvlinesize, mb_x, mb_y;
2254 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2255 const int old_slice_type = h->slice_type;
2256 const int pixel_shift = h->pixel_shift;
2257 const int block_h = 16 >> h->chroma_y_shift;
2259 if (h->deblocking_filter) {
2260 for (mb_x = start_x; mb_x < end_x; mb_x++)
2261 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2263 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2264 h->slice_num = h->slice_table[mb_xy];
2265 mb_type = h->cur_pic.mb_type[mb_xy];
2266 h->list_count = h->list_counts[mb_xy];
2270 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2274 dest_y = h->cur_pic.f.data[0] +
2275 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2276 dest_cb = h->cur_pic.f.data[1] +
2277 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2278 mb_y * h->uvlinesize * block_h;
2279 dest_cr = h->cur_pic.f.data[2] +
2280 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2281 mb_y * h->uvlinesize * block_h;
2282 // FIXME simplify above
2285 linesize = h->mb_linesize = h->linesize * 2;
2286 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2287 if (mb_y & 1) { // FIXME move out of this function?
2288 dest_y -= h->linesize * 15;
2289 dest_cb -= h->uvlinesize * (block_h - 1);
2290 dest_cr -= h->uvlinesize * (block_h - 1);
2293 linesize = h->mb_linesize = h->linesize;
2294 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2296 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2298 if (fill_filter_caches(h, mb_type))
2300 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2301 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2303 if (FRAME_MBAFF(h)) {
2304 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2305 linesize, uvlinesize);
2307 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2308 dest_cr, linesize, uvlinesize);
2312 h->slice_type = old_slice_type;
2314 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2315 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2316 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2319 static void predict_field_decoding_flag(H264Context *h)
2321 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2322 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2323 h->cur_pic.mb_type[mb_xy - 1] :
2324 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2325 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2326 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2330 * Draw edges and report progress for the last MB row.
2332 static void decode_finish_row(H264Context *h)
2334 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2335 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2336 int height = 16 << FRAME_MBAFF(h);
2337 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2339 if (h->deblocking_filter) {
2340 if ((top + height) >= pic_height)
2341 height += deblock_border;
2342 top -= deblock_border;
2345 if (top >= pic_height || (top + height) < 0)
2348 height = FFMIN(height, pic_height - top);
2350 height = top + height;
2354 ff_h264_draw_horiz_band(h, top, height);
2356 if (h->droppable || h->er.error_occurred)
2359 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2360 h->picture_structure == PICT_BOTTOM_FIELD);
2363 static void er_add_slice(H264Context *h, int startx, int starty,
2364 int endx, int endy, int status)
2366 if (CONFIG_ERROR_RESILIENCE) {
2367 ERContext *er = &h->er;
2369 ff_er_add_slice(er, startx, starty, endx, endy, status);
2373 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2375 H264Context *h = *(void **)arg;
2376 int lf_x_start = h->mb_x;
2378 h->mb_skip_run = -1;
2380 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2382 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2383 avctx->codec_id != AV_CODEC_ID_H264 ||
2384 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2386 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2387 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2389 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2390 prev_status &= ~ VP_START;
2391 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2392 h->er.error_occurred = 1;
2398 align_get_bits(&h->gb);
2401 ff_init_cabac_decoder(&h->cabac,
2402 h->gb.buffer + get_bits_count(&h->gb) / 8,
2403 (get_bits_left(&h->gb) + 7) / 8);
2405 ff_h264_init_cabac_states(h);
2409 int ret = ff_h264_decode_mb_cabac(h);
2411 // STOP_TIMER("decode_mb_cabac")
2414 ff_h264_hl_decode_mb(h);
2416 // FIXME optimal? or let mb_decode decode 16x32 ?
2417 if (ret >= 0 && FRAME_MBAFF(h)) {
2420 ret = ff_h264_decode_mb_cabac(h);
2423 ff_h264_hl_decode_mb(h);
2426 eos = get_cabac_terminate(&h->cabac);
2428 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2429 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2430 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2431 h->mb_y, ER_MB_END);
2432 if (h->mb_x >= lf_x_start)
2433 loop_filter(h, lf_x_start, h->mb_x + 1);
2436 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
2437 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", h->cabac.bytestream_end - h->cabac.bytestream);
2438 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
2439 av_log(h->avctx, AV_LOG_ERROR,
2440 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2442 h->cabac.bytestream_end - h->cabac.bytestream);
2443 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2444 h->mb_y, ER_MB_ERROR);
2445 return AVERROR_INVALIDDATA;
2448 if (++h->mb_x >= h->mb_width) {
2449 loop_filter(h, lf_x_start, h->mb_x);
2450 h->mb_x = lf_x_start = 0;
2451 decode_finish_row(h);
2453 if (FIELD_OR_MBAFF_PICTURE(h)) {
2455 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2456 predict_field_decoding_flag(h);
2460 if (eos || h->mb_y >= h->mb_height) {
2461 tprintf(h->avctx, "slice end %d %d\n",
2462 get_bits_count(&h->gb), h->gb.size_in_bits);
2463 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2464 h->mb_y, ER_MB_END);
2465 if (h->mb_x > lf_x_start)
2466 loop_filter(h, lf_x_start, h->mb_x);
2472 int ret = ff_h264_decode_mb_cavlc(h);
2475 ff_h264_hl_decode_mb(h);
2477 // FIXME optimal? or let mb_decode decode 16x32 ?
2478 if (ret >= 0 && FRAME_MBAFF(h)) {
2480 ret = ff_h264_decode_mb_cavlc(h);
2483 ff_h264_hl_decode_mb(h);
2488 av_log(h->avctx, AV_LOG_ERROR,
2489 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2490 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2491 h->mb_y, ER_MB_ERROR);
2495 if (++h->mb_x >= h->mb_width) {
2496 loop_filter(h, lf_x_start, h->mb_x);
2497 h->mb_x = lf_x_start = 0;
2498 decode_finish_row(h);
2500 if (FIELD_OR_MBAFF_PICTURE(h)) {
2502 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2503 predict_field_decoding_flag(h);
2505 if (h->mb_y >= h->mb_height) {
2506 tprintf(h->avctx, "slice end %d %d\n",
2507 get_bits_count(&h->gb), h->gb.size_in_bits);
2509 if ( get_bits_left(&h->gb) == 0
2510 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2511 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2512 h->mb_x - 1, h->mb_y, ER_MB_END);
2516 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2517 h->mb_x, h->mb_y, ER_MB_END);
2519 return AVERROR_INVALIDDATA;
2524 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2525 tprintf(h->avctx, "slice end %d %d\n",
2526 get_bits_count(&h->gb), h->gb.size_in_bits);
2528 if (get_bits_left(&h->gb) == 0) {
2529 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2530 h->mb_x - 1, h->mb_y, ER_MB_END);
2531 if (h->mb_x > lf_x_start)
2532 loop_filter(h, lf_x_start, h->mb_x);
2536 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2537 h->mb_y, ER_MB_ERROR);
2539 return AVERROR_INVALIDDATA;
2547 * Call decode_slice() for each context.
2549 * @param h h264 master context
2550 * @param context_count number of contexts to execute
2552 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2554 AVCodecContext *const avctx = h->avctx;
2558 av_assert0(h->mb_y < h->mb_height);
2560 if (h->avctx->hwaccel ||
2561 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2563 if (context_count == 1) {
2564 return decode_slice(avctx, &h);
2566 av_assert0(context_count > 0);
2567 for (i = 1; i < context_count; i++) {
2568 hx = h->thread_context[i];
2569 if (CONFIG_ERROR_RESILIENCE) {
2570 hx->er.error_count = 0;
2572 hx->x264_build = h->x264_build;
2575 avctx->execute(avctx, decode_slice, h->thread_context,
2576 NULL, context_count, sizeof(void *));
2578 /* pull back stuff from slices to master context */
2579 hx = h->thread_context[context_count - 1];
2582 h->droppable = hx->droppable;
2583 h->picture_structure = hx->picture_structure;
2584 if (CONFIG_ERROR_RESILIENCE) {
2585 for (i = 1; i < context_count; i++)
2586 h->er.error_count += h->thread_context[i]->er.error_count;