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"
35 #include "error_resilience.h"
39 #include "h264chroma.h"
40 #include "h264_mvpred.h"
43 #include "mpegutils.h"
44 #include "rectangle.h"
48 static const uint8_t rem6[QP_MAX_NUM + 1] = {
49 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
50 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
51 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
52 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
56 static const uint8_t div6[QP_MAX_NUM + 1] = {
57 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
58 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
59 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
60 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
64 static const uint8_t field_scan[16+1] = {
65 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
66 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
67 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
68 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
71 static const uint8_t field_scan8x8[64+1] = {
72 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
73 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
74 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
75 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
76 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
77 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
78 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
79 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
80 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
81 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
82 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
83 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
84 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
85 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
86 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
87 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
90 static const uint8_t field_scan8x8_cavlc[64+1] = {
91 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
92 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
93 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
94 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
95 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
96 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
97 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
98 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
99 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
100 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
101 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
102 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
103 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
104 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
105 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
106 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
109 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
110 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
111 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
112 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
113 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
114 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
115 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
116 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
117 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
118 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
119 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
120 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
121 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
122 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
123 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
124 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
125 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
126 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
129 static const uint8_t dequant4_coeff_init[6][3] = {
138 static const uint8_t dequant8_coeff_init_scan[16] = {
139 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
142 static const uint8_t dequant8_coeff_init[6][6] = {
143 { 20, 18, 32, 19, 25, 24 },
144 { 22, 19, 35, 21, 28, 26 },
145 { 26, 23, 42, 24, 33, 31 },
146 { 28, 25, 45, 26, 35, 33 },
147 { 32, 28, 51, 30, 40, 38 },
148 { 36, 32, 58, 34, 46, 43 },
151 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
152 #if CONFIG_H264_DXVA2_HWACCEL
153 AV_PIX_FMT_DXVA2_VLD,
155 #if CONFIG_H264_VAAPI_HWACCEL
156 AV_PIX_FMT_VAAPI_VLD,
158 #if CONFIG_H264_VDA_HWACCEL
162 #if CONFIG_H264_VDPAU_HWACCEL
169 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
170 #if CONFIG_H264_DXVA2_HWACCEL
171 AV_PIX_FMT_DXVA2_VLD,
173 #if CONFIG_H264_VAAPI_HWACCEL
174 AV_PIX_FMT_VAAPI_VLD,
176 #if CONFIG_H264_VDA_HWACCEL
180 #if CONFIG_H264_VDPAU_HWACCEL
188 static void release_unused_pictures(H264Context *h, int remove_current)
192 /* release non reference frames */
193 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
194 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
195 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
196 ff_h264_unref_picture(h, &h->DPB[i]);
201 static int alloc_scratch_buffers(H264Context *h, int linesize)
203 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
205 if (h->bipred_scratchpad)
208 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
209 // edge emu needs blocksize + filter length - 1
210 // (= 21x21 for h264)
211 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
213 if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
214 av_freep(&h->bipred_scratchpad);
215 av_freep(&h->edge_emu_buffer);
216 return AVERROR(ENOMEM);
222 static int init_table_pools(H264Context *h)
224 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
225 const int mb_array_size = h->mb_stride * h->mb_height;
226 const int b4_stride = h->mb_width * 4 + 1;
227 const int b4_array_size = b4_stride * h->mb_height * 4;
229 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
231 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
232 sizeof(uint32_t), av_buffer_allocz);
233 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
234 sizeof(int16_t), av_buffer_allocz);
235 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
237 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
238 !h->ref_index_pool) {
239 av_buffer_pool_uninit(&h->qscale_table_pool);
240 av_buffer_pool_uninit(&h->mb_type_pool);
241 av_buffer_pool_uninit(&h->motion_val_pool);
242 av_buffer_pool_uninit(&h->ref_index_pool);
243 return AVERROR(ENOMEM);
249 static int alloc_picture(H264Context *h, H264Picture *pic)
253 av_assert0(!pic->f.data[0]);
256 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
257 AV_GET_BUFFER_FLAG_REF : 0);
261 h->linesize = pic->f.linesize[0];
262 h->uvlinesize = pic->f.linesize[1];
263 pic->crop = h->sps.crop;
264 pic->crop_top = h->sps.crop_top;
265 pic->crop_left= h->sps.crop_left;
267 if (h->avctx->hwaccel) {
268 const AVHWAccel *hwaccel = h->avctx->hwaccel;
269 av_assert0(!pic->hwaccel_picture_private);
270 if (hwaccel->frame_priv_data_size) {
271 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
272 if (!pic->hwaccel_priv_buf)
273 return AVERROR(ENOMEM);
274 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
277 if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
278 int h_chroma_shift, v_chroma_shift;
279 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
280 &h_chroma_shift, &v_chroma_shift);
282 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
283 memset(pic->f.data[1] + pic->f.linesize[1]*i,
284 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
285 memset(pic->f.data[2] + pic->f.linesize[2]*i,
286 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
290 if (!h->qscale_table_pool) {
291 ret = init_table_pools(h);
296 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
297 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
298 if (!pic->qscale_table_buf || !pic->mb_type_buf)
301 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
302 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
304 for (i = 0; i < 2; i++) {
305 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
306 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
307 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
310 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
311 pic->ref_index[i] = pic->ref_index_buf[i]->data;
316 ff_h264_unref_picture(h, pic);
317 return (ret < 0) ? ret : AVERROR(ENOMEM);
320 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
324 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
329 static int find_unused_picture(H264Context *h)
333 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
334 if (pic_is_unused(h, &h->DPB[i]))
337 if (i == H264_MAX_PICTURE_COUNT)
338 return AVERROR_INVALIDDATA;
340 if (h->DPB[i].needs_realloc) {
341 h->DPB[i].needs_realloc = 0;
342 ff_h264_unref_picture(h, &h->DPB[i]);
349 static void init_dequant8_coeff_table(H264Context *h)
352 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
354 for (i = 0; i < 6; i++) {
355 h->dequant8_coeff[i] = h->dequant8_buffer[i];
356 for (j = 0; j < i; j++)
357 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
358 64 * sizeof(uint8_t))) {
359 h->dequant8_coeff[i] = h->dequant8_buffer[j];
365 for (q = 0; q < max_qp + 1; q++) {
368 for (x = 0; x < 64; x++)
369 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
370 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
371 h->pps.scaling_matrix8[i][x]) << shift;
376 static void init_dequant4_coeff_table(H264Context *h)
379 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
380 for (i = 0; i < 6; i++) {
381 h->dequant4_coeff[i] = h->dequant4_buffer[i];
382 for (j = 0; j < i; j++)
383 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
384 16 * sizeof(uint8_t))) {
385 h->dequant4_coeff[i] = h->dequant4_buffer[j];
391 for (q = 0; q < max_qp + 1; q++) {
392 int shift = div6[q] + 2;
394 for (x = 0; x < 16; x++)
395 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
396 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
397 h->pps.scaling_matrix4[i][x]) << shift;
402 void h264_init_dequant_tables(H264Context *h)
405 init_dequant4_coeff_table(h);
406 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
408 if (h->pps.transform_8x8_mode)
409 init_dequant8_coeff_table(h);
410 if (h->sps.transform_bypass) {
411 for (i = 0; i < 6; i++)
412 for (x = 0; x < 16; x++)
413 h->dequant4_coeff[i][0][x] = 1 << 6;
414 if (h->pps.transform_8x8_mode)
415 for (i = 0; i < 6; i++)
416 for (x = 0; x < 64; x++)
417 h->dequant8_coeff[i][0][x] = 1 << 6;
422 * Mimic alloc_tables(), but for every context thread.
424 static void clone_tables(H264Context *dst, H264Context *src, int i)
426 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
427 dst->non_zero_count = src->non_zero_count;
428 dst->slice_table = src->slice_table;
429 dst->cbp_table = src->cbp_table;
430 dst->mb2b_xy = src->mb2b_xy;
431 dst->mb2br_xy = src->mb2br_xy;
432 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
433 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
434 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
435 dst->direct_table = src->direct_table;
436 dst->list_counts = src->list_counts;
438 dst->cur_pic_ptr = src->cur_pic_ptr;
439 dst->cur_pic = src->cur_pic;
440 dst->bipred_scratchpad = NULL;
441 dst->edge_emu_buffer = NULL;
442 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
443 src->sps.chroma_format_idc);
446 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
447 #undef REBASE_PICTURE
448 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
449 (((pic) && (pic) >= (old_ctx)->DPB && \
450 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
451 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
453 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
454 H264Context *new_base,
455 H264Context *old_base)
459 for (i = 0; i < count; i++) {
460 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
461 IN_RANGE(from[i], old_base->DPB,
462 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
464 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
468 static int copy_parameter_set(void **to, void **from, int count, int size)
472 for (i = 0; i < count; i++) {
473 if (to[i] && !from[i]) {
475 } else if (from[i] && !to[i]) {
476 to[i] = av_malloc(size);
478 return AVERROR(ENOMEM);
482 memcpy(to[i], from[i], size);
488 #define copy_fields(to, from, start_field, end_field) \
489 memcpy(&(to)->start_field, &(from)->start_field, \
490 (char *)&(to)->end_field - (char *)&(to)->start_field)
492 static int h264_slice_header_init(H264Context *h, int reinit);
494 int ff_h264_update_thread_context(AVCodecContext *dst,
495 const AVCodecContext *src)
497 H264Context *h = dst->priv_data, *h1 = src->priv_data;
498 int inited = h->context_initialized, err = 0;
499 int context_reinitialized = 0;
506 (h->width != h1->width ||
507 h->height != h1->height ||
508 h->mb_width != h1->mb_width ||
509 h->mb_height != h1->mb_height ||
510 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
511 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
512 h->sps.colorspace != h1->sps.colorspace)) {
514 /* set bits_per_raw_sample to the previous value. the check for changed
515 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
516 * the current value */
517 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
519 av_freep(&h->bipred_scratchpad);
521 h->width = h1->width;
522 h->height = h1->height;
523 h->mb_height = h1->mb_height;
524 h->mb_width = h1->mb_width;
525 h->mb_num = h1->mb_num;
526 h->mb_stride = h1->mb_stride;
527 h->b_stride = h1->b_stride;
529 if ((ret = copy_parameter_set((void **)h->sps_buffers,
530 (void **)h1->sps_buffers,
531 MAX_SPS_COUNT, sizeof(SPS))) < 0)
534 if ((ret = copy_parameter_set((void **)h->pps_buffers,
535 (void **)h1->pps_buffers,
536 MAX_PPS_COUNT, sizeof(PPS))) < 0)
540 if ((err = h264_slice_header_init(h, 1)) < 0) {
541 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
544 context_reinitialized = 1;
547 h264_set_parameter_from_sps(h);
548 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
549 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
552 /* update linesize on resize for h264. The h264 decoder doesn't
553 * necessarily call ff_MPV_frame_start in the new thread */
554 h->linesize = h1->linesize;
555 h->uvlinesize = h1->uvlinesize;
557 /* copy block_offset since frame_start may not be called */
558 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
561 for (i = 0; i < MAX_SPS_COUNT; i++)
562 av_freep(h->sps_buffers + i);
564 for (i = 0; i < MAX_PPS_COUNT; i++)
565 av_freep(h->pps_buffers + i);
567 av_freep(&h->rbsp_buffer[0]);
568 av_freep(&h->rbsp_buffer[1]);
569 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
570 memcpy(&h->cabac, &h1->cabac,
571 sizeof(H264Context) - offsetof(H264Context, cabac));
572 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
574 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
575 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
577 memset(&h->er, 0, sizeof(h->er));
578 memset(&h->mb, 0, sizeof(h->mb));
579 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
580 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
581 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
585 h->qscale_table_pool = NULL;
586 h->mb_type_pool = NULL;
587 h->ref_index_pool = NULL;
588 h->motion_val_pool = NULL;
589 for (i = 0; i < 2; i++) {
590 h->rbsp_buffer[i] = NULL;
591 h->rbsp_buffer_size[i] = 0;
594 if (h1->context_initialized) {
595 h->context_initialized = 0;
597 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
598 av_frame_unref(&h->cur_pic.f);
599 h->cur_pic.tf.f = &h->cur_pic.f;
601 ret = ff_h264_alloc_tables(h);
603 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
606 ret = ff_h264_context_init(h);
608 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
613 h->bipred_scratchpad = NULL;
614 h->edge_emu_buffer = NULL;
616 h->thread_context[0] = h;
617 h->context_initialized = h1->context_initialized;
620 h->avctx->coded_height = h1->avctx->coded_height;
621 h->avctx->coded_width = h1->avctx->coded_width;
622 h->avctx->width = h1->avctx->width;
623 h->avctx->height = h1->avctx->height;
624 h->coded_picture_number = h1->coded_picture_number;
625 h->first_field = h1->first_field;
626 h->picture_structure = h1->picture_structure;
627 h->qscale = h1->qscale;
628 h->droppable = h1->droppable;
629 h->low_delay = h1->low_delay;
631 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
632 ff_h264_unref_picture(h, &h->DPB[i]);
633 if (h1->DPB && h1->DPB[i].f.buf[0] &&
634 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
638 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
639 ff_h264_unref_picture(h, &h->cur_pic);
640 if (h1->cur_pic.f.buf[0] && (ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
643 h->workaround_bugs = h1->workaround_bugs;
644 h->low_delay = h1->low_delay;
645 h->droppable = h1->droppable;
647 // extradata/NAL handling
648 h->is_avc = h1->is_avc;
651 if ((ret = copy_parameter_set((void **)h->sps_buffers,
652 (void **)h1->sps_buffers,
653 MAX_SPS_COUNT, sizeof(SPS))) < 0)
656 if ((ret = copy_parameter_set((void **)h->pps_buffers,
657 (void **)h1->pps_buffers,
658 MAX_PPS_COUNT, sizeof(PPS))) < 0)
662 // Dequantization matrices
663 // FIXME these are big - can they be only copied when PPS changes?
664 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
666 for (i = 0; i < 6; i++)
667 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
668 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
670 for (i = 0; i < 6; i++)
671 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
672 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
674 h->dequant_coeff_pps = h1->dequant_coeff_pps;
677 copy_fields(h, h1, poc_lsb, redundant_pic_count);
680 copy_fields(h, h1, short_ref, cabac_init_idc);
682 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
683 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
684 copy_picture_range(h->delayed_pic, h1->delayed_pic,
685 MAX_DELAYED_PIC_COUNT + 2, h, h1);
687 h->frame_recovered = h1->frame_recovered;
689 if (context_reinitialized)
690 ff_h264_set_parameter_from_sps(h);
696 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
697 h->prev_poc_msb = h->poc_msb;
698 h->prev_poc_lsb = h->poc_lsb;
700 h->prev_frame_num_offset = h->frame_num_offset;
701 h->prev_frame_num = h->frame_num;
702 h->outputed_poc = h->next_outputed_poc;
704 h->recovery_frame = h1->recovery_frame;
709 static int h264_frame_start(H264Context *h)
713 const int pixel_shift = h->pixel_shift;
715 1<<(h->sps.bit_depth_luma-1),
716 1<<(h->sps.bit_depth_chroma-1),
717 1<<(h->sps.bit_depth_chroma-1),
721 if (!ff_thread_can_start_frame(h->avctx)) {
722 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
726 release_unused_pictures(h, 1);
727 h->cur_pic_ptr = NULL;
729 i = find_unused_picture(h);
731 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
736 pic->reference = h->droppable ? 0 : h->picture_structure;
737 pic->f.coded_picture_number = h->coded_picture_number++;
738 pic->field_picture = h->picture_structure != PICT_FRAME;
741 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
743 * See decode_nal_units().
745 pic->f.key_frame = 0;
748 pic->invalid_gap = 0;
749 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
751 if ((ret = alloc_picture(h, pic)) < 0)
753 if(!h->frame_recovered && !h->avctx->hwaccel &&
754 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
755 avpriv_color_frame(&pic->f, c);
757 h->cur_pic_ptr = pic;
758 ff_h264_unref_picture(h, &h->cur_pic);
759 if (CONFIG_ERROR_RESILIENCE) {
760 ff_h264_set_erpic(&h->er.cur_pic, NULL);
763 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
766 if (CONFIG_ERROR_RESILIENCE) {
767 ff_er_frame_start(&h->er);
768 ff_h264_set_erpic(&h->er.last_pic, NULL);
769 ff_h264_set_erpic(&h->er.next_pic, NULL);
772 assert(h->linesize && h->uvlinesize);
774 for (i = 0; i < 16; i++) {
775 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
776 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
778 for (i = 0; i < 16; i++) {
779 h->block_offset[16 + i] =
780 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
781 h->block_offset[48 + 16 + i] =
782 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
785 /* We mark the current picture as non-reference after allocating it, so
786 * that if we break out due to an error it can be released automatically
787 * in the next ff_MPV_frame_start().
789 h->cur_pic_ptr->reference = 0;
791 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
793 h->next_output_pic = NULL;
795 assert(h->cur_pic_ptr->long_ref == 0);
800 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
801 uint8_t *src_cb, uint8_t *src_cr,
802 int linesize, int uvlinesize,
807 const int pixel_shift = h->pixel_shift;
808 int chroma444 = CHROMA444(h);
809 int chroma422 = CHROMA422(h);
812 src_cb -= uvlinesize;
813 src_cr -= uvlinesize;
815 if (!simple && FRAME_MBAFF(h)) {
818 top_border = h->top_borders[0][h->mb_x];
819 AV_COPY128(top_border, src_y + 15 * linesize);
821 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
822 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
825 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
826 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
827 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
828 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
830 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
831 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
833 } else if (chroma422) {
835 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
836 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
838 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
839 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
843 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
844 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
846 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
847 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
852 } else if (MB_MBAFF(h)) {
858 top_border = h->top_borders[top_idx][h->mb_x];
859 /* There are two lines saved, the line above the top macroblock
860 * of a pair, and the line above the bottom macroblock. */
861 AV_COPY128(top_border, src_y + 16 * linesize);
863 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
865 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
868 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
869 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
870 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
871 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
873 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
874 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
876 } else if (chroma422) {
878 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
879 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
881 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
882 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
886 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
887 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
889 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
890 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
897 * Initialize implicit_weight table.
898 * @param field 0/1 initialize the weight for interlaced MBAFF
899 * -1 initializes the rest
901 static void implicit_weight_table(H264Context *h, int field)
903 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
905 for (i = 0; i < 2; i++) {
906 h->luma_weight_flag[i] = 0;
907 h->chroma_weight_flag[i] = 0;
911 if (h->picture_structure == PICT_FRAME) {
912 cur_poc = h->cur_pic_ptr->poc;
914 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
916 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
917 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
919 h->use_weight_chroma = 0;
923 ref_count0 = h->ref_count[0];
924 ref_count1 = h->ref_count[1];
926 cur_poc = h->cur_pic_ptr->field_poc[field];
928 ref_count0 = 16 + 2 * h->ref_count[0];
929 ref_count1 = 16 + 2 * h->ref_count[1];
933 h->use_weight_chroma = 2;
934 h->luma_log2_weight_denom = 5;
935 h->chroma_log2_weight_denom = 5;
937 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
938 int poc0 = h->ref_list[0][ref0].poc;
939 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
941 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
942 int poc1 = h->ref_list[1][ref1].poc;
943 int td = av_clip(poc1 - poc0, -128, 127);
945 int tb = av_clip(cur_poc - poc0, -128, 127);
946 int tx = (16384 + (FFABS(td) >> 1)) / td;
947 int dist_scale_factor = (tb * tx + 32) >> 8;
948 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
949 w = 64 - dist_scale_factor;
953 h->implicit_weight[ref0][ref1][0] =
954 h->implicit_weight[ref0][ref1][1] = w;
956 h->implicit_weight[ref0][ref1][field] = w;
963 * initialize scan tables
965 static void init_scan_tables(H264Context *h)
968 for (i = 0; i < 16; i++) {
969 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
970 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
971 h->field_scan[i] = TRANSPOSE(field_scan[i]);
974 for (i = 0; i < 64; i++) {
975 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
976 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
977 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
978 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
979 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
982 if (h->sps.transform_bypass) { // FIXME same ugly
983 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
984 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
985 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
986 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
987 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
988 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
990 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
991 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
992 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
993 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
994 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
995 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
1000 * Replicate H264 "master" context to thread contexts.
1002 static int clone_slice(H264Context *dst, H264Context *src)
1004 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
1005 dst->cur_pic_ptr = src->cur_pic_ptr;
1006 dst->cur_pic = src->cur_pic;
1007 dst->linesize = src->linesize;
1008 dst->uvlinesize = src->uvlinesize;
1009 dst->first_field = src->first_field;
1011 dst->prev_poc_msb = src->prev_poc_msb;
1012 dst->prev_poc_lsb = src->prev_poc_lsb;
1013 dst->prev_frame_num_offset = src->prev_frame_num_offset;
1014 dst->prev_frame_num = src->prev_frame_num;
1015 dst->short_ref_count = src->short_ref_count;
1017 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1018 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1019 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1021 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1022 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1027 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1029 switch (h->sps.bit_depth_luma) {
1032 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1033 return AV_PIX_FMT_GBRP9;
1035 return AV_PIX_FMT_YUV444P9;
1036 } else if (CHROMA422(h))
1037 return AV_PIX_FMT_YUV422P9;
1039 return AV_PIX_FMT_YUV420P9;
1043 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1044 return AV_PIX_FMT_GBRP10;
1046 return AV_PIX_FMT_YUV444P10;
1047 } else if (CHROMA422(h))
1048 return AV_PIX_FMT_YUV422P10;
1050 return AV_PIX_FMT_YUV420P10;
1054 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1055 return AV_PIX_FMT_GBRP12;
1057 return AV_PIX_FMT_YUV444P12;
1058 } else if (CHROMA422(h))
1059 return AV_PIX_FMT_YUV422P12;
1061 return AV_PIX_FMT_YUV420P12;
1065 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1066 return AV_PIX_FMT_GBRP14;
1068 return AV_PIX_FMT_YUV444P14;
1069 } else if (CHROMA422(h))
1070 return AV_PIX_FMT_YUV422P14;
1072 return AV_PIX_FMT_YUV420P14;
1076 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1077 av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
1078 return AV_PIX_FMT_GBR24P;
1079 } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) {
1080 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1082 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
1083 : AV_PIX_FMT_YUV444P;
1084 } else if (CHROMA422(h)) {
1085 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
1086 : AV_PIX_FMT_YUV422P;
1089 const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ?
1090 h->avctx->codec->pix_fmts :
1091 h->avctx->color_range == AVCOL_RANGE_JPEG ?
1092 h264_hwaccel_pixfmt_list_jpeg_420 :
1093 h264_hwaccel_pixfmt_list_420;
1095 for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
1096 if (fmt[i] == h->avctx->pix_fmt && !force_callback)
1098 return ff_thread_get_format(h->avctx, fmt);
1102 av_log(h->avctx, AV_LOG_ERROR,
1103 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1104 return AVERROR_INVALIDDATA;
1108 /* export coded and cropped frame dimensions to AVCodecContext */
1109 static int init_dimensions(H264Context *h)
1111 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1112 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1113 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1114 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1116 /* handle container cropping */
1118 FFALIGN(h->avctx->width, 16) == h->width &&
1119 FFALIGN(h->avctx->height, 16) == h->height) {
1120 width = h->avctx->width;
1121 height = h->avctx->height;
1124 if (width <= 0 || height <= 0) {
1125 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1127 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1128 return AVERROR_INVALIDDATA;
1130 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1131 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
1138 h->avctx->coded_width = h->width;
1139 h->avctx->coded_height = h->height;
1140 h->avctx->width = width;
1141 h->avctx->height = height;
1146 static int h264_slice_header_init(H264Context *h, int reinit)
1148 int nb_slices = (HAVE_THREADS &&
1149 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1150 h->avctx->thread_count : 1;
1153 ff_set_sar(h->avctx, h->sps.sar);
1154 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1155 &h->chroma_x_shift, &h->chroma_y_shift);
1157 if (h->sps.timing_info_present_flag) {
1158 int64_t den = h->sps.time_scale;
1159 if (h->x264_build < 44U)
1161 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
1162 h->sps.num_units_in_tick, den, 1 << 30);
1166 ff_h264_free_tables(h, 0);
1168 h->prev_interlaced_frame = 1;
1170 init_scan_tables(h);
1171 ret = ff_h264_alloc_tables(h);
1173 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1177 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1180 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1182 max_slices = H264_MAX_THREADS;
1183 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1184 " reducing to %d\n", nb_slices, max_slices);
1185 nb_slices = max_slices;
1187 h->slice_context_count = nb_slices;
1189 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1190 ret = ff_h264_context_init(h);
1192 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1196 for (i = 1; i < h->slice_context_count; i++) {
1198 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1200 return AVERROR(ENOMEM);
1201 c->avctx = h->avctx;
1202 if (CONFIG_ERROR_RESILIENCE) {
1206 c->h264dsp = h->h264dsp;
1207 c->h264qpel = h->h264qpel;
1208 c->h264chroma = h->h264chroma;
1211 c->pixel_shift = h->pixel_shift;
1212 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1213 c->width = h->width;
1214 c->height = h->height;
1215 c->linesize = h->linesize;
1216 c->uvlinesize = h->uvlinesize;
1217 c->chroma_x_shift = h->chroma_x_shift;
1218 c->chroma_y_shift = h->chroma_y_shift;
1219 c->qscale = h->qscale;
1220 c->droppable = h->droppable;
1221 c->data_partitioning = h->data_partitioning;
1222 c->low_delay = h->low_delay;
1223 c->mb_width = h->mb_width;
1224 c->mb_height = h->mb_height;
1225 c->mb_stride = h->mb_stride;
1226 c->mb_num = h->mb_num;
1227 c->flags = h->flags;
1228 c->workaround_bugs = h->workaround_bugs;
1229 c->pict_type = h->pict_type;
1231 init_scan_tables(c);
1232 clone_tables(c, h, i);
1233 c->context_initialized = 1;
1236 for (i = 0; i < h->slice_context_count; i++)
1237 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1238 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1243 h->context_initialized = 1;
1248 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1251 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1252 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1253 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1260 * Decode a slice header.
1261 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1263 * @param h h264context
1264 * @param h0 h264 master context (differs from 'h' when doing sliced based
1265 * parallel decoding)
1267 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1269 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1271 unsigned int first_mb_in_slice;
1272 unsigned int pps_id;
1274 unsigned int slice_type, tmp, i, j;
1275 int last_pic_structure, last_pic_droppable;
1277 int needs_reinit = 0;
1278 int field_pic_flag, bottom_field_flag;
1280 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1281 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1283 first_mb_in_slice = get_ue_golomb_long(&h->gb);
1285 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1286 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1287 ff_h264_field_end(h, 1);
1290 h0->current_slice = 0;
1291 if (!h0->first_field) {
1292 if (h->cur_pic_ptr && !h->droppable) {
1293 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1294 h->picture_structure == PICT_BOTTOM_FIELD);
1296 h->cur_pic_ptr = NULL;
1300 slice_type = get_ue_golomb_31(&h->gb);
1301 if (slice_type > 9) {
1302 av_log(h->avctx, AV_LOG_ERROR,
1303 "slice type %d too large at %d %d\n",
1304 slice_type, h->mb_x, h->mb_y);
1305 return AVERROR_INVALIDDATA;
1307 if (slice_type > 4) {
1309 h->slice_type_fixed = 1;
1311 h->slice_type_fixed = 0;
1313 slice_type = golomb_to_pict_type[slice_type];
1314 h->slice_type = slice_type;
1315 h->slice_type_nos = slice_type & 3;
1317 if (h->nal_unit_type == NAL_IDR_SLICE &&
1318 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1319 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1320 return AVERROR_INVALIDDATA;
1324 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1325 (h->avctx->skip_frame >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1326 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1327 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1328 h->avctx->skip_frame >= AVDISCARD_ALL) {
1329 return SLICE_SKIPED;
1332 // to make a few old functions happy, it's wrong though
1333 h->pict_type = h->slice_type;
1335 pps_id = get_ue_golomb(&h->gb);
1336 if (pps_id >= MAX_PPS_COUNT) {
1337 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1338 return AVERROR_INVALIDDATA;
1340 if (!h0->pps_buffers[pps_id]) {
1341 av_log(h->avctx, AV_LOG_ERROR,
1342 "non-existing PPS %u referenced\n",
1344 return AVERROR_INVALIDDATA;
1346 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1347 av_log(h->avctx, AV_LOG_ERROR,
1348 "PPS change from %d to %d forbidden\n",
1349 h0->au_pps_id, pps_id);
1350 return AVERROR_INVALIDDATA;
1352 h->pps = *h0->pps_buffers[pps_id];
1354 if (!h0->sps_buffers[h->pps.sps_id]) {
1355 av_log(h->avctx, AV_LOG_ERROR,
1356 "non-existing SPS %u referenced\n",
1358 return AVERROR_INVALIDDATA;
1361 if (h->pps.sps_id != h->sps.sps_id ||
1362 h->pps.sps_id != h->current_sps_id ||
1363 h0->sps_buffers[h->pps.sps_id]->new) {
1365 h->sps = *h0->sps_buffers[h->pps.sps_id];
1367 if (h->mb_width != h->sps.mb_width ||
1368 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1369 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1370 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1374 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1375 h->chroma_format_idc != h->sps.chroma_format_idc) {
1376 h->bit_depth_luma = h->sps.bit_depth_luma;
1377 h->chroma_format_idc = h->sps.chroma_format_idc;
1380 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1384 h->avctx->profile = ff_h264_get_profile(&h->sps);
1385 h->avctx->level = h->sps.level_idc;
1386 h->avctx->refs = h->sps.ref_frame_count;
1388 must_reinit = (h->context_initialized &&
1389 ( 16*h->sps.mb_width != h->avctx->coded_width
1390 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1391 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1392 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1393 || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
1394 || h->mb_width != h->sps.mb_width
1395 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1397 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1400 h->mb_width = h->sps.mb_width;
1401 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1402 h->mb_num = h->mb_width * h->mb_height;
1403 h->mb_stride = h->mb_width + 1;
1405 h->b_stride = h->mb_width * 4;
1407 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1409 h->width = 16 * h->mb_width;
1410 h->height = 16 * h->mb_height;
1412 ret = init_dimensions(h);
1416 if (h->sps.video_signal_type_present_flag) {
1417 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1419 if (h->sps.colour_description_present_flag) {
1420 if (h->avctx->colorspace != h->sps.colorspace)
1422 h->avctx->color_primaries = h->sps.color_primaries;
1423 h->avctx->color_trc = h->sps.color_trc;
1424 h->avctx->colorspace = h->sps.colorspace;
1428 if (h->context_initialized &&
1429 (must_reinit || needs_reinit)) {
1431 av_log(h->avctx, AV_LOG_ERROR,
1432 "changing width %d -> %d / height %d -> %d on "
1434 h->width, h->avctx->coded_width,
1435 h->height, h->avctx->coded_height,
1436 h0->current_slice + 1);
1437 return AVERROR_INVALIDDATA;
1440 ff_h264_flush_change(h);
1442 if ((ret = get_pixel_format(h, 1)) < 0)
1444 h->avctx->pix_fmt = ret;
1446 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1447 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1449 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1450 av_log(h->avctx, AV_LOG_ERROR,
1451 "h264_slice_header_init() failed\n");
1455 if (!h->context_initialized) {
1457 av_log(h->avctx, AV_LOG_ERROR,
1458 "Cannot (re-)initialize context during parallel decoding.\n");
1459 return AVERROR_PATCHWELCOME;
1462 if ((ret = get_pixel_format(h, 1)) < 0)
1464 h->avctx->pix_fmt = ret;
1466 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1467 av_log(h->avctx, AV_LOG_ERROR,
1468 "h264_slice_header_init() failed\n");
1473 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1474 h->dequant_coeff_pps = pps_id;
1475 h264_init_dequant_tables(h);
1478 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1481 h->mb_aff_frame = 0;
1482 last_pic_structure = h0->picture_structure;
1483 last_pic_droppable = h0->droppable;
1484 h->droppable = h->nal_ref_idc == 0;
1485 if (h->sps.frame_mbs_only_flag) {
1486 h->picture_structure = PICT_FRAME;
1488 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1489 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1492 field_pic_flag = get_bits1(&h->gb);
1493 if (field_pic_flag) {
1494 bottom_field_flag = get_bits1(&h->gb);
1495 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1497 h->picture_structure = PICT_FRAME;
1498 h->mb_aff_frame = h->sps.mb_aff;
1501 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1503 if (h0->current_slice != 0) {
1504 if (last_pic_structure != h->picture_structure ||
1505 last_pic_droppable != h->droppable) {
1506 av_log(h->avctx, AV_LOG_ERROR,
1507 "Changing field mode (%d -> %d) between slices is not allowed\n",
1508 last_pic_structure, h->picture_structure);
1509 h->picture_structure = last_pic_structure;
1510 h->droppable = last_pic_droppable;
1511 return AVERROR_INVALIDDATA;
1512 } else if (!h0->cur_pic_ptr) {
1513 av_log(h->avctx, AV_LOG_ERROR,
1514 "unset cur_pic_ptr on slice %d\n",
1515 h0->current_slice + 1);
1516 return AVERROR_INVALIDDATA;
1519 /* Shorten frame num gaps so we don't have to allocate reference
1520 * frames just to throw them away */
1521 if (h->frame_num != h->prev_frame_num) {
1522 int unwrap_prev_frame_num = h->prev_frame_num;
1523 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1525 if (unwrap_prev_frame_num > h->frame_num)
1526 unwrap_prev_frame_num -= max_frame_num;
1528 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1529 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1530 if (unwrap_prev_frame_num < 0)
1531 unwrap_prev_frame_num += max_frame_num;
1533 h->prev_frame_num = unwrap_prev_frame_num;
1537 /* See if we have a decoded first field looking for a pair...
1538 * Here, we're using that to see if we should mark previously
1539 * decode frames as "finished".
1540 * We have to do that before the "dummy" in-between frame allocation,
1541 * since that can modify h->cur_pic_ptr. */
1542 if (h0->first_field) {
1543 assert(h0->cur_pic_ptr);
1544 assert(h0->cur_pic_ptr->f.buf[0]);
1545 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1547 /* Mark old field/frame as completed */
1548 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1549 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1550 last_pic_structure == PICT_BOTTOM_FIELD);
1553 /* figure out if we have a complementary field pair */
1554 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1555 /* Previous field is unmatched. Don't display it, but let it
1556 * remain for reference if marked as such. */
1557 if (last_pic_structure != PICT_FRAME) {
1558 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1559 last_pic_structure == PICT_TOP_FIELD);
1562 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1563 /* This and previous field were reference, but had
1564 * different frame_nums. Consider this field first in
1565 * pair. Throw away previous field except for reference
1567 if (last_pic_structure != PICT_FRAME) {
1568 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1569 last_pic_structure == PICT_TOP_FIELD);
1572 /* Second field in complementary pair */
1573 if (!((last_pic_structure == PICT_TOP_FIELD &&
1574 h->picture_structure == PICT_BOTTOM_FIELD) ||
1575 (last_pic_structure == PICT_BOTTOM_FIELD &&
1576 h->picture_structure == PICT_TOP_FIELD))) {
1577 av_log(h->avctx, AV_LOG_ERROR,
1578 "Invalid field mode combination %d/%d\n",
1579 last_pic_structure, h->picture_structure);
1580 h->picture_structure = last_pic_structure;
1581 h->droppable = last_pic_droppable;
1582 return AVERROR_INVALIDDATA;
1583 } else if (last_pic_droppable != h->droppable) {
1584 avpriv_request_sample(h->avctx,
1585 "Found reference and non-reference fields in the same frame, which");
1586 h->picture_structure = last_pic_structure;
1587 h->droppable = last_pic_droppable;
1588 return AVERROR_PATCHWELCOME;
1594 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1595 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1596 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1597 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1598 h->frame_num, h->prev_frame_num);
1599 if (!h->sps.gaps_in_frame_num_allowed_flag)
1600 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1601 h->last_pocs[i] = INT_MIN;
1602 ret = h264_frame_start(h);
1604 h0->first_field = 0;
1608 h->prev_frame_num++;
1609 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1610 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1611 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1612 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1613 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1614 ret = ff_generate_sliding_window_mmcos(h, 1);
1615 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1617 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1618 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1620 /* Error concealment: If a ref is missing, copy the previous ref
1622 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1623 * many assumptions about there being no actual duplicates.
1624 * FIXME: This does not copy padding for out-of-frame motion
1625 * vectors. Given we are concealing a lost frame, this probably
1626 * is not noticeable by comparison, but it should be fixed. */
1627 if (h->short_ref_count) {
1629 av_image_copy(h->short_ref[0]->f.data,
1630 h->short_ref[0]->f.linesize,
1631 (const uint8_t **)prev->f.data,
1636 h->short_ref[0]->poc = prev->poc + 2;
1638 h->short_ref[0]->frame_num = h->prev_frame_num;
1642 /* See if we have a decoded first field looking for a pair...
1643 * We're using that to see whether to continue decoding in that
1644 * frame, or to allocate a new one. */
1645 if (h0->first_field) {
1646 assert(h0->cur_pic_ptr);
1647 assert(h0->cur_pic_ptr->f.buf[0]);
1648 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1650 /* figure out if we have a complementary field pair */
1651 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1652 /* Previous field is unmatched. Don't display it, but let it
1653 * remain for reference if marked as such. */
1654 h0->cur_pic_ptr = NULL;
1655 h0->first_field = FIELD_PICTURE(h);
1657 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1658 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1659 h0->picture_structure==PICT_BOTTOM_FIELD);
1660 /* This and the previous field had different frame_nums.
1661 * Consider this field first in pair. Throw away previous
1662 * one except for reference purposes. */
1663 h0->first_field = 1;
1664 h0->cur_pic_ptr = NULL;
1666 /* Second field in complementary pair */
1667 h0->first_field = 0;
1671 /* Frame or first field in a potentially complementary pair */
1672 h0->first_field = FIELD_PICTURE(h);
1675 if (!FIELD_PICTURE(h) || h0->first_field) {
1676 if (h264_frame_start(h) < 0) {
1677 h0->first_field = 0;
1678 return AVERROR_INVALIDDATA;
1681 release_unused_pictures(h, 0);
1683 /* Some macroblocks can be accessed before they're available in case
1684 * of lost slices, MBAFF or threading. */
1685 if (FIELD_PICTURE(h)) {
1686 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1687 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1689 memset(h->slice_table, -1,
1690 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1692 h0->last_slice_type = -1;
1694 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1697 /* can't be in alloc_tables because linesize isn't known there.
1698 * FIXME: redo bipred weight to not require extra buffer? */
1699 for (i = 0; i < h->slice_context_count; i++)
1700 if (h->thread_context[i]) {
1701 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1706 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1708 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1709 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1710 first_mb_in_slice >= h->mb_num) {
1711 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1712 return AVERROR_INVALIDDATA;
1714 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1715 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1716 FIELD_OR_MBAFF_PICTURE(h);
1717 if (h->picture_structure == PICT_BOTTOM_FIELD)
1718 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1719 av_assert1(h->mb_y < h->mb_height);
1721 if (h->picture_structure == PICT_FRAME) {
1722 h->curr_pic_num = h->frame_num;
1723 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1725 h->curr_pic_num = 2 * h->frame_num + 1;
1726 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1729 if (h->nal_unit_type == NAL_IDR_SLICE)
1730 get_ue_golomb(&h->gb); /* idr_pic_id */
1732 if (h->sps.poc_type == 0) {
1733 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1735 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1736 h->delta_poc_bottom = get_se_golomb(&h->gb);
1739 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1740 h->delta_poc[0] = get_se_golomb(&h->gb);
1742 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1743 h->delta_poc[1] = get_se_golomb(&h->gb);
1746 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1748 if (h->pps.redundant_pic_cnt_present)
1749 h->redundant_pic_count = get_ue_golomb(&h->gb);
1751 ret = ff_set_ref_count(h);
1755 if (slice_type != AV_PICTURE_TYPE_I &&
1756 (h0->current_slice == 0 ||
1757 slice_type != h0->last_slice_type ||
1758 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
1760 ff_h264_fill_default_ref_list(h);
1763 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1764 ret = ff_h264_decode_ref_pic_list_reordering(h);
1766 h->ref_count[1] = h->ref_count[0] = 0;
1771 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1772 (h->pps.weighted_bipred_idc == 1 &&
1773 h->slice_type_nos == AV_PICTURE_TYPE_B))
1774 ff_pred_weight_table(h);
1775 else if (h->pps.weighted_bipred_idc == 2 &&
1776 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1777 implicit_weight_table(h, -1);
1780 for (i = 0; i < 2; i++) {
1781 h->luma_weight_flag[i] = 0;
1782 h->chroma_weight_flag[i] = 0;
1786 // If frame-mt is enabled, only update mmco tables for the first slice
1787 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1788 // or h->mmco, which will cause ref list mix-ups and decoding errors
1789 // further down the line. This may break decoding if the first slice is
1790 // corrupt, thus we only do this if frame-mt is enabled.
1791 if (h->nal_ref_idc) {
1792 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1793 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1794 h0->current_slice == 0);
1795 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1796 return AVERROR_INVALIDDATA;
1799 if (FRAME_MBAFF(h)) {
1800 ff_h264_fill_mbaff_ref_list(h);
1802 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1803 implicit_weight_table(h, 0);
1804 implicit_weight_table(h, 1);
1808 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1809 ff_h264_direct_dist_scale_factor(h);
1810 ff_h264_direct_ref_list_init(h);
1812 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1813 tmp = get_ue_golomb_31(&h->gb);
1815 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1816 return AVERROR_INVALIDDATA;
1818 h->cabac_init_idc = tmp;
1821 h->last_qscale_diff = 0;
1822 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1823 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1824 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1825 return AVERROR_INVALIDDATA;
1828 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1829 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1830 // FIXME qscale / qp ... stuff
1831 if (h->slice_type == AV_PICTURE_TYPE_SP)
1832 get_bits1(&h->gb); /* sp_for_switch_flag */
1833 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1834 h->slice_type == AV_PICTURE_TYPE_SI)
1835 get_se_golomb(&h->gb); /* slice_qs_delta */
1837 h->deblocking_filter = 1;
1838 h->slice_alpha_c0_offset = 0;
1839 h->slice_beta_offset = 0;
1840 if (h->pps.deblocking_filter_parameters_present) {
1841 tmp = get_ue_golomb_31(&h->gb);
1843 av_log(h->avctx, AV_LOG_ERROR,
1844 "deblocking_filter_idc %u out of range\n", tmp);
1845 return AVERROR_INVALIDDATA;
1847 h->deblocking_filter = tmp;
1848 if (h->deblocking_filter < 2)
1849 h->deblocking_filter ^= 1; // 1<->0
1851 if (h->deblocking_filter) {
1852 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1853 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1854 if (h->slice_alpha_c0_offset > 12 ||
1855 h->slice_alpha_c0_offset < -12 ||
1856 h->slice_beta_offset > 12 ||
1857 h->slice_beta_offset < -12) {
1858 av_log(h->avctx, AV_LOG_ERROR,
1859 "deblocking filter parameters %d %d out of range\n",
1860 h->slice_alpha_c0_offset, h->slice_beta_offset);
1861 return AVERROR_INVALIDDATA;
1866 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1867 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1868 h->nal_unit_type != NAL_IDR_SLICE) ||
1869 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1870 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1871 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1872 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1873 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1874 h->nal_ref_idc == 0))
1875 h->deblocking_filter = 0;
1877 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1878 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1879 /* Cheat slightly for speed:
1880 * Do not bother to deblock across slices. */
1881 h->deblocking_filter = 2;
1883 h0->max_contexts = 1;
1884 if (!h0->single_decode_warning) {
1885 av_log(h->avctx, AV_LOG_INFO,
1886 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1887 "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"
1888 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1889 "but will generate non-standard-compliant output.\n");
1890 h0->single_decode_warning = 1;
1893 av_log(h->avctx, AV_LOG_ERROR,
1894 "Deblocking switched inside frame.\n");
1895 return SLICE_SINGLETHREAD;
1900 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1902 h->pps.chroma_qp_index_offset[0],
1903 h->pps.chroma_qp_index_offset[1]) +
1904 6 * (h->sps.bit_depth_luma - 8);
1906 h0->last_slice_type = slice_type;
1907 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
1908 h->slice_num = ++h0->current_slice;
1911 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
1912 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
1913 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
1914 && h->slice_num >= MAX_SLICES) {
1915 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1916 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);
1919 for (j = 0; j < 2; j++) {
1921 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1922 for (i = 0; i < 16; i++) {
1924 if (j < h->list_count && i < h->ref_count[j] &&
1925 h->ref_list[j][i].f.buf[0]) {
1927 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1928 for (k = 0; k < h->short_ref_count; k++)
1929 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1933 for (k = 0; k < h->long_ref_count; k++)
1934 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1935 id_list[i] = h->short_ref_count + k;
1943 for (i = 0; i < 16; i++)
1944 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1946 ref2frm[18 + 1] = -1;
1947 for (i = 16; i < 48; i++)
1948 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1949 (h->ref_list[j][i].reference & 3);
1952 if (h->ref_count[0]) ff_h264_set_erpic(&h->er.last_pic, &h->ref_list[0][0]);
1953 if (h->ref_count[1]) ff_h264_set_erpic(&h->er.next_pic, &h->ref_list[1][0]);
1955 h->er.ref_count = h->ref_count[0];
1956 h0->au_pps_id = pps_id;
1958 h0->sps_buffers[h->pps.sps_id]->new = 0;
1959 h->current_sps_id = h->pps.sps_id;
1961 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1962 av_log(h->avctx, AV_LOG_DEBUG,
1963 "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",
1965 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1967 av_get_picture_type_char(h->slice_type),
1968 h->slice_type_fixed ? " fix" : "",
1969 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1970 pps_id, h->frame_num,
1971 h->cur_pic_ptr->field_poc[0],
1972 h->cur_pic_ptr->field_poc[1],
1973 h->ref_count[0], h->ref_count[1],
1975 h->deblocking_filter,
1976 h->slice_alpha_c0_offset, h->slice_beta_offset,
1978 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1979 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1985 int ff_h264_get_slice_type(const H264Context *h)
1987 switch (h->slice_type) {
1988 case AV_PICTURE_TYPE_P:
1990 case AV_PICTURE_TYPE_B:
1992 case AV_PICTURE_TYPE_I:
1994 case AV_PICTURE_TYPE_SP:
1996 case AV_PICTURE_TYPE_SI:
1999 return AVERROR_INVALIDDATA;
2003 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2004 int mb_type, int top_xy,
2005 int left_xy[LEFT_MBS],
2007 int left_type[LEFT_MBS],
2008 int mb_xy, int list)
2010 int b_stride = h->b_stride;
2011 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
2012 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
2013 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2014 if (USES_LIST(top_type, list)) {
2015 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2016 const int b8_xy = 4 * top_xy + 2;
2017 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2018 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2019 ref_cache[0 - 1 * 8] =
2020 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2021 ref_cache[2 - 1 * 8] =
2022 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2024 AV_ZERO128(mv_dst - 1 * 8);
2025 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2028 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2029 if (USES_LIST(left_type[LTOP], list)) {
2030 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2031 const int b8_xy = 4 * left_xy[LTOP] + 1;
2032 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2033 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2034 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2035 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2036 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2038 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2039 ref_cache[-1 + 16] =
2040 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2042 AV_ZERO32(mv_dst - 1 + 0);
2043 AV_ZERO32(mv_dst - 1 + 8);
2044 AV_ZERO32(mv_dst - 1 + 16);
2045 AV_ZERO32(mv_dst - 1 + 24);
2048 ref_cache[-1 + 16] =
2049 ref_cache[-1 + 24] = LIST_NOT_USED;
2054 if (!USES_LIST(mb_type, list)) {
2055 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2056 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2057 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2058 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2059 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2064 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2065 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2066 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2067 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2068 AV_WN32A(&ref_cache[0 * 8], ref01);
2069 AV_WN32A(&ref_cache[1 * 8], ref01);
2070 AV_WN32A(&ref_cache[2 * 8], ref23);
2071 AV_WN32A(&ref_cache[3 * 8], ref23);
2075 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
2076 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2077 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2078 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2079 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2085 * @return non zero if the loop filter can be skipped
2087 static int fill_filter_caches(H264Context *h, int mb_type)
2089 const int mb_xy = h->mb_xy;
2090 int top_xy, left_xy[LEFT_MBS];
2091 int top_type, left_type[LEFT_MBS];
2095 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
2097 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2098 * stuff, I can't imagine that these complex rules are worth it. */
2100 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2101 if (FRAME_MBAFF(h)) {
2102 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2103 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2105 if (left_mb_field_flag != curr_mb_field_flag)
2106 left_xy[LTOP] -= h->mb_stride;
2108 if (curr_mb_field_flag)
2109 top_xy += h->mb_stride &
2110 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2111 if (left_mb_field_flag != curr_mb_field_flag)
2112 left_xy[LBOT] += h->mb_stride;
2116 h->top_mb_xy = top_xy;
2117 h->left_mb_xy[LTOP] = left_xy[LTOP];
2118 h->left_mb_xy[LBOT] = left_xy[LBOT];
2120 /* For sufficiently low qp, filtering wouldn't do anything.
2121 * This is a conservative estimate: could also check beta_offset
2122 * and more accurate chroma_qp. */
2123 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2124 int qp = h->cur_pic.qscale_table[mb_xy];
2125 if (qp <= qp_thresh &&
2126 (left_xy[LTOP] < 0 ||
2127 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2129 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2130 if (!FRAME_MBAFF(h))
2132 if ((left_xy[LTOP] < 0 ||
2133 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2134 (top_xy < h->mb_stride ||
2135 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2140 top_type = h->cur_pic.mb_type[top_xy];
2141 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2142 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2143 if (h->deblocking_filter == 2) {
2144 if (h->slice_table[top_xy] != h->slice_num)
2146 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
2147 left_type[LTOP] = left_type[LBOT] = 0;
2149 if (h->slice_table[top_xy] == 0xFFFF)
2151 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2152 left_type[LTOP] = left_type[LBOT] = 0;
2154 h->top_type = top_type;
2155 h->left_type[LTOP] = left_type[LTOP];
2156 h->left_type[LBOT] = left_type[LBOT];
2158 if (IS_INTRA(mb_type))
2161 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2162 top_type, left_type, mb_xy, 0);
2163 if (h->list_count == 2)
2164 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2165 top_type, left_type, mb_xy, 1);
2167 nnz = h->non_zero_count[mb_xy];
2168 nnz_cache = h->non_zero_count_cache;
2169 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2170 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2171 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2172 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2173 h->cbp = h->cbp_table[mb_xy];
2176 nnz = h->non_zero_count[top_xy];
2177 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2180 if (left_type[LTOP]) {
2181 nnz = h->non_zero_count[left_xy[LTOP]];
2182 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2183 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2184 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2185 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2188 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2189 * from what the loop filter needs */
2190 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2191 if (IS_8x8DCT(top_type)) {
2192 nnz_cache[4 + 8 * 0] =
2193 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2194 nnz_cache[6 + 8 * 0] =
2195 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2197 if (IS_8x8DCT(left_type[LTOP])) {
2198 nnz_cache[3 + 8 * 1] =
2199 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2201 if (IS_8x8DCT(left_type[LBOT])) {
2202 nnz_cache[3 + 8 * 3] =
2203 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2206 if (IS_8x8DCT(mb_type)) {
2207 nnz_cache[scan8[0]] =
2208 nnz_cache[scan8[1]] =
2209 nnz_cache[scan8[2]] =
2210 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2212 nnz_cache[scan8[0 + 4]] =
2213 nnz_cache[scan8[1 + 4]] =
2214 nnz_cache[scan8[2 + 4]] =
2215 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2217 nnz_cache[scan8[0 + 8]] =
2218 nnz_cache[scan8[1 + 8]] =
2219 nnz_cache[scan8[2 + 8]] =
2220 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2222 nnz_cache[scan8[0 + 12]] =
2223 nnz_cache[scan8[1 + 12]] =
2224 nnz_cache[scan8[2 + 12]] =
2225 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2232 static void loop_filter(H264Context *h, int start_x, int end_x)
2234 uint8_t *dest_y, *dest_cb, *dest_cr;
2235 int linesize, uvlinesize, mb_x, mb_y;
2236 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2237 const int old_slice_type = h->slice_type;
2238 const int pixel_shift = h->pixel_shift;
2239 const int block_h = 16 >> h->chroma_y_shift;
2241 if (h->deblocking_filter) {
2242 for (mb_x = start_x; mb_x < end_x; mb_x++)
2243 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2245 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2246 h->slice_num = h->slice_table[mb_xy];
2247 mb_type = h->cur_pic.mb_type[mb_xy];
2248 h->list_count = h->list_counts[mb_xy];
2252 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2256 dest_y = h->cur_pic.f.data[0] +
2257 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2258 dest_cb = h->cur_pic.f.data[1] +
2259 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2260 mb_y * h->uvlinesize * block_h;
2261 dest_cr = h->cur_pic.f.data[2] +
2262 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2263 mb_y * h->uvlinesize * block_h;
2264 // FIXME simplify above
2267 linesize = h->mb_linesize = h->linesize * 2;
2268 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2269 if (mb_y & 1) { // FIXME move out of this function?
2270 dest_y -= h->linesize * 15;
2271 dest_cb -= h->uvlinesize * (block_h - 1);
2272 dest_cr -= h->uvlinesize * (block_h - 1);
2275 linesize = h->mb_linesize = h->linesize;
2276 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2278 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2280 if (fill_filter_caches(h, mb_type))
2282 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2283 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2285 if (FRAME_MBAFF(h)) {
2286 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2287 linesize, uvlinesize);
2289 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2290 dest_cr, linesize, uvlinesize);
2294 h->slice_type = old_slice_type;
2296 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2297 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2298 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2301 static void predict_field_decoding_flag(H264Context *h)
2303 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2304 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2305 h->cur_pic.mb_type[mb_xy - 1] :
2306 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2307 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2308 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2312 * Draw edges and report progress for the last MB row.
2314 static void decode_finish_row(H264Context *h)
2316 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2317 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2318 int height = 16 << FRAME_MBAFF(h);
2319 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2321 if (h->deblocking_filter) {
2322 if ((top + height) >= pic_height)
2323 height += deblock_border;
2324 top -= deblock_border;
2327 if (top >= pic_height || (top + height) < 0)
2330 height = FFMIN(height, pic_height - top);
2332 height = top + height;
2336 ff_h264_draw_horiz_band(h, top, height);
2338 if (h->droppable || h->er.error_occurred)
2341 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2342 h->picture_structure == PICT_BOTTOM_FIELD);
2345 static void er_add_slice(H264Context *h, int startx, int starty,
2346 int endx, int endy, int status)
2348 if (CONFIG_ERROR_RESILIENCE) {
2349 ERContext *er = &h->er;
2351 ff_er_add_slice(er, startx, starty, endx, endy, status);
2355 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2357 H264Context *h = *(void **)arg;
2358 int lf_x_start = h->mb_x;
2360 h->mb_skip_run = -1;
2362 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2364 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2365 avctx->codec_id != AV_CODEC_ID_H264 ||
2366 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2368 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2369 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2371 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2372 prev_status &= ~ VP_START;
2373 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2374 h->er.error_occurred = 1;
2380 align_get_bits(&h->gb);
2383 ff_init_cabac_decoder(&h->cabac,
2384 h->gb.buffer + get_bits_count(&h->gb) / 8,
2385 (get_bits_left(&h->gb) + 7) / 8);
2387 ff_h264_init_cabac_states(h);
2391 int ret = ff_h264_decode_mb_cabac(h);
2393 // STOP_TIMER("decode_mb_cabac")
2396 ff_h264_hl_decode_mb(h);
2398 // FIXME optimal? or let mb_decode decode 16x32 ?
2399 if (ret >= 0 && FRAME_MBAFF(h)) {
2402 ret = ff_h264_decode_mb_cabac(h);
2405 ff_h264_hl_decode_mb(h);
2408 eos = get_cabac_terminate(&h->cabac);
2410 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2411 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2412 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2413 h->mb_y, ER_MB_END);
2414 if (h->mb_x >= lf_x_start)
2415 loop_filter(h, lf_x_start, h->mb_x + 1);
2418 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
2419 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", h->cabac.bytestream_end - h->cabac.bytestream);
2420 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
2421 av_log(h->avctx, AV_LOG_ERROR,
2422 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2424 h->cabac.bytestream_end - h->cabac.bytestream);
2425 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2426 h->mb_y, ER_MB_ERROR);
2427 return AVERROR_INVALIDDATA;
2430 if (++h->mb_x >= h->mb_width) {
2431 loop_filter(h, lf_x_start, h->mb_x);
2432 h->mb_x = lf_x_start = 0;
2433 decode_finish_row(h);
2435 if (FIELD_OR_MBAFF_PICTURE(h)) {
2437 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2438 predict_field_decoding_flag(h);
2442 if (eos || h->mb_y >= h->mb_height) {
2443 tprintf(h->avctx, "slice end %d %d\n",
2444 get_bits_count(&h->gb), h->gb.size_in_bits);
2445 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2446 h->mb_y, ER_MB_END);
2447 if (h->mb_x > lf_x_start)
2448 loop_filter(h, lf_x_start, h->mb_x);
2454 int ret = ff_h264_decode_mb_cavlc(h);
2457 ff_h264_hl_decode_mb(h);
2459 // FIXME optimal? or let mb_decode decode 16x32 ?
2460 if (ret >= 0 && FRAME_MBAFF(h)) {
2462 ret = ff_h264_decode_mb_cavlc(h);
2465 ff_h264_hl_decode_mb(h);
2470 av_log(h->avctx, AV_LOG_ERROR,
2471 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2472 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2473 h->mb_y, ER_MB_ERROR);
2477 if (++h->mb_x >= h->mb_width) {
2478 loop_filter(h, lf_x_start, h->mb_x);
2479 h->mb_x = lf_x_start = 0;
2480 decode_finish_row(h);
2482 if (FIELD_OR_MBAFF_PICTURE(h)) {
2484 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2485 predict_field_decoding_flag(h);
2487 if (h->mb_y >= h->mb_height) {
2488 tprintf(h->avctx, "slice end %d %d\n",
2489 get_bits_count(&h->gb), h->gb.size_in_bits);
2491 if ( get_bits_left(&h->gb) == 0
2492 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2493 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2494 h->mb_x - 1, h->mb_y,
2499 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2503 return AVERROR_INVALIDDATA;
2508 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2509 tprintf(h->avctx, "slice end %d %d\n",
2510 get_bits_count(&h->gb), h->gb.size_in_bits);
2512 if (get_bits_left(&h->gb) == 0) {
2513 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2514 h->mb_x - 1, h->mb_y,
2516 if (h->mb_x > lf_x_start)
2517 loop_filter(h, lf_x_start, h->mb_x);
2521 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2522 h->mb_y, ER_MB_ERROR);
2524 return AVERROR_INVALIDDATA;
2532 * Call decode_slice() for each context.
2534 * @param h h264 master context
2535 * @param context_count number of contexts to execute
2537 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2539 AVCodecContext *const avctx = h->avctx;
2543 av_assert0(h->mb_y < h->mb_height);
2545 if (h->avctx->hwaccel ||
2546 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2548 if (context_count == 1) {
2549 return decode_slice(avctx, &h);
2551 av_assert0(context_count > 0);
2552 for (i = 1; i < context_count; i++) {
2553 hx = h->thread_context[i];
2554 if (CONFIG_ERROR_RESILIENCE) {
2555 hx->er.error_count = 0;
2557 hx->x264_build = h->x264_build;
2560 avctx->execute(avctx, decode_slice, h->thread_context,
2561 NULL, context_count, sizeof(void *));
2563 /* pull back stuff from slices to master context */
2564 hx = h->thread_context[context_count - 1];
2567 h->droppable = hx->droppable;
2568 h->picture_structure = hx->picture_structure;
2569 if (CONFIG_ERROR_RESILIENCE) {
2570 for (i = 1; i < context_count; i++)
2571 h->er.error_count += h->thread_context[i]->er.error_count;