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 Libav.
7 * Libav 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 * Libav 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 Libav; 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, 3,
53 static const uint8_t div6[QP_MAX_NUM + 1] = {
54 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
55 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
56 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
59 static const uint8_t field_scan[16] = {
60 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
61 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
62 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
63 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
66 static const uint8_t field_scan8x8[64] = {
67 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
68 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
69 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
70 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
71 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
72 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
73 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
74 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
75 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
76 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
77 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
78 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
79 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
80 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
81 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
82 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
85 static const uint8_t field_scan8x8_cavlc[64] = {
86 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
87 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
88 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
89 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
90 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
91 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
92 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
93 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
94 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
95 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
96 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
97 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
98 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
99 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
100 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
101 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
104 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
105 static const uint8_t zigzag_scan8x8_cavlc[64] = {
106 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
107 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
108 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
109 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
110 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
111 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
112 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
113 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
114 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
115 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
116 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
117 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
118 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
119 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
120 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
121 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
124 static const uint8_t dequant4_coeff_init[6][3] = {
133 static const uint8_t dequant8_coeff_init_scan[16] = {
134 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
137 static const uint8_t dequant8_coeff_init[6][6] = {
138 { 20, 18, 32, 19, 25, 24 },
139 { 22, 19, 35, 21, 28, 26 },
140 { 26, 23, 42, 24, 33, 31 },
141 { 28, 25, 45, 26, 35, 33 },
142 { 32, 28, 51, 30, 40, 38 },
143 { 36, 32, 58, 34, 46, 43 },
146 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
147 #if CONFIG_H264_DXVA2_HWACCEL
148 AV_PIX_FMT_DXVA2_VLD,
150 #if CONFIG_H264_VAAPI_HWACCEL
151 AV_PIX_FMT_VAAPI_VLD,
153 #if CONFIG_H264_VDA_HWACCEL
157 #if CONFIG_H264_VDPAU_HWACCEL
164 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
165 #if CONFIG_H264_DXVA2_HWACCEL
166 AV_PIX_FMT_DXVA2_VLD,
168 #if CONFIG_H264_VAAPI_HWACCEL
169 AV_PIX_FMT_VAAPI_VLD,
171 #if CONFIG_H264_VDA_HWACCEL
175 #if CONFIG_H264_VDPAU_HWACCEL
183 static void release_unused_pictures(H264Context *h, int remove_current)
187 /* release non reference frames */
188 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
189 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
190 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
191 ff_h264_unref_picture(h, &h->DPB[i]);
196 static int alloc_scratch_buffers(H264Context *h, int linesize)
198 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
200 if (h->bipred_scratchpad)
203 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
204 // edge emu needs blocksize + filter length - 1
205 // (= 21x21 for h264)
206 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
208 if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
209 av_freep(&h->bipred_scratchpad);
210 av_freep(&h->edge_emu_buffer);
211 return AVERROR(ENOMEM);
217 static int init_table_pools(H264Context *h)
219 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
220 const int mb_array_size = h->mb_stride * h->mb_height;
221 const int b4_stride = h->mb_width * 4 + 1;
222 const int b4_array_size = b4_stride * h->mb_height * 4;
224 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
226 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
227 sizeof(uint32_t), av_buffer_allocz);
228 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
229 sizeof(int16_t), av_buffer_allocz);
230 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
232 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
233 !h->ref_index_pool) {
234 av_buffer_pool_uninit(&h->qscale_table_pool);
235 av_buffer_pool_uninit(&h->mb_type_pool);
236 av_buffer_pool_uninit(&h->motion_val_pool);
237 av_buffer_pool_uninit(&h->ref_index_pool);
238 return AVERROR(ENOMEM);
244 static int alloc_picture(H264Context *h, H264Picture *pic)
248 av_assert0(!pic->f.data[0]);
251 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
252 AV_GET_BUFFER_FLAG_REF : 0);
256 h->linesize = pic->f.linesize[0];
257 h->uvlinesize = pic->f.linesize[1];
259 if (h->avctx->hwaccel) {
260 const AVHWAccel *hwaccel = h->avctx->hwaccel;
261 av_assert0(!pic->hwaccel_picture_private);
262 if (hwaccel->frame_priv_data_size) {
263 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
264 if (!pic->hwaccel_priv_buf)
265 return AVERROR(ENOMEM);
266 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
270 if (!h->qscale_table_pool) {
271 ret = init_table_pools(h);
276 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
277 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
278 if (!pic->qscale_table_buf || !pic->mb_type_buf)
281 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
282 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
284 for (i = 0; i < 2; i++) {
285 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
286 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
287 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
290 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
291 pic->ref_index[i] = pic->ref_index_buf[i]->data;
296 ff_h264_unref_picture(h, pic);
297 return (ret < 0) ? ret : AVERROR(ENOMEM);
300 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
304 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
309 static int find_unused_picture(H264Context *h)
313 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
314 if (pic_is_unused(h, &h->DPB[i]))
317 if (i == H264_MAX_PICTURE_COUNT)
318 return AVERROR_INVALIDDATA;
320 if (h->DPB[i].needs_realloc) {
321 h->DPB[i].needs_realloc = 0;
322 ff_h264_unref_picture(h, &h->DPB[i]);
329 static void init_dequant8_coeff_table(H264Context *h)
332 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
334 for (i = 0; i < 6; i++) {
335 h->dequant8_coeff[i] = h->dequant8_buffer[i];
336 for (j = 0; j < i; j++)
337 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
338 64 * sizeof(uint8_t))) {
339 h->dequant8_coeff[i] = h->dequant8_buffer[j];
345 for (q = 0; q < max_qp + 1; q++) {
348 for (x = 0; x < 64; x++)
349 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
350 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
351 h->pps.scaling_matrix8[i][x]) << shift;
356 static void init_dequant4_coeff_table(H264Context *h)
359 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
360 for (i = 0; i < 6; i++) {
361 h->dequant4_coeff[i] = h->dequant4_buffer[i];
362 for (j = 0; j < i; j++)
363 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
364 16 * sizeof(uint8_t))) {
365 h->dequant4_coeff[i] = h->dequant4_buffer[j];
371 for (q = 0; q < max_qp + 1; q++) {
372 int shift = div6[q] + 2;
374 for (x = 0; x < 16; x++)
375 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
376 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
377 h->pps.scaling_matrix4[i][x]) << shift;
382 void h264_init_dequant_tables(H264Context *h)
385 init_dequant4_coeff_table(h);
386 if (h->pps.transform_8x8_mode)
387 init_dequant8_coeff_table(h);
388 if (h->sps.transform_bypass) {
389 for (i = 0; i < 6; i++)
390 for (x = 0; x < 16; x++)
391 h->dequant4_coeff[i][0][x] = 1 << 6;
392 if (h->pps.transform_8x8_mode)
393 for (i = 0; i < 6; i++)
394 for (x = 0; x < 64; x++)
395 h->dequant8_coeff[i][0][x] = 1 << 6;
400 * Mimic alloc_tables(), but for every context thread.
402 static void clone_tables(H264Context *dst, H264Context *src, int i)
404 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
405 dst->non_zero_count = src->non_zero_count;
406 dst->slice_table = src->slice_table;
407 dst->cbp_table = src->cbp_table;
408 dst->mb2b_xy = src->mb2b_xy;
409 dst->mb2br_xy = src->mb2br_xy;
410 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
411 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
412 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
413 dst->direct_table = src->direct_table;
414 dst->list_counts = src->list_counts;
416 dst->cur_pic_ptr = src->cur_pic_ptr;
417 dst->cur_pic = src->cur_pic;
418 dst->bipred_scratchpad = NULL;
419 dst->edge_emu_buffer = NULL;
420 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
421 src->sps.chroma_format_idc);
424 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
425 #undef REBASE_PICTURE
426 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
427 ((pic && pic >= old_ctx->DPB && \
428 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
429 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
431 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
432 H264Context *new_base,
433 H264Context *old_base)
437 for (i = 0; i < count; i++) {
438 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
439 IN_RANGE(from[i], old_base->DPB,
440 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
442 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
446 static int copy_parameter_set(void **to, void **from, int count, int size)
450 for (i = 0; i < count; i++) {
451 if (to[i] && !from[i]) {
453 } else if (from[i] && !to[i]) {
454 to[i] = av_malloc(size);
456 return AVERROR(ENOMEM);
460 memcpy(to[i], from[i], size);
466 #define copy_fields(to, from, start_field, end_field) \
467 memcpy(&to->start_field, &from->start_field, \
468 (char *)&to->end_field - (char *)&to->start_field)
470 static int h264_slice_header_init(H264Context *h, int reinit);
472 int ff_h264_update_thread_context(AVCodecContext *dst,
473 const AVCodecContext *src)
475 H264Context *h = dst->priv_data, *h1 = src->priv_data;
476 int inited = h->context_initialized, err = 0;
477 int context_reinitialized = 0;
480 if (dst == src || !h1->context_initialized)
484 (h->width != h1->width ||
485 h->height != h1->height ||
486 h->mb_width != h1->mb_width ||
487 h->mb_height != h1->mb_height ||
488 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
489 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
490 h->sps.colorspace != h1->sps.colorspace)) {
492 /* set bits_per_raw_sample to the previous value. the check for changed
493 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
494 * the current value */
495 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
497 av_freep(&h->bipred_scratchpad);
499 h->width = h1->width;
500 h->height = h1->height;
501 h->mb_height = h1->mb_height;
502 h->mb_width = h1->mb_width;
503 h->mb_num = h1->mb_num;
504 h->mb_stride = h1->mb_stride;
505 h->b_stride = h1->b_stride;
507 if ((err = h264_slice_header_init(h, 1)) < 0) {
508 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
511 context_reinitialized = 1;
513 /* update linesize on resize. The decoder doesn't
514 * necessarily call h264_frame_start in the new thread */
515 h->linesize = h1->linesize;
516 h->uvlinesize = h1->uvlinesize;
518 /* copy block_offset since frame_start may not be called */
519 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
523 for (i = 0; i < MAX_SPS_COUNT; i++)
524 av_freep(h->sps_buffers + i);
526 for (i = 0; i < MAX_PPS_COUNT; i++)
527 av_freep(h->pps_buffers + i);
529 memcpy(h, h1, sizeof(*h1));
530 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
531 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
532 memset(&h->er, 0, sizeof(h->er));
533 memset(&h->mb, 0, sizeof(h->mb));
534 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
535 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
536 h->context_initialized = 0;
538 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
539 av_frame_unref(&h->cur_pic.f);
540 h->cur_pic.tf.f = &h->cur_pic.f;
544 h->qscale_table_pool = NULL;
545 h->mb_type_pool = NULL;
546 h->ref_index_pool = NULL;
547 h->motion_val_pool = NULL;
549 ret = ff_h264_alloc_tables(h);
551 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
554 ret = ff_h264_context_init(h);
556 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
560 for (i = 0; i < 2; i++) {
561 h->rbsp_buffer[i] = NULL;
562 h->rbsp_buffer_size[i] = 0;
564 h->bipred_scratchpad = NULL;
565 h->edge_emu_buffer = NULL;
567 h->thread_context[0] = h;
569 h->context_initialized = 1;
572 h->avctx->coded_height = h1->avctx->coded_height;
573 h->avctx->coded_width = h1->avctx->coded_width;
574 h->avctx->width = h1->avctx->width;
575 h->avctx->height = h1->avctx->height;
576 h->coded_picture_number = h1->coded_picture_number;
577 h->first_field = h1->first_field;
578 h->picture_structure = h1->picture_structure;
579 h->qscale = h1->qscale;
580 h->droppable = h1->droppable;
581 h->low_delay = h1->low_delay;
583 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
584 ff_h264_unref_picture(h, &h->DPB[i]);
585 if (h1->DPB[i].f.buf[0] &&
586 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
590 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
591 ff_h264_unref_picture(h, &h->cur_pic);
592 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
595 h->workaround_bugs = h1->workaround_bugs;
596 h->low_delay = h1->low_delay;
597 h->droppable = h1->droppable;
599 /* frame_start may not be called for the next thread (if it's decoding
600 * a bottom field) so this has to be allocated here */
601 err = alloc_scratch_buffers(h, h1->linesize);
605 // extradata/NAL handling
606 h->is_avc = h1->is_avc;
609 if ((ret = copy_parameter_set((void **)h->sps_buffers,
610 (void **)h1->sps_buffers,
611 MAX_SPS_COUNT, sizeof(SPS))) < 0)
614 if ((ret = copy_parameter_set((void **)h->pps_buffers,
615 (void **)h1->pps_buffers,
616 MAX_PPS_COUNT, sizeof(PPS))) < 0)
620 // Dequantization matrices
621 // FIXME these are big - can they be only copied when PPS changes?
622 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
624 for (i = 0; i < 6; i++)
625 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
626 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
628 for (i = 0; i < 6; i++)
629 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
630 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
632 h->dequant_coeff_pps = h1->dequant_coeff_pps;
635 copy_fields(h, h1, poc_lsb, redundant_pic_count);
638 copy_fields(h, h1, short_ref, cabac_init_idc);
640 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
641 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
642 copy_picture_range(h->delayed_pic, h1->delayed_pic,
643 MAX_DELAYED_PIC_COUNT + 2, h, h1);
645 h->last_slice_type = h1->last_slice_type;
647 if (context_reinitialized)
648 ff_h264_set_parameter_from_sps(h);
654 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
655 h->prev_poc_msb = h->poc_msb;
656 h->prev_poc_lsb = h->poc_lsb;
658 h->prev_frame_num_offset = h->frame_num_offset;
659 h->prev_frame_num = h->frame_num;
660 h->outputed_poc = h->next_outputed_poc;
662 h->recovery_frame = h1->recovery_frame;
663 h->frame_recovered = h1->frame_recovered;
668 static int h264_frame_start(H264Context *h)
672 const int pixel_shift = h->pixel_shift;
674 release_unused_pictures(h, 1);
675 h->cur_pic_ptr = NULL;
677 i = find_unused_picture(h);
679 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
684 pic->reference = h->droppable ? 0 : h->picture_structure;
685 pic->f.coded_picture_number = h->coded_picture_number++;
686 pic->field_picture = h->picture_structure != PICT_FRAME;
688 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
690 * See decode_nal_units().
692 pic->f.key_frame = 0;
696 if ((ret = alloc_picture(h, pic)) < 0)
699 h->cur_pic_ptr = pic;
700 ff_h264_unref_picture(h, &h->cur_pic);
701 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
704 if (CONFIG_ERROR_RESILIENCE)
705 ff_er_frame_start(&h->er);
707 assert(h->linesize && h->uvlinesize);
709 for (i = 0; i < 16; i++) {
710 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
711 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
713 for (i = 0; i < 16; i++) {
714 h->block_offset[16 + i] =
715 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
716 h->block_offset[48 + 16 + i] =
717 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
720 /* can't be in alloc_tables because linesize isn't known there.
721 * FIXME: redo bipred weight to not require extra buffer? */
722 for (i = 0; i < h->slice_context_count; i++)
723 if (h->thread_context[i]) {
724 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
729 /* Some macroblocks can be accessed before they're available in case
730 * of lost slices, MBAFF or threading. */
731 memset(h->slice_table, -1,
732 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
734 /* We mark the current picture as non-reference after allocating it, so
735 * that if we break out due to an error it can be released automatically
736 * in the next ff_mpv_frame_start().
738 h->cur_pic_ptr->reference = 0;
740 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
742 h->next_output_pic = NULL;
744 assert(h->cur_pic_ptr->long_ref == 0);
749 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
750 uint8_t *src_cb, uint8_t *src_cr,
751 int linesize, int uvlinesize,
756 const int pixel_shift = h->pixel_shift;
757 int chroma444 = CHROMA444(h);
758 int chroma422 = CHROMA422(h);
761 src_cb -= uvlinesize;
762 src_cr -= uvlinesize;
764 if (!simple && FRAME_MBAFF(h)) {
767 top_border = h->top_borders[0][h->mb_x];
768 AV_COPY128(top_border, src_y + 15 * linesize);
770 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
771 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
774 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
775 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
776 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
777 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
779 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
780 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
782 } else if (chroma422) {
784 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
785 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
787 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
788 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
792 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
793 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
795 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
796 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
801 } else if (MB_MBAFF(h)) {
807 top_border = h->top_borders[top_idx][h->mb_x];
808 /* There are two lines saved, the line above the top macroblock
809 * of a pair, and the line above the bottom macroblock. */
810 AV_COPY128(top_border, src_y + 16 * linesize);
812 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
814 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
817 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
818 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
819 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
820 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
822 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
823 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
825 } else if (chroma422) {
827 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
828 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
830 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
831 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
835 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
836 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
838 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
839 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
846 * Initialize implicit_weight table.
847 * @param field 0/1 initialize the weight for interlaced MBAFF
848 * -1 initializes the rest
850 static void implicit_weight_table(H264Context *h, int field)
852 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
854 for (i = 0; i < 2; i++) {
855 h->luma_weight_flag[i] = 0;
856 h->chroma_weight_flag[i] = 0;
860 if (h->picture_structure == PICT_FRAME) {
861 cur_poc = h->cur_pic_ptr->poc;
863 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
865 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
866 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
868 h->use_weight_chroma = 0;
872 ref_count0 = h->ref_count[0];
873 ref_count1 = h->ref_count[1];
875 cur_poc = h->cur_pic_ptr->field_poc[field];
877 ref_count0 = 16 + 2 * h->ref_count[0];
878 ref_count1 = 16 + 2 * h->ref_count[1];
882 h->use_weight_chroma = 2;
883 h->luma_log2_weight_denom = 5;
884 h->chroma_log2_weight_denom = 5;
886 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
887 int poc0 = h->ref_list[0][ref0].poc;
888 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
890 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
891 int poc1 = h->ref_list[1][ref1].poc;
892 int td = av_clip(poc1 - poc0, -128, 127);
894 int tb = av_clip(cur_poc - poc0, -128, 127);
895 int tx = (16384 + (FFABS(td) >> 1)) / td;
896 int dist_scale_factor = (tb * tx + 32) >> 8;
897 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
898 w = 64 - dist_scale_factor;
902 h->implicit_weight[ref0][ref1][0] =
903 h->implicit_weight[ref0][ref1][1] = w;
905 h->implicit_weight[ref0][ref1][field] = w;
912 * initialize scan tables
914 static void init_scan_tables(H264Context *h)
917 for (i = 0; i < 16; i++) {
918 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
919 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
920 h->field_scan[i] = TRANSPOSE(field_scan[i]);
923 for (i = 0; i < 64; i++) {
924 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
925 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
926 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
927 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
928 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
931 if (h->sps.transform_bypass) { // FIXME same ugly
932 h->zigzag_scan_q0 = zigzag_scan;
933 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
934 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
935 h->field_scan_q0 = field_scan;
936 h->field_scan8x8_q0 = field_scan8x8;
937 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
939 h->zigzag_scan_q0 = h->zigzag_scan;
940 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
941 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
942 h->field_scan_q0 = h->field_scan;
943 h->field_scan8x8_q0 = h->field_scan8x8;
944 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
949 * Replicate H264 "master" context to thread contexts.
951 static int clone_slice(H264Context *dst, H264Context *src)
953 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
954 dst->cur_pic_ptr = src->cur_pic_ptr;
955 dst->cur_pic = src->cur_pic;
956 dst->linesize = src->linesize;
957 dst->uvlinesize = src->uvlinesize;
958 dst->first_field = src->first_field;
960 dst->prev_poc_msb = src->prev_poc_msb;
961 dst->prev_poc_lsb = src->prev_poc_lsb;
962 dst->prev_frame_num_offset = src->prev_frame_num_offset;
963 dst->prev_frame_num = src->prev_frame_num;
964 dst->short_ref_count = src->short_ref_count;
966 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
967 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
968 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
970 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
971 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
976 static enum AVPixelFormat get_pixel_format(H264Context *h)
978 switch (h->sps.bit_depth_luma) {
981 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
982 return AV_PIX_FMT_GBRP9;
984 return AV_PIX_FMT_YUV444P9;
985 } else if (CHROMA422(h))
986 return AV_PIX_FMT_YUV422P9;
988 return AV_PIX_FMT_YUV420P9;
992 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
993 return AV_PIX_FMT_GBRP10;
995 return AV_PIX_FMT_YUV444P10;
996 } else if (CHROMA422(h))
997 return AV_PIX_FMT_YUV422P10;
999 return AV_PIX_FMT_YUV420P10;
1003 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1004 return AV_PIX_FMT_GBRP;
1006 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
1007 : AV_PIX_FMT_YUV444P;
1008 } else if (CHROMA422(h)) {
1009 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
1010 : AV_PIX_FMT_YUV422P;
1012 return ff_get_format(h->avctx, h->avctx->codec->pix_fmts ?
1013 h->avctx->codec->pix_fmts :
1014 h->avctx->color_range == AVCOL_RANGE_JPEG ?
1015 h264_hwaccel_pixfmt_list_jpeg_420 :
1016 h264_hwaccel_pixfmt_list_420);
1020 av_log(h->avctx, AV_LOG_ERROR,
1021 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1022 return AVERROR_INVALIDDATA;
1026 /* export coded and cropped frame dimensions to AVCodecContext */
1027 static int init_dimensions(H264Context *h)
1029 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1030 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1032 /* handle container cropping */
1034 FFALIGN(h->avctx->width, 16) == h->width &&
1035 FFALIGN(h->avctx->height, 16) == h->height) {
1036 width = h->avctx->width;
1037 height = h->avctx->height;
1040 if (width <= 0 || height <= 0) {
1041 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1043 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1044 return AVERROR_INVALIDDATA;
1046 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1047 h->sps.crop_bottom =
1057 h->avctx->coded_width = h->width;
1058 h->avctx->coded_height = h->height;
1059 h->avctx->width = width;
1060 h->avctx->height = height;
1065 static int h264_slice_header_init(H264Context *h, int reinit)
1067 int nb_slices = (HAVE_THREADS &&
1068 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1069 h->avctx->thread_count : 1;
1072 ff_set_sar(h->avctx, h->sps.sar);
1073 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1074 &h->chroma_x_shift, &h->chroma_y_shift);
1076 if (h->sps.timing_info_present_flag) {
1077 int64_t den = h->sps.time_scale;
1078 if (h->x264_build < 44U)
1080 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
1081 h->sps.num_units_in_tick, den, 1 << 30);
1085 ff_h264_free_tables(h, 0);
1087 h->prev_interlaced_frame = 1;
1089 init_scan_tables(h);
1090 ret = ff_h264_alloc_tables(h);
1092 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1096 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1099 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1101 max_slices = H264_MAX_THREADS;
1102 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1103 " reducing to %d\n", nb_slices, max_slices);
1104 nb_slices = max_slices;
1106 h->slice_context_count = nb_slices;
1108 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1109 ret = ff_h264_context_init(h);
1111 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1115 for (i = 1; i < h->slice_context_count; i++) {
1117 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1119 return AVERROR(ENOMEM);
1120 c->avctx = h->avctx;
1123 c->h264dsp = h->h264dsp;
1124 c->h264qpel = h->h264qpel;
1125 c->h264chroma = h->h264chroma;
1128 c->pixel_shift = h->pixel_shift;
1129 c->width = h->width;
1130 c->height = h->height;
1131 c->linesize = h->linesize;
1132 c->uvlinesize = h->uvlinesize;
1133 c->chroma_x_shift = h->chroma_x_shift;
1134 c->chroma_y_shift = h->chroma_y_shift;
1135 c->qscale = h->qscale;
1136 c->droppable = h->droppable;
1137 c->data_partitioning = h->data_partitioning;
1138 c->low_delay = h->low_delay;
1139 c->mb_width = h->mb_width;
1140 c->mb_height = h->mb_height;
1141 c->mb_stride = h->mb_stride;
1142 c->mb_num = h->mb_num;
1143 c->flags = h->flags;
1144 c->workaround_bugs = h->workaround_bugs;
1145 c->pict_type = h->pict_type;
1147 init_scan_tables(c);
1148 clone_tables(c, h, i);
1149 c->context_initialized = 1;
1152 for (i = 0; i < h->slice_context_count; i++)
1153 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1154 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1159 h->context_initialized = 1;
1165 * Decode a slice header.
1166 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1168 * @param h h264context
1169 * @param h0 h264 master context (differs from 'h' when doing sliced based
1170 * parallel decoding)
1172 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1174 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1176 unsigned int first_mb_in_slice;
1177 unsigned int pps_id;
1179 unsigned int slice_type, tmp, i, j;
1180 int default_ref_list_done = 0;
1181 int last_pic_structure, last_pic_droppable;
1182 int needs_reinit = 0;
1183 int field_pic_flag, bottom_field_flag;
1185 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1186 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1188 first_mb_in_slice = get_ue_golomb(&h->gb);
1190 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1191 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1192 ff_h264_field_end(h, 1);
1195 h0->current_slice = 0;
1196 if (!h0->first_field) {
1197 if (h->cur_pic_ptr && !h->droppable) {
1198 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1199 h->picture_structure == PICT_BOTTOM_FIELD);
1201 h->cur_pic_ptr = NULL;
1205 slice_type = get_ue_golomb_31(&h->gb);
1206 if (slice_type > 9) {
1207 av_log(h->avctx, AV_LOG_ERROR,
1208 "slice type %d too large at %d %d\n",
1209 slice_type, h->mb_x, h->mb_y);
1210 return AVERROR_INVALIDDATA;
1212 if (slice_type > 4) {
1214 h->slice_type_fixed = 1;
1216 h->slice_type_fixed = 0;
1218 slice_type = golomb_to_pict_type[slice_type];
1219 if (slice_type == AV_PICTURE_TYPE_I ||
1220 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
1221 default_ref_list_done = 1;
1223 h->slice_type = slice_type;
1224 h->slice_type_nos = slice_type & 3;
1226 if (h->nal_unit_type == NAL_IDR_SLICE &&
1227 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1228 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1229 return AVERROR_INVALIDDATA;
1232 // to make a few old functions happy, it's wrong though
1233 h->pict_type = h->slice_type;
1235 pps_id = get_ue_golomb(&h->gb);
1236 if (pps_id >= MAX_PPS_COUNT) {
1237 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1238 return AVERROR_INVALIDDATA;
1240 if (!h0->pps_buffers[pps_id]) {
1241 av_log(h->avctx, AV_LOG_ERROR,
1242 "non-existing PPS %u referenced\n",
1244 return AVERROR_INVALIDDATA;
1246 h->pps = *h0->pps_buffers[pps_id];
1248 if (!h0->sps_buffers[h->pps.sps_id]) {
1249 av_log(h->avctx, AV_LOG_ERROR,
1250 "non-existing SPS %u referenced\n",
1252 return AVERROR_INVALIDDATA;
1255 if (h->pps.sps_id != h->sps.sps_id ||
1256 h0->sps_buffers[h->pps.sps_id]->new) {
1257 h0->sps_buffers[h->pps.sps_id]->new = 0;
1259 h->sps = *h0->sps_buffers[h->pps.sps_id];
1261 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1262 h->chroma_format_idc != h->sps.chroma_format_idc) {
1263 h->bit_depth_luma = h->sps.bit_depth_luma;
1264 h->chroma_format_idc = h->sps.chroma_format_idc;
1267 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1271 h->avctx->profile = ff_h264_get_profile(&h->sps);
1272 h->avctx->level = h->sps.level_idc;
1273 h->avctx->refs = h->sps.ref_frame_count;
1275 if (h->mb_width != h->sps.mb_width ||
1276 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1279 h->mb_width = h->sps.mb_width;
1280 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1281 h->mb_num = h->mb_width * h->mb_height;
1282 h->mb_stride = h->mb_width + 1;
1284 h->b_stride = h->mb_width * 4;
1286 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1288 h->width = 16 * h->mb_width;
1289 h->height = 16 * h->mb_height;
1291 ret = init_dimensions(h);
1295 if (h->sps.video_signal_type_present_flag) {
1296 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1298 if (h->sps.colour_description_present_flag) {
1299 if (h->avctx->colorspace != h->sps.colorspace)
1301 h->avctx->color_primaries = h->sps.color_primaries;
1302 h->avctx->color_trc = h->sps.color_trc;
1303 h->avctx->colorspace = h->sps.colorspace;
1307 if (h->context_initialized && needs_reinit) {
1309 av_log(h->avctx, AV_LOG_ERROR,
1310 "changing width %d -> %d / height %d -> %d on "
1312 h->width, h->avctx->coded_width,
1313 h->height, h->avctx->coded_height,
1314 h0->current_slice + 1);
1315 return AVERROR_INVALIDDATA;
1318 ff_h264_flush_change(h);
1320 if ((ret = get_pixel_format(h)) < 0)
1322 h->avctx->pix_fmt = ret;
1324 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1325 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1327 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1328 av_log(h->avctx, AV_LOG_ERROR,
1329 "h264_slice_header_init() failed\n");
1333 if (!h->context_initialized) {
1335 av_log(h->avctx, AV_LOG_ERROR,
1336 "Cannot (re-)initialize context during parallel decoding.\n");
1337 return AVERROR_PATCHWELCOME;
1340 if ((ret = get_pixel_format(h)) < 0)
1342 h->avctx->pix_fmt = ret;
1344 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1345 av_log(h->avctx, AV_LOG_ERROR,
1346 "h264_slice_header_init() failed\n");
1351 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1352 h->dequant_coeff_pps = pps_id;
1353 h264_init_dequant_tables(h);
1356 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1359 h->mb_aff_frame = 0;
1360 last_pic_structure = h0->picture_structure;
1361 last_pic_droppable = h0->droppable;
1362 h->droppable = h->nal_ref_idc == 0;
1363 if (h->sps.frame_mbs_only_flag) {
1364 h->picture_structure = PICT_FRAME;
1366 field_pic_flag = get_bits1(&h->gb);
1367 if (field_pic_flag) {
1368 bottom_field_flag = get_bits1(&h->gb);
1369 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1371 h->picture_structure = PICT_FRAME;
1372 h->mb_aff_frame = h->sps.mb_aff;
1375 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1377 if (h0->current_slice != 0) {
1378 if (last_pic_structure != h->picture_structure ||
1379 last_pic_droppable != h->droppable) {
1380 av_log(h->avctx, AV_LOG_ERROR,
1381 "Changing field mode (%d -> %d) between slices is not allowed\n",
1382 last_pic_structure, h->picture_structure);
1383 h->picture_structure = last_pic_structure;
1384 h->droppable = last_pic_droppable;
1385 return AVERROR_INVALIDDATA;
1386 } else if (!h0->cur_pic_ptr) {
1387 av_log(h->avctx, AV_LOG_ERROR,
1388 "unset cur_pic_ptr on slice %d\n",
1389 h0->current_slice + 1);
1390 return AVERROR_INVALIDDATA;
1393 /* Shorten frame num gaps so we don't have to allocate reference
1394 * frames just to throw them away */
1395 if (h->frame_num != h->prev_frame_num) {
1396 int unwrap_prev_frame_num = h->prev_frame_num;
1397 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1399 if (unwrap_prev_frame_num > h->frame_num)
1400 unwrap_prev_frame_num -= max_frame_num;
1402 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1403 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1404 if (unwrap_prev_frame_num < 0)
1405 unwrap_prev_frame_num += max_frame_num;
1407 h->prev_frame_num = unwrap_prev_frame_num;
1411 /* See if we have a decoded first field looking for a pair...
1412 * Here, we're using that to see if we should mark previously
1413 * decode frames as "finished".
1414 * We have to do that before the "dummy" in-between frame allocation,
1415 * since that can modify s->current_picture_ptr. */
1416 if (h0->first_field) {
1417 assert(h0->cur_pic_ptr);
1418 assert(h0->cur_pic_ptr->f.buf[0]);
1419 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1421 /* figure out if we have a complementary field pair */
1422 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1423 /* Previous field is unmatched. Don't display it, but let it
1424 * remain for reference if marked as such. */
1425 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1426 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1427 last_pic_structure == PICT_TOP_FIELD);
1430 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1431 /* This and previous field were reference, but had
1432 * different frame_nums. Consider this field first in
1433 * pair. Throw away previous field except for reference
1435 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1436 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1437 last_pic_structure == PICT_TOP_FIELD);
1440 /* Second field in complementary pair */
1441 if (!((last_pic_structure == PICT_TOP_FIELD &&
1442 h->picture_structure == PICT_BOTTOM_FIELD) ||
1443 (last_pic_structure == PICT_BOTTOM_FIELD &&
1444 h->picture_structure == PICT_TOP_FIELD))) {
1445 av_log(h->avctx, AV_LOG_ERROR,
1446 "Invalid field mode combination %d/%d\n",
1447 last_pic_structure, h->picture_structure);
1448 h->picture_structure = last_pic_structure;
1449 h->droppable = last_pic_droppable;
1450 return AVERROR_INVALIDDATA;
1451 } else if (last_pic_droppable != h->droppable) {
1452 avpriv_request_sample(h->avctx,
1453 "Found reference and non-reference fields in the same frame, which");
1454 h->picture_structure = last_pic_structure;
1455 h->droppable = last_pic_droppable;
1456 return AVERROR_PATCHWELCOME;
1462 while (h->frame_num != h->prev_frame_num &&
1463 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1464 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1465 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1466 h->frame_num, h->prev_frame_num);
1467 ret = h264_frame_start(h);
1469 h0->first_field = 0;
1473 h->prev_frame_num++;
1474 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1475 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1476 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1477 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1478 ret = ff_generate_sliding_window_mmcos(h, 1);
1479 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1481 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1482 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1484 /* Error concealment: If a ref is missing, copy the previous ref
1486 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1487 * many assumptions about there being no actual duplicates.
1488 * FIXME: This does not copy padding for out-of-frame motion
1489 * vectors. Given we are concealing a lost frame, this probably
1490 * is not noticeable by comparison, but it should be fixed. */
1491 if (h->short_ref_count) {
1493 av_image_copy(h->short_ref[0]->f.data,
1494 h->short_ref[0]->f.linesize,
1495 (const uint8_t **)prev->f.data,
1500 h->short_ref[0]->poc = prev->poc + 2;
1502 h->short_ref[0]->frame_num = h->prev_frame_num;
1506 /* See if we have a decoded first field looking for a pair...
1507 * We're using that to see whether to continue decoding in that
1508 * frame, or to allocate a new one. */
1509 if (h0->first_field) {
1510 assert(h0->cur_pic_ptr);
1511 assert(h0->cur_pic_ptr->f.buf[0]);
1512 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1514 /* figure out if we have a complementary field pair */
1515 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1516 /* Previous field is unmatched. Don't display it, but let it
1517 * remain for reference if marked as such. */
1518 h0->cur_pic_ptr = NULL;
1519 h0->first_field = FIELD_PICTURE(h);
1521 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1522 /* This and the previous field had different frame_nums.
1523 * Consider this field first in pair. Throw away previous
1524 * one except for reference purposes. */
1525 h0->first_field = 1;
1526 h0->cur_pic_ptr = NULL;
1528 /* Second field in complementary pair */
1529 h0->first_field = 0;
1533 /* Frame or first field in a potentially complementary pair */
1534 h0->first_field = FIELD_PICTURE(h);
1537 if (!FIELD_PICTURE(h) || h0->first_field) {
1538 if (h264_frame_start(h) < 0) {
1539 h0->first_field = 0;
1540 return AVERROR_INVALIDDATA;
1543 release_unused_pictures(h, 0);
1546 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1549 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1551 assert(h->mb_num == h->mb_width * h->mb_height);
1552 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1553 first_mb_in_slice >= h->mb_num) {
1554 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1555 return AVERROR_INVALIDDATA;
1557 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1558 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1559 FIELD_OR_MBAFF_PICTURE(h);
1560 if (h->picture_structure == PICT_BOTTOM_FIELD)
1561 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1562 assert(h->mb_y < h->mb_height);
1564 if (h->picture_structure == PICT_FRAME) {
1565 h->curr_pic_num = h->frame_num;
1566 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1568 h->curr_pic_num = 2 * h->frame_num + 1;
1569 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1572 if (h->nal_unit_type == NAL_IDR_SLICE)
1573 get_ue_golomb(&h->gb); /* idr_pic_id */
1575 if (h->sps.poc_type == 0) {
1576 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1578 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1579 h->delta_poc_bottom = get_se_golomb(&h->gb);
1582 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1583 h->delta_poc[0] = get_se_golomb(&h->gb);
1585 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1586 h->delta_poc[1] = get_se_golomb(&h->gb);
1589 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1591 if (h->pps.redundant_pic_cnt_present)
1592 h->redundant_pic_count = get_ue_golomb(&h->gb);
1594 ret = ff_set_ref_count(h);
1598 default_ref_list_done = 0;
1600 if (!default_ref_list_done)
1601 ff_h264_fill_default_ref_list(h);
1603 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1604 ret = ff_h264_decode_ref_pic_list_reordering(h);
1606 h->ref_count[1] = h->ref_count[0] = 0;
1611 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1612 (h->pps.weighted_bipred_idc == 1 &&
1613 h->slice_type_nos == AV_PICTURE_TYPE_B))
1614 ff_pred_weight_table(h);
1615 else if (h->pps.weighted_bipred_idc == 2 &&
1616 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1617 implicit_weight_table(h, -1);
1620 for (i = 0; i < 2; i++) {
1621 h->luma_weight_flag[i] = 0;
1622 h->chroma_weight_flag[i] = 0;
1626 // If frame-mt is enabled, only update mmco tables for the first slice
1627 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1628 // or h->mmco, which will cause ref list mix-ups and decoding errors
1629 // further down the line. This may break decoding if the first slice is
1630 // corrupt, thus we only do this if frame-mt is enabled.
1631 if (h->nal_ref_idc) {
1632 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1633 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1634 h0->current_slice == 0);
1635 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1636 return AVERROR_INVALIDDATA;
1639 if (FRAME_MBAFF(h)) {
1640 ff_h264_fill_mbaff_ref_list(h);
1642 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1643 implicit_weight_table(h, 0);
1644 implicit_weight_table(h, 1);
1648 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1649 ff_h264_direct_dist_scale_factor(h);
1650 ff_h264_direct_ref_list_init(h);
1652 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1653 tmp = get_ue_golomb_31(&h->gb);
1655 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1656 return AVERROR_INVALIDDATA;
1658 h->cabac_init_idc = tmp;
1661 h->last_qscale_diff = 0;
1662 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1663 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1664 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1665 return AVERROR_INVALIDDATA;
1668 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1669 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1670 // FIXME qscale / qp ... stuff
1671 if (h->slice_type == AV_PICTURE_TYPE_SP)
1672 get_bits1(&h->gb); /* sp_for_switch_flag */
1673 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1674 h->slice_type == AV_PICTURE_TYPE_SI)
1675 get_se_golomb(&h->gb); /* slice_qs_delta */
1677 h->deblocking_filter = 1;
1678 h->slice_alpha_c0_offset = 0;
1679 h->slice_beta_offset = 0;
1680 if (h->pps.deblocking_filter_parameters_present) {
1681 tmp = get_ue_golomb_31(&h->gb);
1683 av_log(h->avctx, AV_LOG_ERROR,
1684 "deblocking_filter_idc %u out of range\n", tmp);
1685 return AVERROR_INVALIDDATA;
1687 h->deblocking_filter = tmp;
1688 if (h->deblocking_filter < 2)
1689 h->deblocking_filter ^= 1; // 1<->0
1691 if (h->deblocking_filter) {
1692 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1693 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1694 if (h->slice_alpha_c0_offset > 12 ||
1695 h->slice_alpha_c0_offset < -12 ||
1696 h->slice_beta_offset > 12 ||
1697 h->slice_beta_offset < -12) {
1698 av_log(h->avctx, AV_LOG_ERROR,
1699 "deblocking filter parameters %d %d out of range\n",
1700 h->slice_alpha_c0_offset, h->slice_beta_offset);
1701 return AVERROR_INVALIDDATA;
1706 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1707 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1708 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1709 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1710 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1711 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1712 h->nal_ref_idc == 0))
1713 h->deblocking_filter = 0;
1715 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1716 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1717 /* Cheat slightly for speed:
1718 * Do not bother to deblock across slices. */
1719 h->deblocking_filter = 2;
1721 h0->max_contexts = 1;
1722 if (!h0->single_decode_warning) {
1723 av_log(h->avctx, AV_LOG_INFO,
1724 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1725 h0->single_decode_warning = 1;
1728 av_log(h->avctx, AV_LOG_ERROR,
1729 "Deblocking switched inside frame.\n");
1735 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1737 h->pps.chroma_qp_index_offset[0],
1738 h->pps.chroma_qp_index_offset[1]) +
1739 6 * (h->sps.bit_depth_luma - 8);
1741 h0->last_slice_type = slice_type;
1742 h->slice_num = ++h0->current_slice;
1743 if (h->slice_num >= MAX_SLICES) {
1744 av_log(h->avctx, AV_LOG_ERROR,
1745 "Too many slices, increase MAX_SLICES and recompile\n");
1748 for (j = 0; j < 2; j++) {
1750 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1751 for (i = 0; i < 16; i++) {
1753 if (j < h->list_count && i < h->ref_count[j] &&
1754 h->ref_list[j][i].f.buf[0]) {
1756 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1757 for (k = 0; k < h->short_ref_count; k++)
1758 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1762 for (k = 0; k < h->long_ref_count; k++)
1763 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1764 id_list[i] = h->short_ref_count + k;
1772 for (i = 0; i < 16; i++)
1773 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1775 ref2frm[18 + 1] = -1;
1776 for (i = 16; i < 48; i++)
1777 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1778 (h->ref_list[j][i].reference & 3);
1781 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1782 av_log(h->avctx, AV_LOG_DEBUG,
1783 "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",
1785 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1787 av_get_picture_type_char(h->slice_type),
1788 h->slice_type_fixed ? " fix" : "",
1789 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1790 pps_id, h->frame_num,
1791 h->cur_pic_ptr->field_poc[0],
1792 h->cur_pic_ptr->field_poc[1],
1793 h->ref_count[0], h->ref_count[1],
1795 h->deblocking_filter,
1796 h->slice_alpha_c0_offset, h->slice_beta_offset,
1798 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1799 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1805 int ff_h264_get_slice_type(const H264Context *h)
1807 switch (h->slice_type) {
1808 case AV_PICTURE_TYPE_P:
1810 case AV_PICTURE_TYPE_B:
1812 case AV_PICTURE_TYPE_I:
1814 case AV_PICTURE_TYPE_SP:
1816 case AV_PICTURE_TYPE_SI:
1819 return AVERROR_INVALIDDATA;
1823 static av_always_inline void fill_filter_caches_inter(H264Context *h,
1824 int mb_type, int top_xy,
1825 int left_xy[LEFT_MBS],
1827 int left_type[LEFT_MBS],
1828 int mb_xy, int list)
1830 int b_stride = h->b_stride;
1831 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
1832 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
1833 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1834 if (USES_LIST(top_type, list)) {
1835 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1836 const int b8_xy = 4 * top_xy + 2;
1837 int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1838 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1839 ref_cache[0 - 1 * 8] =
1840 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1841 ref_cache[2 - 1 * 8] =
1842 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1844 AV_ZERO128(mv_dst - 1 * 8);
1845 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1848 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1849 if (USES_LIST(left_type[LTOP], list)) {
1850 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1851 const int b8_xy = 4 * left_xy[LTOP] + 1;
1852 int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1853 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1854 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1855 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1856 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1858 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1859 ref_cache[-1 + 16] =
1860 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1862 AV_ZERO32(mv_dst - 1 + 0);
1863 AV_ZERO32(mv_dst - 1 + 8);
1864 AV_ZERO32(mv_dst - 1 + 16);
1865 AV_ZERO32(mv_dst - 1 + 24);
1868 ref_cache[-1 + 16] =
1869 ref_cache[-1 + 24] = LIST_NOT_USED;
1874 if (!USES_LIST(mb_type, list)) {
1875 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1876 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1877 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1878 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1879 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1884 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1885 int (*ref2frm)[64] = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1886 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1887 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1888 AV_WN32A(&ref_cache[0 * 8], ref01);
1889 AV_WN32A(&ref_cache[1 * 8], ref01);
1890 AV_WN32A(&ref_cache[2 * 8], ref23);
1891 AV_WN32A(&ref_cache[3 * 8], ref23);
1895 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
1896 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1897 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1898 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1899 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1905 * @return non zero if the loop filter can be skipped
1907 static int fill_filter_caches(H264Context *h, int mb_type)
1909 const int mb_xy = h->mb_xy;
1910 int top_xy, left_xy[LEFT_MBS];
1911 int top_type, left_type[LEFT_MBS];
1915 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
1917 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1918 * stuff, I can't imagine that these complex rules are worth it. */
1920 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1921 if (FRAME_MBAFF(h)) {
1922 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1923 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1925 if (left_mb_field_flag != curr_mb_field_flag)
1926 left_xy[LTOP] -= h->mb_stride;
1928 if (curr_mb_field_flag)
1929 top_xy += h->mb_stride &
1930 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1931 if (left_mb_field_flag != curr_mb_field_flag)
1932 left_xy[LBOT] += h->mb_stride;
1936 h->top_mb_xy = top_xy;
1937 h->left_mb_xy[LTOP] = left_xy[LTOP];
1938 h->left_mb_xy[LBOT] = left_xy[LBOT];
1940 /* For sufficiently low qp, filtering wouldn't do anything.
1941 * This is a conservative estimate: could also check beta_offset
1942 * and more accurate chroma_qp. */
1943 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1944 int qp = h->cur_pic.qscale_table[mb_xy];
1945 if (qp <= qp_thresh &&
1946 (left_xy[LTOP] < 0 ||
1947 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1949 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1950 if (!FRAME_MBAFF(h))
1952 if ((left_xy[LTOP] < 0 ||
1953 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1954 (top_xy < h->mb_stride ||
1955 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1960 top_type = h->cur_pic.mb_type[top_xy];
1961 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1962 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1963 if (h->deblocking_filter == 2) {
1964 if (h->slice_table[top_xy] != h->slice_num)
1966 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
1967 left_type[LTOP] = left_type[LBOT] = 0;
1969 if (h->slice_table[top_xy] == 0xFFFF)
1971 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1972 left_type[LTOP] = left_type[LBOT] = 0;
1974 h->top_type = top_type;
1975 h->left_type[LTOP] = left_type[LTOP];
1976 h->left_type[LBOT] = left_type[LBOT];
1978 if (IS_INTRA(mb_type))
1981 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1982 top_type, left_type, mb_xy, 0);
1983 if (h->list_count == 2)
1984 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1985 top_type, left_type, mb_xy, 1);
1987 nnz = h->non_zero_count[mb_xy];
1988 nnz_cache = h->non_zero_count_cache;
1989 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1990 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1991 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1992 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1993 h->cbp = h->cbp_table[mb_xy];
1996 nnz = h->non_zero_count[top_xy];
1997 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2000 if (left_type[LTOP]) {
2001 nnz = h->non_zero_count[left_xy[LTOP]];
2002 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2003 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2004 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2005 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2008 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2009 * from what the loop filter needs */
2010 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2011 if (IS_8x8DCT(top_type)) {
2012 nnz_cache[4 + 8 * 0] =
2013 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2014 nnz_cache[6 + 8 * 0] =
2015 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2017 if (IS_8x8DCT(left_type[LTOP])) {
2018 nnz_cache[3 + 8 * 1] =
2019 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2021 if (IS_8x8DCT(left_type[LBOT])) {
2022 nnz_cache[3 + 8 * 3] =
2023 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2026 if (IS_8x8DCT(mb_type)) {
2027 nnz_cache[scan8[0]] =
2028 nnz_cache[scan8[1]] =
2029 nnz_cache[scan8[2]] =
2030 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2032 nnz_cache[scan8[0 + 4]] =
2033 nnz_cache[scan8[1 + 4]] =
2034 nnz_cache[scan8[2 + 4]] =
2035 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2037 nnz_cache[scan8[0 + 8]] =
2038 nnz_cache[scan8[1 + 8]] =
2039 nnz_cache[scan8[2 + 8]] =
2040 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2042 nnz_cache[scan8[0 + 12]] =
2043 nnz_cache[scan8[1 + 12]] =
2044 nnz_cache[scan8[2 + 12]] =
2045 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2052 static void loop_filter(H264Context *h, int start_x, int end_x)
2054 uint8_t *dest_y, *dest_cb, *dest_cr;
2055 int linesize, uvlinesize, mb_x, mb_y;
2056 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2057 const int old_slice_type = h->slice_type;
2058 const int pixel_shift = h->pixel_shift;
2059 const int block_h = 16 >> h->chroma_y_shift;
2061 if (h->deblocking_filter) {
2062 for (mb_x = start_x; mb_x < end_x; mb_x++)
2063 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2065 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2066 h->slice_num = h->slice_table[mb_xy];
2067 mb_type = h->cur_pic.mb_type[mb_xy];
2068 h->list_count = h->list_counts[mb_xy];
2072 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2076 dest_y = h->cur_pic.f.data[0] +
2077 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2078 dest_cb = h->cur_pic.f.data[1] +
2079 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2080 mb_y * h->uvlinesize * block_h;
2081 dest_cr = h->cur_pic.f.data[2] +
2082 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2083 mb_y * h->uvlinesize * block_h;
2084 // FIXME simplify above
2087 linesize = h->mb_linesize = h->linesize * 2;
2088 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2089 if (mb_y & 1) { // FIXME move out of this function?
2090 dest_y -= h->linesize * 15;
2091 dest_cb -= h->uvlinesize * (block_h - 1);
2092 dest_cr -= h->uvlinesize * (block_h - 1);
2095 linesize = h->mb_linesize = h->linesize;
2096 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2098 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2100 if (fill_filter_caches(h, mb_type))
2102 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2103 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2105 if (FRAME_MBAFF(h)) {
2106 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2107 linesize, uvlinesize);
2109 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2110 dest_cr, linesize, uvlinesize);
2114 h->slice_type = old_slice_type;
2116 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2117 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2118 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2121 static void predict_field_decoding_flag(H264Context *h)
2123 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2124 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2125 h->cur_pic.mb_type[mb_xy - 1] :
2126 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2127 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2128 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2132 * Draw edges and report progress for the last MB row.
2134 static void decode_finish_row(H264Context *h)
2136 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2137 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2138 int height = 16 << FRAME_MBAFF(h);
2139 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2141 if (h->deblocking_filter) {
2142 if ((top + height) >= pic_height)
2143 height += deblock_border;
2144 top -= deblock_border;
2147 if (top >= pic_height || (top + height) < 0)
2150 height = FFMIN(height, pic_height - top);
2152 height = top + height;
2156 ff_h264_draw_horiz_band(h, top, height);
2161 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2162 h->picture_structure == PICT_BOTTOM_FIELD);
2165 static void er_add_slice(H264Context *h, int startx, int starty,
2166 int endx, int endy, int status)
2168 #if CONFIG_ERROR_RESILIENCE
2169 ERContext *er = &h->er;
2171 er->ref_count = h->ref_count[0];
2172 ff_er_add_slice(er, startx, starty, endx, endy, status);
2176 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2178 H264Context *h = *(void **)arg;
2179 int lf_x_start = h->mb_x;
2181 h->mb_skip_run = -1;
2183 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2184 avctx->codec_id != AV_CODEC_ID_H264 ||
2185 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2189 align_get_bits(&h->gb);
2192 ff_init_cabac_decoder(&h->cabac,
2193 h->gb.buffer + get_bits_count(&h->gb) / 8,
2194 (get_bits_left(&h->gb) + 7) / 8);
2196 ff_h264_init_cabac_states(h);
2200 int ret = ff_h264_decode_mb_cabac(h);
2202 // STOP_TIMER("decode_mb_cabac")
2205 ff_h264_hl_decode_mb(h);
2207 // FIXME optimal? or let mb_decode decode 16x32 ?
2208 if (ret >= 0 && FRAME_MBAFF(h)) {
2211 ret = ff_h264_decode_mb_cabac(h);
2214 ff_h264_hl_decode_mb(h);
2217 eos = get_cabac_terminate(&h->cabac);
2219 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2220 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2221 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2222 h->mb_y, ER_MB_END);
2223 if (h->mb_x >= lf_x_start)
2224 loop_filter(h, lf_x_start, h->mb_x + 1);
2227 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2228 av_log(h->avctx, AV_LOG_ERROR,
2229 "error while decoding MB %d %d, bytestream %td\n",
2231 h->cabac.bytestream_end - h->cabac.bytestream);
2232 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2233 h->mb_y, ER_MB_ERROR);
2234 return AVERROR_INVALIDDATA;
2237 if (++h->mb_x >= h->mb_width) {
2238 loop_filter(h, lf_x_start, h->mb_x);
2239 h->mb_x = lf_x_start = 0;
2240 decode_finish_row(h);
2242 if (FIELD_OR_MBAFF_PICTURE(h)) {
2244 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2245 predict_field_decoding_flag(h);
2249 if (eos || h->mb_y >= h->mb_height) {
2250 tprintf(h->avctx, "slice end %d %d\n",
2251 get_bits_count(&h->gb), h->gb.size_in_bits);
2252 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2253 h->mb_y, ER_MB_END);
2254 if (h->mb_x > lf_x_start)
2255 loop_filter(h, lf_x_start, h->mb_x);
2261 int ret = ff_h264_decode_mb_cavlc(h);
2264 ff_h264_hl_decode_mb(h);
2266 // FIXME optimal? or let mb_decode decode 16x32 ?
2267 if (ret >= 0 && FRAME_MBAFF(h)) {
2269 ret = ff_h264_decode_mb_cavlc(h);
2272 ff_h264_hl_decode_mb(h);
2277 av_log(h->avctx, AV_LOG_ERROR,
2278 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2279 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2280 h->mb_y, ER_MB_ERROR);
2284 if (++h->mb_x >= h->mb_width) {
2285 loop_filter(h, lf_x_start, h->mb_x);
2286 h->mb_x = lf_x_start = 0;
2287 decode_finish_row(h);
2289 if (FIELD_OR_MBAFF_PICTURE(h)) {
2291 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2292 predict_field_decoding_flag(h);
2294 if (h->mb_y >= h->mb_height) {
2295 tprintf(h->avctx, "slice end %d %d\n",
2296 get_bits_count(&h->gb), h->gb.size_in_bits);
2298 if (get_bits_left(&h->gb) == 0) {
2299 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2300 h->mb_x - 1, h->mb_y, ER_MB_END);
2304 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2305 h->mb_x - 1, h->mb_y, ER_MB_END);
2307 return AVERROR_INVALIDDATA;
2312 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2313 tprintf(h->avctx, "slice end %d %d\n",
2314 get_bits_count(&h->gb), h->gb.size_in_bits);
2316 if (get_bits_left(&h->gb) == 0) {
2317 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2318 h->mb_x - 1, h->mb_y, ER_MB_END);
2319 if (h->mb_x > lf_x_start)
2320 loop_filter(h, lf_x_start, h->mb_x);
2324 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2325 h->mb_y, ER_MB_ERROR);
2327 return AVERROR_INVALIDDATA;
2335 * Call decode_slice() for each context.
2337 * @param h h264 master context
2338 * @param context_count number of contexts to execute
2340 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2342 AVCodecContext *const avctx = h->avctx;
2346 if (h->mb_y >= h->mb_height) {
2347 av_log(h->avctx, AV_LOG_ERROR,
2348 "Input contains more MB rows than the frame height.\n");
2349 return AVERROR_INVALIDDATA;
2352 if (h->avctx->hwaccel)
2354 if (context_count == 1) {
2355 return decode_slice(avctx, &h);
2357 for (i = 1; i < context_count; i++) {
2358 hx = h->thread_context[i];
2359 hx->er.error_count = 0;
2362 avctx->execute(avctx, decode_slice, h->thread_context,
2363 NULL, context_count, sizeof(void *));
2365 /* pull back stuff from slices to master context */
2366 hx = h->thread_context[context_count - 1];
2369 h->droppable = hx->droppable;
2370 h->picture_structure = hx->picture_structure;
2371 for (i = 1; i < context_count; i++)
2372 h->er.error_count += h->thread_context[i]->er.error_count;