h264: Split h264 slice decoding from nal decoding
[libav.git] / libavcodec / h264_slice.c
CommitLineData
f51d0f39
LB
1/*
2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of Libav.
6 *
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.
11 *
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.
16 *
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
20 */
21
22/**
23 * @file
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28#include "libavutil/avassert.h"
29#include "libavutil/imgutils.h"
30#include "libavutil/timer.h"
31#include "internal.h"
32#include "cabac.h"
33#include "cabac_functions.h"
34#include "dsputil.h"
35#include "error_resilience.h"
36#include "avcodec.h"
37#include "h264.h"
38#include "h264data.h"
39#include "h264chroma.h"
40#include "h264_mvpred.h"
41#include "golomb.h"
42#include "mathops.h"
43#include "mpegutils.h"
44#include "rectangle.h"
45#include "thread.h"
46
47
48static 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, 3,
52};
53
54static const uint8_t div6[QP_MAX_NUM + 1] = {
55 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
56 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
57 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
58};
59
60static const uint8_t field_scan[16] = {
61 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
62 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
63 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
64 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
65};
66
67static const uint8_t field_scan8x8[64] = {
68 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
69 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
70 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
71 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
72 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
73 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
74 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
75 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
76 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
77 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
78 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
79 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
80 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
81 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
82 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
83 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
84};
85
86static const uint8_t field_scan8x8_cavlc[64] = {
87 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
88 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
89 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
90 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
91 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
92 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
93 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
94 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
95 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
96 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
97 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
98 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
99 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
100 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
101 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
102 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
103};
104
105// zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
106static const uint8_t zigzag_scan8x8_cavlc[64] = {
107 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
108 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
109 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
110 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
111 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
112 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
113 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
114 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
115 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
116 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
117 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
118 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
119 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
120 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
121 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
122 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
123};
124
125static const uint8_t dequant4_coeff_init[6][3] = {
126 { 10, 13, 16 },
127 { 11, 14, 18 },
128 { 13, 16, 20 },
129 { 14, 18, 23 },
130 { 16, 20, 25 },
131 { 18, 23, 29 },
132};
133
134static const uint8_t dequant8_coeff_init_scan[16] = {
135 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
136};
137
138static const uint8_t dequant8_coeff_init[6][6] = {
139 { 20, 18, 32, 19, 25, 24 },
140 { 22, 19, 35, 21, 28, 26 },
141 { 26, 23, 42, 24, 33, 31 },
142 { 28, 25, 45, 26, 35, 33 },
143 { 32, 28, 51, 30, 40, 38 },
144 { 36, 32, 58, 34, 46, 43 },
145};
146
147static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
148#if CONFIG_H264_DXVA2_HWACCEL
149 AV_PIX_FMT_DXVA2_VLD,
150#endif
151#if CONFIG_H264_VAAPI_HWACCEL
152 AV_PIX_FMT_VAAPI_VLD,
153#endif
154#if CONFIG_H264_VDA_HWACCEL
155 AV_PIX_FMT_VDA_VLD,
156#endif
157#if CONFIG_H264_VDPAU_HWACCEL
158 AV_PIX_FMT_VDPAU,
159#endif
160 AV_PIX_FMT_YUV420P,
161 AV_PIX_FMT_NONE
162};
163
164static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
165#if CONFIG_H264_DXVA2_HWACCEL
166 AV_PIX_FMT_DXVA2_VLD,
167#endif
168#if CONFIG_H264_VAAPI_HWACCEL
169 AV_PIX_FMT_VAAPI_VLD,
170#endif
171#if CONFIG_H264_VDA_HWACCEL
172 AV_PIX_FMT_VDA_VLD,
173#endif
174#if CONFIG_H264_VDPAU_HWACCEL
175 AV_PIX_FMT_VDPAU,
176#endif
177 AV_PIX_FMT_YUVJ420P,
178 AV_PIX_FMT_NONE
179};
180
181
182static void release_unused_pictures(H264Context *h, int remove_current)
183{
184 int i;
185
186 /* release non reference frames */
187 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
188 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
189 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
190 ff_h264_unref_picture(h, &h->DPB[i]);
191 }
192 }
193}
194
195static int alloc_scratch_buffers(H264Context *h, int linesize)
196{
197 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
198
199 if (h->bipred_scratchpad)
200 return 0;
201
202 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
203 // edge emu needs blocksize + filter length - 1
204 // (= 21x21 for h264)
205 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
206
207 if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
208 av_freep(&h->bipred_scratchpad);
209 av_freep(&h->edge_emu_buffer);
210 return AVERROR(ENOMEM);
211 }
212
213 return 0;
214}
215
216static int init_table_pools(H264Context *h)
217{
218 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
219 const int mb_array_size = h->mb_stride * h->mb_height;
220 const int b4_stride = h->mb_width * 4 + 1;
221 const int b4_array_size = b4_stride * h->mb_height * 4;
222
223 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
224 av_buffer_allocz);
225 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
226 sizeof(uint32_t), av_buffer_allocz);
227 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
228 sizeof(int16_t), av_buffer_allocz);
229 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
230
231 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
232 !h->ref_index_pool) {
233 av_buffer_pool_uninit(&h->qscale_table_pool);
234 av_buffer_pool_uninit(&h->mb_type_pool);
235 av_buffer_pool_uninit(&h->motion_val_pool);
236 av_buffer_pool_uninit(&h->ref_index_pool);
237 return AVERROR(ENOMEM);
238 }
239
240 return 0;
241}
242
243static int alloc_picture(H264Context *h, H264Picture *pic)
244{
245 int i, ret = 0;
246
247 av_assert0(!pic->f.data[0]);
248
249 pic->tf.f = &pic->f;
250 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
251 AV_GET_BUFFER_FLAG_REF : 0);
252 if (ret < 0)
253 goto fail;
254
255 h->linesize = pic->f.linesize[0];
256 h->uvlinesize = pic->f.linesize[1];
257
258 if (h->avctx->hwaccel) {
259 const AVHWAccel *hwaccel = h->avctx->hwaccel;
260 av_assert0(!pic->hwaccel_picture_private);
261 if (hwaccel->priv_data_size) {
262 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size);
263 if (!pic->hwaccel_priv_buf)
264 return AVERROR(ENOMEM);
265 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
266 }
267 }
268
269 if (!h->qscale_table_pool) {
270 ret = init_table_pools(h);
271 if (ret < 0)
272 goto fail;
273 }
274
275 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
276 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
277 if (!pic->qscale_table_buf || !pic->mb_type_buf)
278 goto fail;
279
280 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
281 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
282
283 for (i = 0; i < 2; i++) {
284 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
285 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
286 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
287 goto fail;
288
289 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
290 pic->ref_index[i] = pic->ref_index_buf[i]->data;
291 }
292
293 return 0;
294fail:
295 ff_h264_unref_picture(h, pic);
296 return (ret < 0) ? ret : AVERROR(ENOMEM);
297}
298
299static inline int pic_is_unused(H264Context *h, H264Picture *pic)
300{
301 if (!pic->f.buf[0])
302 return 1;
303 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
304 return 1;
305 return 0;
306}
307
308static int find_unused_picture(H264Context *h)
309{
310 int i;
311
312 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
313 if (pic_is_unused(h, &h->DPB[i]))
314 break;
315 }
316 if (i == H264_MAX_PICTURE_COUNT)
317 return AVERROR_INVALIDDATA;
318
319 if (h->DPB[i].needs_realloc) {
320 h->DPB[i].needs_realloc = 0;
321 ff_h264_unref_picture(h, &h->DPB[i]);
322 }
323
324 return i;
325}
326
327
328static void init_dequant8_coeff_table(H264Context *h)
329{
330 int i, j, q, x;
331 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
332
333 for (i = 0; i < 6; i++) {
334 h->dequant8_coeff[i] = h->dequant8_buffer[i];
335 for (j = 0; j < i; j++)
336 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
337 64 * sizeof(uint8_t))) {
338 h->dequant8_coeff[i] = h->dequant8_buffer[j];
339 break;
340 }
341 if (j < i)
342 continue;
343
344 for (q = 0; q < max_qp + 1; q++) {
345 int shift = div6[q];
346 int idx = rem6[q];
347 for (x = 0; x < 64; x++)
348 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
349 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
350 h->pps.scaling_matrix8[i][x]) << shift;
351 }
352 }
353}
354
355static void init_dequant4_coeff_table(H264Context *h)
356{
357 int i, j, q, x;
358 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
359 for (i = 0; i < 6; i++) {
360 h->dequant4_coeff[i] = h->dequant4_buffer[i];
361 for (j = 0; j < i; j++)
362 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
363 16 * sizeof(uint8_t))) {
364 h->dequant4_coeff[i] = h->dequant4_buffer[j];
365 break;
366 }
367 if (j < i)
368 continue;
369
370 for (q = 0; q < max_qp + 1; q++) {
371 int shift = div6[q] + 2;
372 int idx = rem6[q];
373 for (x = 0; x < 16; x++)
374 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
375 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
376 h->pps.scaling_matrix4[i][x]) << shift;
377 }
378 }
379}
380
381void h264_init_dequant_tables(H264Context *h)
382{
383 int i, x;
384 init_dequant4_coeff_table(h);
385 if (h->pps.transform_8x8_mode)
386 init_dequant8_coeff_table(h);
387 if (h->sps.transform_bypass) {
388 for (i = 0; i < 6; i++)
389 for (x = 0; x < 16; x++)
390 h->dequant4_coeff[i][0][x] = 1 << 6;
391 if (h->pps.transform_8x8_mode)
392 for (i = 0; i < 6; i++)
393 for (x = 0; x < 64; x++)
394 h->dequant8_coeff[i][0][x] = 1 << 6;
395 }
396}
397
398/**
399 * Mimic alloc_tables(), but for every context thread.
400 */
401static void clone_tables(H264Context *dst, H264Context *src, int i)
402{
403 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
404 dst->non_zero_count = src->non_zero_count;
405 dst->slice_table = src->slice_table;
406 dst->cbp_table = src->cbp_table;
407 dst->mb2b_xy = src->mb2b_xy;
408 dst->mb2br_xy = src->mb2br_xy;
409 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
410 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
411 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
412 dst->direct_table = src->direct_table;
413 dst->list_counts = src->list_counts;
414 dst->DPB = src->DPB;
415 dst->cur_pic_ptr = src->cur_pic_ptr;
416 dst->cur_pic = src->cur_pic;
417 dst->bipred_scratchpad = NULL;
418 dst->edge_emu_buffer = NULL;
419 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
420 src->sps.chroma_format_idc);
421}
422
423#define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
424#undef REBASE_PICTURE
425#define REBASE_PICTURE(pic, new_ctx, old_ctx) \
426 ((pic && pic >= old_ctx->DPB && \
427 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
428 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
429
430static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
431 H264Context *new_base,
432 H264Context *old_base)
433{
434 int i;
435
436 for (i = 0; i < count; i++) {
437 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
438 IN_RANGE(from[i], old_base->DPB,
439 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
440 !from[i]));
441 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
442 }
443}
444
445static int copy_parameter_set(void **to, void **from, int count, int size)
446{
447 int i;
448
449 for (i = 0; i < count; i++) {
450 if (to[i] && !from[i]) {
451 av_freep(&to[i]);
452 } else if (from[i] && !to[i]) {
453 to[i] = av_malloc(size);
454 if (!to[i])
455 return AVERROR(ENOMEM);
456 }
457
458 if (from[i])
459 memcpy(to[i], from[i], size);
460 }
461
462 return 0;
463}
464
465#define copy_fields(to, from, start_field, end_field) \
466 memcpy(&to->start_field, &from->start_field, \
467 (char *)&to->end_field - (char *)&to->start_field)
468
469static int h264_slice_header_init(H264Context *h, int reinit);
470
471int ff_h264_update_thread_context(AVCodecContext *dst,
472 const AVCodecContext *src)
473{
474 H264Context *h = dst->priv_data, *h1 = src->priv_data;
475 int inited = h->context_initialized, err = 0;
476 int context_reinitialized = 0;
477 int i, ret;
478
479 if (dst == src || !h1->context_initialized)
480 return 0;
481
482 if (inited &&
483 (h->width != h1->width ||
484 h->height != h1->height ||
485 h->mb_width != h1->mb_width ||
486 h->mb_height != h1->mb_height ||
487 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
488 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
489 h->sps.colorspace != h1->sps.colorspace)) {
490
491 /* set bits_per_raw_sample to the previous value. the check for changed
492 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
493 * the current value */
494 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
495
496 av_freep(&h->bipred_scratchpad);
497
498 h->width = h1->width;
499 h->height = h1->height;
500 h->mb_height = h1->mb_height;
501 h->mb_width = h1->mb_width;
502 h->mb_num = h1->mb_num;
503 h->mb_stride = h1->mb_stride;
504 h->b_stride = h1->b_stride;
505
506 if ((err = h264_slice_header_init(h, 1)) < 0) {
507 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
508 return err;
509 }
510 context_reinitialized = 1;
511
512 /* update linesize on resize. The decoder doesn't
513 * necessarily call h264_frame_start in the new thread */
514 h->linesize = h1->linesize;
515 h->uvlinesize = h1->uvlinesize;
516
517 /* copy block_offset since frame_start may not be called */
518 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
519 }
520
521 if (!inited) {
522 for (i = 0; i < MAX_SPS_COUNT; i++)
523 av_freep(h->sps_buffers + i);
524
525 for (i = 0; i < MAX_PPS_COUNT; i++)
526 av_freep(h->pps_buffers + i);
527
528 memcpy(h, h1, sizeof(*h1));
529 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
530 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
531 memset(&h->er, 0, sizeof(h->er));
532 memset(&h->mb, 0, sizeof(h->mb));
533 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
534 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
535 h->context_initialized = 0;
536
537 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
538 av_frame_unref(&h->cur_pic.f);
539 h->cur_pic.tf.f = &h->cur_pic.f;
540
541 h->avctx = dst;
542 h->DPB = NULL;
543 h->qscale_table_pool = NULL;
544 h->mb_type_pool = NULL;
545 h->ref_index_pool = NULL;
546 h->motion_val_pool = NULL;
547
548 ret = ff_h264_alloc_tables(h);
549 if (ret < 0) {
550 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
551 return ret;
552 }
553 ret = ff_h264_context_init(h);
554 if (ret < 0) {
555 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
556 return ret;
557 }
558
559 for (i = 0; i < 2; i++) {
560 h->rbsp_buffer[i] = NULL;
561 h->rbsp_buffer_size[i] = 0;
562 }
563 h->bipred_scratchpad = NULL;
564 h->edge_emu_buffer = NULL;
565
566 h->thread_context[0] = h;
567
568 h->context_initialized = 1;
569 }
570
571 h->avctx->coded_height = h1->avctx->coded_height;
572 h->avctx->coded_width = h1->avctx->coded_width;
573 h->avctx->width = h1->avctx->width;
574 h->avctx->height = h1->avctx->height;
575 h->coded_picture_number = h1->coded_picture_number;
576 h->first_field = h1->first_field;
577 h->picture_structure = h1->picture_structure;
578 h->qscale = h1->qscale;
579 h->droppable = h1->droppable;
580 h->low_delay = h1->low_delay;
581
582 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
583 ff_h264_unref_picture(h, &h->DPB[i]);
584 if (h1->DPB[i].f.buf[0] &&
585 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
586 return ret;
587 }
588
589 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
590 ff_h264_unref_picture(h, &h->cur_pic);
591 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
592 return ret;
593
594 h->workaround_bugs = h1->workaround_bugs;
595 h->low_delay = h1->low_delay;
596 h->droppable = h1->droppable;
597
598 /* frame_start may not be called for the next thread (if it's decoding
599 * a bottom field) so this has to be allocated here */
600 err = alloc_scratch_buffers(h, h1->linesize);
601 if (err < 0)
602 return err;
603
604 // extradata/NAL handling
605 h->is_avc = h1->is_avc;
606
607 // SPS/PPS
608 if ((ret = copy_parameter_set((void **)h->sps_buffers,
609 (void **)h1->sps_buffers,
610 MAX_SPS_COUNT, sizeof(SPS))) < 0)
611 return ret;
612 h->sps = h1->sps;
613 if ((ret = copy_parameter_set((void **)h->pps_buffers,
614 (void **)h1->pps_buffers,
615 MAX_PPS_COUNT, sizeof(PPS))) < 0)
616 return ret;
617 h->pps = h1->pps;
618
619 // Dequantization matrices
620 // FIXME these are big - can they be only copied when PPS changes?
621 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
622
623 for (i = 0; i < 6; i++)
624 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
625 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
626
627 for (i = 0; i < 6; i++)
628 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
629 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
630
631 h->dequant_coeff_pps = h1->dequant_coeff_pps;
632
633 // POC timing
634 copy_fields(h, h1, poc_lsb, redundant_pic_count);
635
636 // reference lists
637 copy_fields(h, h1, short_ref, cabac_init_idc);
638
639 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
640 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
641 copy_picture_range(h->delayed_pic, h1->delayed_pic,
642 MAX_DELAYED_PIC_COUNT + 2, h, h1);
643
644 h->last_slice_type = h1->last_slice_type;
645
646 if (context_reinitialized)
647 ff_h264_set_parameter_from_sps(h);
648
649 if (!h->cur_pic_ptr)
650 return 0;
651
652 if (!h->droppable) {
653 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
654 h->prev_poc_msb = h->poc_msb;
655 h->prev_poc_lsb = h->poc_lsb;
656 }
657 h->prev_frame_num_offset = h->frame_num_offset;
658 h->prev_frame_num = h->frame_num;
659 h->outputed_poc = h->next_outputed_poc;
660
661 h->recovery_frame = h1->recovery_frame;
662 h->frame_recovered = h1->frame_recovered;
663
664 return err;
665}
666
667static int h264_frame_start(H264Context *h)
668{
669 H264Picture *pic;
670 int i, ret;
671 const int pixel_shift = h->pixel_shift;
672
673 release_unused_pictures(h, 1);
674 h->cur_pic_ptr = NULL;
675
676 i = find_unused_picture(h);
677 if (i < 0) {
678 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
679 return i;
680 }
681 pic = &h->DPB[i];
682
683 pic->reference = h->droppable ? 0 : h->picture_structure;
684 pic->f.coded_picture_number = h->coded_picture_number++;
685 pic->field_picture = h->picture_structure != PICT_FRAME;
686 /*
687 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
688 * in later.
689 * See decode_nal_units().
690 */
691 pic->f.key_frame = 0;
692 pic->mmco_reset = 0;
693 pic->recovered = 0;
694
695 if ((ret = alloc_picture(h, pic)) < 0)
696 return ret;
697
698 h->cur_pic_ptr = pic;
699 ff_h264_unref_picture(h, &h->cur_pic);
700 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
701 return ret;
702
703 if (CONFIG_ERROR_RESILIENCE)
704 ff_er_frame_start(&h->er);
705
706 assert(h->linesize && h->uvlinesize);
707
708 for (i = 0; i < 16; i++) {
709 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
710 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
711 }
712 for (i = 0; i < 16; i++) {
713 h->block_offset[16 + i] =
714 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
715 h->block_offset[48 + 16 + i] =
716 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
717 }
718
719 /* can't be in alloc_tables because linesize isn't known there.
720 * FIXME: redo bipred weight to not require extra buffer? */
721 for (i = 0; i < h->slice_context_count; i++)
722 if (h->thread_context[i]) {
723 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
724 if (ret < 0)
725 return ret;
726 }
727
728 /* Some macroblocks can be accessed before they're available in case
729 * of lost slices, MBAFF or threading. */
730 memset(h->slice_table, -1,
731 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
732
733 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
734 // s->current_picture.f.reference /* || h->contains_intra */ || 1;
735
736 /* We mark the current picture as non-reference after allocating it, so
737 * that if we break out due to an error it can be released automatically
738 * in the next ff_MPV_frame_start().
739 */
740 h->cur_pic_ptr->reference = 0;
741
742 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
743
744 h->next_output_pic = NULL;
745
746 assert(h->cur_pic_ptr->long_ref == 0);
747
748 return 0;
749}
750
751static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
752 uint8_t *src_cb, uint8_t *src_cr,
753 int linesize, int uvlinesize,
754 int simple)
755{
756 uint8_t *top_border;
757 int top_idx = 1;
758 const int pixel_shift = h->pixel_shift;
759 int chroma444 = CHROMA444(h);
760 int chroma422 = CHROMA422(h);
761
762 src_y -= linesize;
763 src_cb -= uvlinesize;
764 src_cr -= uvlinesize;
765
766 if (!simple && FRAME_MBAFF(h)) {
767 if (h->mb_y & 1) {
768 if (!MB_MBAFF(h)) {
769 top_border = h->top_borders[0][h->mb_x];
770 AV_COPY128(top_border, src_y + 15 * linesize);
771 if (pixel_shift)
772 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
773 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
774 if (chroma444) {
775 if (pixel_shift) {
776 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
777 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
778 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
779 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
780 } else {
781 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
782 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
783 }
784 } else if (chroma422) {
785 if (pixel_shift) {
786 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
787 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
788 } else {
789 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
790 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
791 }
792 } else {
793 if (pixel_shift) {
794 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
795 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
796 } else {
797 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
798 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
799 }
800 }
801 }
802 }
803 } else if (MB_MBAFF(h)) {
804 top_idx = 0;
805 } else
806 return;
807 }
808
809 top_border = h->top_borders[top_idx][h->mb_x];
810 /* There are two lines saved, the line above the top macroblock
811 * of a pair, and the line above the bottom macroblock. */
812 AV_COPY128(top_border, src_y + 16 * linesize);
813 if (pixel_shift)
814 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
815
816 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
817 if (chroma444) {
818 if (pixel_shift) {
819 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
820 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
821 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
822 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
823 } else {
824 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
825 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
826 }
827 } else if (chroma422) {
828 if (pixel_shift) {
829 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
830 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
831 } else {
832 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
833 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
834 }
835 } else {
836 if (pixel_shift) {
837 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
838 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
839 } else {
840 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
841 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
842 }
843 }
844 }
845}
846
847/**
848 * Initialize implicit_weight table.
849 * @param field 0/1 initialize the weight for interlaced MBAFF
850 * -1 initializes the rest
851 */
852static void implicit_weight_table(H264Context *h, int field)
853{
854 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
855
856 for (i = 0; i < 2; i++) {
857 h->luma_weight_flag[i] = 0;
858 h->chroma_weight_flag[i] = 0;
859 }
860
861 if (field < 0) {
862 if (h->picture_structure == PICT_FRAME) {
863 cur_poc = h->cur_pic_ptr->poc;
864 } else {
865 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
866 }
867 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
868 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
869 h->use_weight = 0;
870 h->use_weight_chroma = 0;
871 return;
872 }
873 ref_start = 0;
874 ref_count0 = h->ref_count[0];
875 ref_count1 = h->ref_count[1];
876 } else {
877 cur_poc = h->cur_pic_ptr->field_poc[field];
878 ref_start = 16;
879 ref_count0 = 16 + 2 * h->ref_count[0];
880 ref_count1 = 16 + 2 * h->ref_count[1];
881 }
882
883 h->use_weight = 2;
884 h->use_weight_chroma = 2;
885 h->luma_log2_weight_denom = 5;
886 h->chroma_log2_weight_denom = 5;
887
888 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
889 int poc0 = h->ref_list[0][ref0].poc;
890 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
891 int w = 32;
892 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
893 int poc1 = h->ref_list[1][ref1].poc;
894 int td = av_clip(poc1 - poc0, -128, 127);
895 if (td) {
896 int tb = av_clip(cur_poc - poc0, -128, 127);
897 int tx = (16384 + (FFABS(td) >> 1)) / td;
898 int dist_scale_factor = (tb * tx + 32) >> 8;
899 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
900 w = 64 - dist_scale_factor;
901 }
902 }
903 if (field < 0) {
904 h->implicit_weight[ref0][ref1][0] =
905 h->implicit_weight[ref0][ref1][1] = w;
906 } else {
907 h->implicit_weight[ref0][ref1][field] = w;
908 }
909 }
910 }
911}
912
913/**
914 * initialize scan tables
915 */
916static void init_scan_tables(H264Context *h)
917{
918 int i;
919 for (i = 0; i < 16; i++) {
920#define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
921 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
922 h->field_scan[i] = TRANSPOSE(field_scan[i]);
923#undef TRANSPOSE
924 }
925 for (i = 0; i < 64; i++) {
926#define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
927 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
928 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
929 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
930 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
931#undef TRANSPOSE
932 }
933 if (h->sps.transform_bypass) { // FIXME same ugly
934 h->zigzag_scan_q0 = zigzag_scan;
935 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
936 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
937 h->field_scan_q0 = field_scan;
938 h->field_scan8x8_q0 = field_scan8x8;
939 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
940 } else {
941 h->zigzag_scan_q0 = h->zigzag_scan;
942 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
943 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
944 h->field_scan_q0 = h->field_scan;
945 h->field_scan8x8_q0 = h->field_scan8x8;
946 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
947 }
948}
949
950/**
951 * Replicate H264 "master" context to thread contexts.
952 */
953static int clone_slice(H264Context *dst, H264Context *src)
954{
955 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
956 dst->cur_pic_ptr = src->cur_pic_ptr;
957 dst->cur_pic = src->cur_pic;
958 dst->linesize = src->linesize;
959 dst->uvlinesize = src->uvlinesize;
960 dst->first_field = src->first_field;
961
962 dst->prev_poc_msb = src->prev_poc_msb;
963 dst->prev_poc_lsb = src->prev_poc_lsb;
964 dst->prev_frame_num_offset = src->prev_frame_num_offset;
965 dst->prev_frame_num = src->prev_frame_num;
966 dst->short_ref_count = src->short_ref_count;
967
968 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
969 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
970 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
971
972 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
973 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
974
975 return 0;
976}
977
978static enum AVPixelFormat get_pixel_format(H264Context *h)
979{
980 switch (h->sps.bit_depth_luma) {
981 case 9:
982 if (CHROMA444(h)) {
983 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
984 return AV_PIX_FMT_GBRP9;
985 } else
986 return AV_PIX_FMT_YUV444P9;
987 } else if (CHROMA422(h))
988 return AV_PIX_FMT_YUV422P9;
989 else
990 return AV_PIX_FMT_YUV420P9;
991 break;
992 case 10:
993 if (CHROMA444(h)) {
994 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
995 return AV_PIX_FMT_GBRP10;
996 } else
997 return AV_PIX_FMT_YUV444P10;
998 } else if (CHROMA422(h))
999 return AV_PIX_FMT_YUV422P10;
1000 else
1001 return AV_PIX_FMT_YUV420P10;
1002 break;
1003 case 8:
1004 if (CHROMA444(h)) {
1005 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1006 return AV_PIX_FMT_GBRP;
1007 } else
1008 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
1009 : AV_PIX_FMT_YUV444P;
1010 } else if (CHROMA422(h)) {
1011 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
1012 : AV_PIX_FMT_YUV422P;
1013 } else {
1014 return h->avctx->get_format(h->avctx, h->avctx->codec->pix_fmts ?
1015 h->avctx->codec->pix_fmts :
1016 h->avctx->color_range == AVCOL_RANGE_JPEG ?
1017 h264_hwaccel_pixfmt_list_jpeg_420 :
1018 h264_hwaccel_pixfmt_list_420);
1019 }
1020 break;
1021 default:
1022 av_log(h->avctx, AV_LOG_ERROR,
1023 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1024 return AVERROR_INVALIDDATA;
1025 }
1026}
1027
1028/* export coded and cropped frame dimensions to AVCodecContext */
1029static int init_dimensions(H264Context *h)
1030{
1031 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1032 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1033
1034 /* handle container cropping */
1035 if (!h->sps.crop &&
1036 FFALIGN(h->avctx->width, 16) == h->width &&
1037 FFALIGN(h->avctx->height, 16) == h->height) {
1038 width = h->avctx->width;
1039 height = h->avctx->height;
1040 }
1041
1042 if (width <= 0 || height <= 0) {
1043 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1044 width, height);
1045 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1046 return AVERROR_INVALIDDATA;
1047
1048 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1049 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
1050 h->sps.crop = 0;
1051
1052 width = h->width;
1053 height = h->height;
1054 }
1055
1056 h->avctx->coded_width = h->width;
1057 h->avctx->coded_height = h->height;
1058 h->avctx->width = width;
1059 h->avctx->height = height;
1060
1061 return 0;
1062}
1063
1064static int h264_slice_header_init(H264Context *h, int reinit)
1065{
1066 int nb_slices = (HAVE_THREADS &&
1067 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1068 h->avctx->thread_count : 1;
1069 int i, ret;
1070
1071 h->avctx->sample_aspect_ratio = h->sps.sar;
1072 av_assert0(h->avctx->sample_aspect_ratio.den);
1073 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1074 &h->chroma_x_shift, &h->chroma_y_shift);
1075
1076 if (h->sps.timing_info_present_flag) {
1077 int64_t den = h->sps.time_scale;
1078 if (h->x264_build < 44U)
1079 den *= 2;
1080 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
1081 h->sps.num_units_in_tick, den, 1 << 30);
1082 }
1083
1084 h->avctx->hwaccel = ff_find_hwaccel(h->avctx);
1085
1086 if (reinit)
1087 ff_h264_free_tables(h, 0);
1088 h->first_field = 0;
1089 h->prev_interlaced_frame = 1;
1090
1091 init_scan_tables(h);
1092 ret = ff_h264_alloc_tables(h);
1093 if (ret < 0) {
1094 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1095 return ret;
1096 }
1097
1098 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1099 int max_slices;
1100 if (h->mb_height)
1101 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1102 else
1103 max_slices = H264_MAX_THREADS;
1104 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1105 " reducing to %d\n", nb_slices, max_slices);
1106 nb_slices = max_slices;
1107 }
1108 h->slice_context_count = nb_slices;
1109
1110 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1111 ret = ff_h264_context_init(h);
1112 if (ret < 0) {
1113 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1114 return ret;
1115 }
1116 } else {
1117 for (i = 1; i < h->slice_context_count; i++) {
1118 H264Context *c;
1119 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1120 if (!c)
1121 return AVERROR(ENOMEM);
1122 c->avctx = h->avctx;
1123 c->dsp = h->dsp;
1124 c->vdsp = h->vdsp;
1125 c->h264dsp = h->h264dsp;
1126 c->h264qpel = h->h264qpel;
1127 c->h264chroma = h->h264chroma;
1128 c->sps = h->sps;
1129 c->pps = h->pps;
1130 c->pixel_shift = h->pixel_shift;
1131 c->width = h->width;
1132 c->height = h->height;
1133 c->linesize = h->linesize;
1134 c->uvlinesize = h->uvlinesize;
1135 c->chroma_x_shift = h->chroma_x_shift;
1136 c->chroma_y_shift = h->chroma_y_shift;
1137 c->qscale = h->qscale;
1138 c->droppable = h->droppable;
1139 c->data_partitioning = h->data_partitioning;
1140 c->low_delay = h->low_delay;
1141 c->mb_width = h->mb_width;
1142 c->mb_height = h->mb_height;
1143 c->mb_stride = h->mb_stride;
1144 c->mb_num = h->mb_num;
1145 c->flags = h->flags;
1146 c->workaround_bugs = h->workaround_bugs;
1147 c->pict_type = h->pict_type;
1148
1149 init_scan_tables(c);
1150 clone_tables(c, h, i);
1151 c->context_initialized = 1;
1152 }
1153
1154 for (i = 0; i < h->slice_context_count; i++)
1155 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1156 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1157 return ret;
1158 }
1159 }
1160
1161 h->context_initialized = 1;
1162
1163 return 0;
1164}
1165
1166/**
1167 * Decode a slice header.
1168 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1169 *
1170 * @param h h264context
1171 * @param h0 h264 master context (differs from 'h' when doing sliced based
1172 * parallel decoding)
1173 *
1174 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1175 */
1176int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1177{
1178 unsigned int first_mb_in_slice;
1179 unsigned int pps_id;
1180 int ret;
1181 unsigned int slice_type, tmp, i, j;
1182 int default_ref_list_done = 0;
1183 int last_pic_structure, last_pic_droppable;
1184 int needs_reinit = 0;
1185 int field_pic_flag, bottom_field_flag;
1186
1187 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1188 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1189
1190 first_mb_in_slice = get_ue_golomb(&h->gb);
1191
1192 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1193 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1194 ff_h264_field_end(h, 1);
1195 }
1196
1197 h0->current_slice = 0;
1198 if (!h0->first_field) {
1199 if (h->cur_pic_ptr && !h->droppable) {
1200 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1201 h->picture_structure == PICT_BOTTOM_FIELD);
1202 }
1203 h->cur_pic_ptr = NULL;
1204 }
1205 }
1206
1207 slice_type = get_ue_golomb_31(&h->gb);
1208 if (slice_type > 9) {
1209 av_log(h->avctx, AV_LOG_ERROR,
1210 "slice type %d too large at %d %d\n",
1211 slice_type, h->mb_x, h->mb_y);
1212 return AVERROR_INVALIDDATA;
1213 }
1214 if (slice_type > 4) {
1215 slice_type -= 5;
1216 h->slice_type_fixed = 1;
1217 } else
1218 h->slice_type_fixed = 0;
1219
1220 slice_type = golomb_to_pict_type[slice_type];
1221 if (slice_type == AV_PICTURE_TYPE_I ||
1222 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
1223 default_ref_list_done = 1;
1224 }
1225 h->slice_type = slice_type;
1226 h->slice_type_nos = slice_type & 3;
1227
1228 if (h->nal_unit_type == NAL_IDR_SLICE &&
1229 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1230 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1231 return AVERROR_INVALIDDATA;
1232 }
1233
1234 // to make a few old functions happy, it's wrong though
1235 h->pict_type = h->slice_type;
1236
1237 pps_id = get_ue_golomb(&h->gb);
1238 if (pps_id >= MAX_PPS_COUNT) {
1239 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1240 return AVERROR_INVALIDDATA;
1241 }
1242 if (!h0->pps_buffers[pps_id]) {
1243 av_log(h->avctx, AV_LOG_ERROR,
1244 "non-existing PPS %u referenced\n",
1245 pps_id);
1246 return AVERROR_INVALIDDATA;
1247 }
1248 h->pps = *h0->pps_buffers[pps_id];
1249
1250 if (!h0->sps_buffers[h->pps.sps_id]) {
1251 av_log(h->avctx, AV_LOG_ERROR,
1252 "non-existing SPS %u referenced\n",
1253 h->pps.sps_id);
1254 return AVERROR_INVALIDDATA;
1255 }
1256
1257 if (h->pps.sps_id != h->sps.sps_id ||
1258 h0->sps_buffers[h->pps.sps_id]->new) {
1259 h0->sps_buffers[h->pps.sps_id]->new = 0;
1260
1261 h->sps = *h0->sps_buffers[h->pps.sps_id];
1262
1263 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1264 h->chroma_format_idc != h->sps.chroma_format_idc) {
1265 h->bit_depth_luma = h->sps.bit_depth_luma;
1266 h->chroma_format_idc = h->sps.chroma_format_idc;
1267 needs_reinit = 1;
1268 }
1269 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1270 return ret;
1271 }
1272
1273 h->avctx->profile = ff_h264_get_profile(&h->sps);
1274 h->avctx->level = h->sps.level_idc;
1275 h->avctx->refs = h->sps.ref_frame_count;
1276
1277 if (h->mb_width != h->sps.mb_width ||
1278 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1279 needs_reinit = 1;
1280
1281 h->mb_width = h->sps.mb_width;
1282 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1283 h->mb_num = h->mb_width * h->mb_height;
1284 h->mb_stride = h->mb_width + 1;
1285
1286 h->b_stride = h->mb_width * 4;
1287
1288 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1289
1290 h->width = 16 * h->mb_width;
1291 h->height = 16 * h->mb_height;
1292
1293 ret = init_dimensions(h);
1294 if (ret < 0)
1295 return ret;
1296
1297 if (h->sps.video_signal_type_present_flag) {
1298 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1299 : AVCOL_RANGE_MPEG;
1300 if (h->sps.colour_description_present_flag) {
1301 if (h->avctx->colorspace != h->sps.colorspace)
1302 needs_reinit = 1;
1303 h->avctx->color_primaries = h->sps.color_primaries;
1304 h->avctx->color_trc = h->sps.color_trc;
1305 h->avctx->colorspace = h->sps.colorspace;
1306 }
1307 }
1308
1309 if (h->context_initialized &&
1310 (h->width != h->avctx->coded_width ||
1311 h->height != h->avctx->coded_height ||
1312 needs_reinit)) {
1313 if (h != h0) {
1314 av_log(h->avctx, AV_LOG_ERROR,
1315 "changing width %d -> %d / height %d -> %d on "
1316 "slice %d\n",
1317 h->width, h->avctx->coded_width,
1318 h->height, h->avctx->coded_height,
1319 h0->current_slice + 1);
1320 return AVERROR_INVALIDDATA;
1321 }
1322
1323 ff_h264_flush_change(h);
1324
1325 if ((ret = get_pixel_format(h)) < 0)
1326 return ret;
1327 h->avctx->pix_fmt = ret;
1328
1329 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1330 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1331
1332 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1333 av_log(h->avctx, AV_LOG_ERROR,
1334 "h264_slice_header_init() failed\n");
1335 return ret;
1336 }
1337 }
1338 if (!h->context_initialized) {
1339 if (h != h0) {
1340 av_log(h->avctx, AV_LOG_ERROR,
1341 "Cannot (re-)initialize context during parallel decoding.\n");
1342 return AVERROR_PATCHWELCOME;
1343 }
1344
1345 if ((ret = get_pixel_format(h)) < 0)
1346 return ret;
1347 h->avctx->pix_fmt = ret;
1348
1349 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1350 av_log(h->avctx, AV_LOG_ERROR,
1351 "h264_slice_header_init() failed\n");
1352 return ret;
1353 }
1354 }
1355
1356 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1357 h->dequant_coeff_pps = pps_id;
1358 h264_init_dequant_tables(h);
1359 }
1360
1361 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1362
1363 h->mb_mbaff = 0;
1364 h->mb_aff_frame = 0;
1365 last_pic_structure = h0->picture_structure;
1366 last_pic_droppable = h0->droppable;
1367 h->droppable = h->nal_ref_idc == 0;
1368 if (h->sps.frame_mbs_only_flag) {
1369 h->picture_structure = PICT_FRAME;
1370 } else {
1371 field_pic_flag = get_bits1(&h->gb);
1372 if (field_pic_flag) {
1373 bottom_field_flag = get_bits1(&h->gb);
1374 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1375 } else {
1376 h->picture_structure = PICT_FRAME;
1377 h->mb_aff_frame = h->sps.mb_aff;
1378 }
1379 }
1380 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1381
1382 if (h0->current_slice != 0) {
1383 if (last_pic_structure != h->picture_structure ||
1384 last_pic_droppable != h->droppable) {
1385 av_log(h->avctx, AV_LOG_ERROR,
1386 "Changing field mode (%d -> %d) between slices is not allowed\n",
1387 last_pic_structure, h->picture_structure);
1388 h->picture_structure = last_pic_structure;
1389 h->droppable = last_pic_droppable;
1390 return AVERROR_INVALIDDATA;
1391 } else if (!h0->cur_pic_ptr) {
1392 av_log(h->avctx, AV_LOG_ERROR,
1393 "unset cur_pic_ptr on slice %d\n",
1394 h0->current_slice + 1);
1395 return AVERROR_INVALIDDATA;
1396 }
1397 } else {
1398 /* Shorten frame num gaps so we don't have to allocate reference
1399 * frames just to throw them away */
1400 if (h->frame_num != h->prev_frame_num) {
1401 int unwrap_prev_frame_num = h->prev_frame_num;
1402 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1403
1404 if (unwrap_prev_frame_num > h->frame_num)
1405 unwrap_prev_frame_num -= max_frame_num;
1406
1407 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1408 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1409 if (unwrap_prev_frame_num < 0)
1410 unwrap_prev_frame_num += max_frame_num;
1411
1412 h->prev_frame_num = unwrap_prev_frame_num;
1413 }
1414 }
1415
1416 /* See if we have a decoded first field looking for a pair...
1417 * Here, we're using that to see if we should mark previously
1418 * decode frames as "finished".
1419 * We have to do that before the "dummy" in-between frame allocation,
1420 * since that can modify s->current_picture_ptr. */
1421 if (h0->first_field) {
1422 assert(h0->cur_pic_ptr);
1423 assert(h0->cur_pic_ptr->f.buf[0]);
1424 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1425
1426 /* figure out if we have a complementary field pair */
1427 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1428 /* Previous field is unmatched. Don't display it, but let it
1429 * remain for reference if marked as such. */
1430 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1431 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1432 last_pic_structure == PICT_TOP_FIELD);
1433 }
1434 } else {
1435 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1436 /* This and previous field were reference, but had
1437 * different frame_nums. Consider this field first in
1438 * pair. Throw away previous field except for reference
1439 * purposes. */
1440 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1441 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1442 last_pic_structure == PICT_TOP_FIELD);
1443 }
1444 } else {
1445 /* Second field in complementary pair */
1446 if (!((last_pic_structure == PICT_TOP_FIELD &&
1447 h->picture_structure == PICT_BOTTOM_FIELD) ||
1448 (last_pic_structure == PICT_BOTTOM_FIELD &&
1449 h->picture_structure == PICT_TOP_FIELD))) {
1450 av_log(h->avctx, AV_LOG_ERROR,
1451 "Invalid field mode combination %d/%d\n",
1452 last_pic_structure, h->picture_structure);
1453 h->picture_structure = last_pic_structure;
1454 h->droppable = last_pic_droppable;
1455 return AVERROR_INVALIDDATA;
1456 } else if (last_pic_droppable != h->droppable) {
1457 avpriv_request_sample(h->avctx,
1458 "Found reference and non-reference fields in the same frame, which");
1459 h->picture_structure = last_pic_structure;
1460 h->droppable = last_pic_droppable;
1461 return AVERROR_PATCHWELCOME;
1462 }
1463 }
1464 }
1465 }
1466
1467 while (h->frame_num != h->prev_frame_num &&
1468 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1469 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1470 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1471 h->frame_num, h->prev_frame_num);
1472 ret = h264_frame_start(h);
1473 if (ret < 0) {
1474 h0->first_field = 0;
1475 return ret;
1476 }
1477
1478 h->prev_frame_num++;
1479 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1480 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1481 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1482 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1483 ret = ff_generate_sliding_window_mmcos(h, 1);
1484 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1485 return ret;
1486 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1487 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1488 return ret;
1489 /* Error concealment: If a ref is missing, copy the previous ref
1490 * in its place.
1491 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1492 * many assumptions about there being no actual duplicates.
1493 * FIXME: This does not copy padding for out-of-frame motion
1494 * vectors. Given we are concealing a lost frame, this probably
1495 * is not noticeable by comparison, but it should be fixed. */
1496 if (h->short_ref_count) {
1497 if (prev) {
1498 av_image_copy(h->short_ref[0]->f.data,
1499 h->short_ref[0]->f.linesize,
1500 (const uint8_t **)prev->f.data,
1501 prev->f.linesize,
1502 h->avctx->pix_fmt,
1503 h->mb_width * 16,
1504 h->mb_height * 16);
1505 h->short_ref[0]->poc = prev->poc + 2;
1506 }
1507 h->short_ref[0]->frame_num = h->prev_frame_num;
1508 }
1509 }
1510
1511 /* See if we have a decoded first field looking for a pair...
1512 * We're using that to see whether to continue decoding in that
1513 * frame, or to allocate a new one. */
1514 if (h0->first_field) {
1515 assert(h0->cur_pic_ptr);
1516 assert(h0->cur_pic_ptr->f.buf[0]);
1517 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1518
1519 /* figure out if we have a complementary field pair */
1520 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1521 /* Previous field is unmatched. Don't display it, but let it
1522 * remain for reference if marked as such. */
1523 h0->cur_pic_ptr = NULL;
1524 h0->first_field = FIELD_PICTURE(h);
1525 } else {
1526 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1527 /* This and the previous field had different frame_nums.
1528 * Consider this field first in pair. Throw away previous
1529 * one except for reference purposes. */
1530 h0->first_field = 1;
1531 h0->cur_pic_ptr = NULL;
1532 } else {
1533 /* Second field in complementary pair */
1534 h0->first_field = 0;
1535 }
1536 }
1537 } else {
1538 /* Frame or first field in a potentially complementary pair */
1539 h0->first_field = FIELD_PICTURE(h);
1540 }
1541
1542 if (!FIELD_PICTURE(h) || h0->first_field) {
1543 if (h264_frame_start(h) < 0) {
1544 h0->first_field = 0;
1545 return AVERROR_INVALIDDATA;
1546 }
1547 } else {
1548 release_unused_pictures(h, 0);
1549 }
1550 }
1551 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1552 return ret;
1553
1554 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1555
1556 assert(h->mb_num == h->mb_width * h->mb_height);
1557 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1558 first_mb_in_slice >= h->mb_num) {
1559 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1560 return AVERROR_INVALIDDATA;
1561 }
1562 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1563 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1564 FIELD_OR_MBAFF_PICTURE(h);
1565 if (h->picture_structure == PICT_BOTTOM_FIELD)
1566 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1567 assert(h->mb_y < h->mb_height);
1568
1569 if (h->picture_structure == PICT_FRAME) {
1570 h->curr_pic_num = h->frame_num;
1571 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1572 } else {
1573 h->curr_pic_num = 2 * h->frame_num + 1;
1574 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1575 }
1576
1577 if (h->nal_unit_type == NAL_IDR_SLICE)
1578 get_ue_golomb(&h->gb); /* idr_pic_id */
1579
1580 if (h->sps.poc_type == 0) {
1581 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1582
1583 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1584 h->delta_poc_bottom = get_se_golomb(&h->gb);
1585 }
1586
1587 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1588 h->delta_poc[0] = get_se_golomb(&h->gb);
1589
1590 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1591 h->delta_poc[1] = get_se_golomb(&h->gb);
1592 }
1593
1594 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1595
1596 if (h->pps.redundant_pic_cnt_present)
1597 h->redundant_pic_count = get_ue_golomb(&h->gb);
1598
1599 ret = ff_set_ref_count(h);
1600 if (ret < 0)
1601 return ret;
1602 else if (ret == 1)
1603 default_ref_list_done = 0;
1604
1605 if (!default_ref_list_done)
1606 ff_h264_fill_default_ref_list(h);
1607
1608 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1609 ret = ff_h264_decode_ref_pic_list_reordering(h);
1610 if (ret < 0) {
1611 h->ref_count[1] = h->ref_count[0] = 0;
1612 return ret;
1613 }
1614 }
1615
1616 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1617 (h->pps.weighted_bipred_idc == 1 &&
1618 h->slice_type_nos == AV_PICTURE_TYPE_B))
1619 ff_pred_weight_table(h);
1620 else if (h->pps.weighted_bipred_idc == 2 &&
1621 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1622 implicit_weight_table(h, -1);
1623 } else {
1624 h->use_weight = 0;
1625 for (i = 0; i < 2; i++) {
1626 h->luma_weight_flag[i] = 0;
1627 h->chroma_weight_flag[i] = 0;
1628 }
1629 }
1630
1631 // If frame-mt is enabled, only update mmco tables for the first slice
1632 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1633 // or h->mmco, which will cause ref list mix-ups and decoding errors
1634 // further down the line. This may break decoding if the first slice is
1635 // corrupt, thus we only do this if frame-mt is enabled.
1636 if (h->nal_ref_idc) {
1637 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1638 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1639 h0->current_slice == 0);
1640 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1641 return AVERROR_INVALIDDATA;
1642 }
1643
1644 if (FRAME_MBAFF(h)) {
1645 ff_h264_fill_mbaff_ref_list(h);
1646
1647 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1648 implicit_weight_table(h, 0);
1649 implicit_weight_table(h, 1);
1650 }
1651 }
1652
1653 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1654 ff_h264_direct_dist_scale_factor(h);
1655 ff_h264_direct_ref_list_init(h);
1656
1657 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1658 tmp = get_ue_golomb_31(&h->gb);
1659 if (tmp > 2) {
1660 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1661 return AVERROR_INVALIDDATA;
1662 }
1663 h->cabac_init_idc = tmp;
1664 }
1665
1666 h->last_qscale_diff = 0;
1667 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1668 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1669 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1670 return AVERROR_INVALIDDATA;
1671 }
1672 h->qscale = tmp;
1673 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1674 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1675 // FIXME qscale / qp ... stuff
1676 if (h->slice_type == AV_PICTURE_TYPE_SP)
1677 get_bits1(&h->gb); /* sp_for_switch_flag */
1678 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1679 h->slice_type == AV_PICTURE_TYPE_SI)
1680 get_se_golomb(&h->gb); /* slice_qs_delta */
1681
1682 h->deblocking_filter = 1;
1683 h->slice_alpha_c0_offset = 0;
1684 h->slice_beta_offset = 0;
1685 if (h->pps.deblocking_filter_parameters_present) {
1686 tmp = get_ue_golomb_31(&h->gb);
1687 if (tmp > 2) {
1688 av_log(h->avctx, AV_LOG_ERROR,
1689 "deblocking_filter_idc %u out of range\n", tmp);
1690 return AVERROR_INVALIDDATA;
1691 }
1692 h->deblocking_filter = tmp;
1693 if (h->deblocking_filter < 2)
1694 h->deblocking_filter ^= 1; // 1<->0
1695
1696 if (h->deblocking_filter) {
1697 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1698 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1699 if (h->slice_alpha_c0_offset > 12 ||
1700 h->slice_alpha_c0_offset < -12 ||
1701 h->slice_beta_offset > 12 ||
1702 h->slice_beta_offset < -12) {
1703 av_log(h->avctx, AV_LOG_ERROR,
1704 "deblocking filter parameters %d %d out of range\n",
1705 h->slice_alpha_c0_offset, h->slice_beta_offset);
1706 return AVERROR_INVALIDDATA;
1707 }
1708 }
1709 }
1710
1711 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1712 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1713 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1714 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1715 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1716 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1717 h->nal_ref_idc == 0))
1718 h->deblocking_filter = 0;
1719
1720 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1721 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1722 /* Cheat slightly for speed:
1723 * Do not bother to deblock across slices. */
1724 h->deblocking_filter = 2;
1725 } else {
1726 h0->max_contexts = 1;
1727 if (!h0->single_decode_warning) {
1728 av_log(h->avctx, AV_LOG_INFO,
1729 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1730 h0->single_decode_warning = 1;
1731 }
1732 if (h != h0) {
1733 av_log(h->avctx, AV_LOG_ERROR,
1734 "Deblocking switched inside frame.\n");
1735 return 1;
1736 }
1737 }
1738 }
1739 h->qp_thresh = 15 -
1740 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1741 FFMAX3(0,
1742 h->pps.chroma_qp_index_offset[0],
1743 h->pps.chroma_qp_index_offset[1]) +
1744 6 * (h->sps.bit_depth_luma - 8);
1745
1746 h0->last_slice_type = slice_type;
1747 h->slice_num = ++h0->current_slice;
1748 if (h->slice_num >= MAX_SLICES) {
1749 av_log(h->avctx, AV_LOG_ERROR,
1750 "Too many slices, increase MAX_SLICES and recompile\n");
1751 }
1752
1753 for (j = 0; j < 2; j++) {
1754 int id_list[16];
1755 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1756 for (i = 0; i < 16; i++) {
1757 id_list[i] = 60;
1758 if (j < h->list_count && i < h->ref_count[j] &&
1759 h->ref_list[j][i].f.buf[0]) {
1760 int k;
1761 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1762 for (k = 0; k < h->short_ref_count; k++)
1763 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1764 id_list[i] = k;
1765 break;
1766 }
1767 for (k = 0; k < h->long_ref_count; k++)
1768 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1769 id_list[i] = h->short_ref_count + k;
1770 break;
1771 }
1772 }
1773 }
1774
1775 ref2frm[0] =
1776 ref2frm[1] = -1;
1777 for (i = 0; i < 16; i++)
1778 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1779 ref2frm[18 + 0] =
1780 ref2frm[18 + 1] = -1;
1781 for (i = 16; i < 48; i++)
1782 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1783 (h->ref_list[j][i].reference & 3);
1784 }
1785
1786 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1787 av_log(h->avctx, AV_LOG_DEBUG,
1788 "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",
1789 h->slice_num,
1790 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1791 first_mb_in_slice,
1792 av_get_picture_type_char(h->slice_type),
1793 h->slice_type_fixed ? " fix" : "",
1794 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1795 pps_id, h->frame_num,
1796 h->cur_pic_ptr->field_poc[0],
1797 h->cur_pic_ptr->field_poc[1],
1798 h->ref_count[0], h->ref_count[1],
1799 h->qscale,
1800 h->deblocking_filter,
1801 h->slice_alpha_c0_offset, h->slice_beta_offset,
1802 h->use_weight,
1803 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1804 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1805 }
1806
1807 return 0;
1808}
1809
1810int ff_h264_get_slice_type(const H264Context *h)
1811{
1812 switch (h->slice_type) {
1813 case AV_PICTURE_TYPE_P:
1814 return 0;
1815 case AV_PICTURE_TYPE_B:
1816 return 1;
1817 case AV_PICTURE_TYPE_I:
1818 return 2;
1819 case AV_PICTURE_TYPE_SP:
1820 return 3;
1821 case AV_PICTURE_TYPE_SI:
1822 return 4;
1823 default:
1824 return AVERROR_INVALIDDATA;
1825 }
1826}
1827
1828static av_always_inline void fill_filter_caches_inter(H264Context *h,
1829 int mb_type, int top_xy,
1830 int left_xy[LEFT_MBS],
1831 int top_type,
1832 int left_type[LEFT_MBS],
1833 int mb_xy, int list)
1834{
1835 int b_stride = h->b_stride;
1836 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
1837 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
1838 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1839 if (USES_LIST(top_type, list)) {
1840 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1841 const int b8_xy = 4 * top_xy + 2;
1842 int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1843 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1844 ref_cache[0 - 1 * 8] =
1845 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1846 ref_cache[2 - 1 * 8] =
1847 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1848 } else {
1849 AV_ZERO128(mv_dst - 1 * 8);
1850 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1851 }
1852
1853 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1854 if (USES_LIST(left_type[LTOP], list)) {
1855 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1856 const int b8_xy = 4 * left_xy[LTOP] + 1;
1857 int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1858 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1859 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1860 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1861 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1862 ref_cache[-1 + 0] =
1863 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1864 ref_cache[-1 + 16] =
1865 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1866 } else {
1867 AV_ZERO32(mv_dst - 1 + 0);
1868 AV_ZERO32(mv_dst - 1 + 8);
1869 AV_ZERO32(mv_dst - 1 + 16);
1870 AV_ZERO32(mv_dst - 1 + 24);
1871 ref_cache[-1 + 0] =
1872 ref_cache[-1 + 8] =
1873 ref_cache[-1 + 16] =
1874 ref_cache[-1 + 24] = LIST_NOT_USED;
1875 }
1876 }
1877 }
1878
1879 if (!USES_LIST(mb_type, list)) {
1880 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1881 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1882 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1883 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1884 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1885 return;
1886 }
1887
1888 {
1889 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1890 int (*ref2frm)[64] = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1891 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1892 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1893 AV_WN32A(&ref_cache[0 * 8], ref01);
1894 AV_WN32A(&ref_cache[1 * 8], ref01);
1895 AV_WN32A(&ref_cache[2 * 8], ref23);
1896 AV_WN32A(&ref_cache[3 * 8], ref23);
1897 }
1898
1899 {
1900 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
1901 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1902 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1903 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1904 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1905 }
1906}
1907
1908/**
1909 *
1910 * @return non zero if the loop filter can be skipped
1911 */
1912static int fill_filter_caches(H264Context *h, int mb_type)
1913{
1914 const int mb_xy = h->mb_xy;
1915 int top_xy, left_xy[LEFT_MBS];
1916 int top_type, left_type[LEFT_MBS];
1917 uint8_t *nnz;
1918 uint8_t *nnz_cache;
1919
1920 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
1921
1922 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1923 * stuff, I can't imagine that these complex rules are worth it. */
1924
1925 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1926 if (FRAME_MBAFF(h)) {
1927 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1928 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1929 if (h->mb_y & 1) {
1930 if (left_mb_field_flag != curr_mb_field_flag)
1931 left_xy[LTOP] -= h->mb_stride;
1932 } else {
1933 if (curr_mb_field_flag)
1934 top_xy += h->mb_stride &
1935 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1936 if (left_mb_field_flag != curr_mb_field_flag)
1937 left_xy[LBOT] += h->mb_stride;
1938 }
1939 }
1940
1941 h->top_mb_xy = top_xy;
1942 h->left_mb_xy[LTOP] = left_xy[LTOP];
1943 h->left_mb_xy[LBOT] = left_xy[LBOT];
1944 {
1945 /* For sufficiently low qp, filtering wouldn't do anything.
1946 * This is a conservative estimate: could also check beta_offset
1947 * and more accurate chroma_qp. */
1948 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1949 int qp = h->cur_pic.qscale_table[mb_xy];
1950 if (qp <= qp_thresh &&
1951 (left_xy[LTOP] < 0 ||
1952 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1953 (top_xy < 0 ||
1954 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1955 if (!FRAME_MBAFF(h))
1956 return 1;
1957 if ((left_xy[LTOP] < 0 ||
1958 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1959 (top_xy < h->mb_stride ||
1960 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1961 return 1;
1962 }
1963 }
1964
1965 top_type = h->cur_pic.mb_type[top_xy];
1966 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1967 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1968 if (h->deblocking_filter == 2) {
1969 if (h->slice_table[top_xy] != h->slice_num)
1970 top_type = 0;
1971 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
1972 left_type[LTOP] = left_type[LBOT] = 0;
1973 } else {
1974 if (h->slice_table[top_xy] == 0xFFFF)
1975 top_type = 0;
1976 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1977 left_type[LTOP] = left_type[LBOT] = 0;
1978 }
1979 h->top_type = top_type;
1980 h->left_type[LTOP] = left_type[LTOP];
1981 h->left_type[LBOT] = left_type[LBOT];
1982
1983 if (IS_INTRA(mb_type))
1984 return 0;
1985
1986 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1987 top_type, left_type, mb_xy, 0);
1988 if (h->list_count == 2)
1989 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1990 top_type, left_type, mb_xy, 1);
1991
1992 nnz = h->non_zero_count[mb_xy];
1993 nnz_cache = h->non_zero_count_cache;
1994 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1995 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1996 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1997 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1998 h->cbp = h->cbp_table[mb_xy];
1999
2000 if (top_type) {
2001 nnz = h->non_zero_count[top_xy];
2002 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2003 }
2004
2005 if (left_type[LTOP]) {
2006 nnz = h->non_zero_count[left_xy[LTOP]];
2007 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2008 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2009 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2010 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2011 }
2012
2013 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2014 * from what the loop filter needs */
2015 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2016 if (IS_8x8DCT(top_type)) {
2017 nnz_cache[4 + 8 * 0] =
2018 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2019 nnz_cache[6 + 8 * 0] =
2020 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2021 }
2022 if (IS_8x8DCT(left_type[LTOP])) {
2023 nnz_cache[3 + 8 * 1] =
2024 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2025 }
2026 if (IS_8x8DCT(left_type[LBOT])) {
2027 nnz_cache[3 + 8 * 3] =
2028 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2029 }
2030
2031 if (IS_8x8DCT(mb_type)) {
2032 nnz_cache[scan8[0]] =
2033 nnz_cache[scan8[1]] =
2034 nnz_cache[scan8[2]] =
2035 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2036
2037 nnz_cache[scan8[0 + 4]] =
2038 nnz_cache[scan8[1 + 4]] =
2039 nnz_cache[scan8[2 + 4]] =
2040 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2041
2042 nnz_cache[scan8[0 + 8]] =
2043 nnz_cache[scan8[1 + 8]] =
2044 nnz_cache[scan8[2 + 8]] =
2045 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2046
2047 nnz_cache[scan8[0 + 12]] =
2048 nnz_cache[scan8[1 + 12]] =
2049 nnz_cache[scan8[2 + 12]] =
2050 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2051 }
2052 }
2053
2054 return 0;
2055}
2056
2057static void loop_filter(H264Context *h, int start_x, int end_x)
2058{
2059 uint8_t *dest_y, *dest_cb, *dest_cr;
2060 int linesize, uvlinesize, mb_x, mb_y;
2061 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2062 const int old_slice_type = h->slice_type;
2063 const int pixel_shift = h->pixel_shift;
2064 const int block_h = 16 >> h->chroma_y_shift;
2065
2066 if (h->deblocking_filter) {
2067 for (mb_x = start_x; mb_x < end_x; mb_x++)
2068 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2069 int mb_xy, mb_type;
2070 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2071 h->slice_num = h->slice_table[mb_xy];
2072 mb_type = h->cur_pic.mb_type[mb_xy];
2073 h->list_count = h->list_counts[mb_xy];
2074
2075 if (FRAME_MBAFF(h))
2076 h->mb_mbaff =
2077 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2078
2079 h->mb_x = mb_x;
2080 h->mb_y = mb_y;
2081 dest_y = h->cur_pic.f.data[0] +
2082 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2083 dest_cb = h->cur_pic.f.data[1] +
2084 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2085 mb_y * h->uvlinesize * block_h;
2086 dest_cr = h->cur_pic.f.data[2] +
2087 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2088 mb_y * h->uvlinesize * block_h;
2089 // FIXME simplify above
2090
2091 if (MB_FIELD(h)) {
2092 linesize = h->mb_linesize = h->linesize * 2;
2093 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2094 if (mb_y & 1) { // FIXME move out of this function?
2095 dest_y -= h->linesize * 15;
2096 dest_cb -= h->uvlinesize * (block_h - 1);
2097 dest_cr -= h->uvlinesize * (block_h - 1);
2098 }
2099 } else {
2100 linesize = h->mb_linesize = h->linesize;
2101 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2102 }
2103 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2104 uvlinesize, 0);
2105 if (fill_filter_caches(h, mb_type))
2106 continue;
2107 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2108 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2109
2110 if (FRAME_MBAFF(h)) {
2111 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2112 linesize, uvlinesize);
2113 } else {
2114 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2115 dest_cr, linesize, uvlinesize);
2116 }
2117 }
2118 }
2119 h->slice_type = old_slice_type;
2120 h->mb_x = end_x;
2121 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2122 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2123 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2124}
2125
2126static void predict_field_decoding_flag(H264Context *h)
2127{
2128 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2129 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2130 h->cur_pic.mb_type[mb_xy - 1] :
2131 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2132 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2133 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2134}
2135
2136/**
2137 * Draw edges and report progress for the last MB row.
2138 */
2139static void decode_finish_row(H264Context *h)
2140{
2141 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2142 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2143 int height = 16 << FRAME_MBAFF(h);
2144 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2145
2146 if (h->deblocking_filter) {
2147 if ((top + height) >= pic_height)
2148 height += deblock_border;
2149 top -= deblock_border;
2150 }
2151
2152 if (top >= pic_height || (top + height) < 0)
2153 return;
2154
2155 height = FFMIN(height, pic_height - top);
2156 if (top < 0) {
2157 height = top + height;
2158 top = 0;
2159 }
2160
2161 ff_h264_draw_horiz_band(h, top, height);
2162
2163 if (h->droppable)
2164 return;
2165
2166 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2167 h->picture_structure == PICT_BOTTOM_FIELD);
2168}
2169
2170static void er_add_slice(H264Context *h, int startx, int starty,
2171 int endx, int endy, int status)
2172{
2173#if CONFIG_ERROR_RESILIENCE
2174 ERContext *er = &h->er;
2175
2176 er->ref_count = h->ref_count[0];
2177 ff_er_add_slice(er, startx, starty, endx, endy, status);
2178#endif
2179}
2180
2181static int decode_slice(struct AVCodecContext *avctx, void *arg)
2182{
2183 H264Context *h = *(void **)arg;
2184 int lf_x_start = h->mb_x;
2185
2186 h->mb_skip_run = -1;
2187
2188 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2189 avctx->codec_id != AV_CODEC_ID_H264 ||
2190 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2191
2192 if (h->pps.cabac) {
2193 /* realign */
2194 align_get_bits(&h->gb);
2195
2196 /* init cabac */
2197 ff_init_cabac_decoder(&h->cabac,
2198 h->gb.buffer + get_bits_count(&h->gb) / 8,
2199 (get_bits_left(&h->gb) + 7) / 8);
2200
2201 ff_h264_init_cabac_states(h);
2202
2203 for (;;) {
2204 // START_TIMER
2205 int ret = ff_h264_decode_mb_cabac(h);
2206 int eos;
2207 // STOP_TIMER("decode_mb_cabac")
2208
2209 if (ret >= 0)
2210 ff_h264_hl_decode_mb(h);
2211
2212 // FIXME optimal? or let mb_decode decode 16x32 ?
2213 if (ret >= 0 && FRAME_MBAFF(h)) {
2214 h->mb_y++;
2215
2216 ret = ff_h264_decode_mb_cabac(h);
2217
2218 if (ret >= 0)
2219 ff_h264_hl_decode_mb(h);
2220 h->mb_y--;
2221 }
2222 eos = get_cabac_terminate(&h->cabac);
2223
2224 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2225 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2226 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2227 h->mb_y, ER_MB_END);
2228 if (h->mb_x >= lf_x_start)
2229 loop_filter(h, lf_x_start, h->mb_x + 1);
2230 return 0;
2231 }
2232 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2233 av_log(h->avctx, AV_LOG_ERROR,
2234 "error while decoding MB %d %d, bytestream %td\n",
2235 h->mb_x, h->mb_y,
2236 h->cabac.bytestream_end - h->cabac.bytestream);
2237 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2238 h->mb_y, ER_MB_ERROR);
2239 return AVERROR_INVALIDDATA;
2240 }
2241
2242 if (++h->mb_x >= h->mb_width) {
2243 loop_filter(h, lf_x_start, h->mb_x);
2244 h->mb_x = lf_x_start = 0;
2245 decode_finish_row(h);
2246 ++h->mb_y;
2247 if (FIELD_OR_MBAFF_PICTURE(h)) {
2248 ++h->mb_y;
2249 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2250 predict_field_decoding_flag(h);
2251 }
2252 }
2253
2254 if (eos || h->mb_y >= h->mb_height) {
2255 tprintf(h->avctx, "slice end %d %d\n",
2256 get_bits_count(&h->gb), h->gb.size_in_bits);
2257 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2258 h->mb_y, ER_MB_END);
2259 if (h->mb_x > lf_x_start)
2260 loop_filter(h, lf_x_start, h->mb_x);
2261 return 0;
2262 }
2263 }
2264 } else {
2265 for (;;) {
2266 int ret = ff_h264_decode_mb_cavlc(h);
2267
2268 if (ret >= 0)
2269 ff_h264_hl_decode_mb(h);
2270
2271 // FIXME optimal? or let mb_decode decode 16x32 ?
2272 if (ret >= 0 && FRAME_MBAFF(h)) {
2273 h->mb_y++;
2274 ret = ff_h264_decode_mb_cavlc(h);
2275
2276 if (ret >= 0)
2277 ff_h264_hl_decode_mb(h);
2278 h->mb_y--;
2279 }
2280
2281 if (ret < 0) {
2282 av_log(h->avctx, AV_LOG_ERROR,
2283 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2284 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2285 h->mb_y, ER_MB_ERROR);
2286 return ret;
2287 }
2288
2289 if (++h->mb_x >= h->mb_width) {
2290 loop_filter(h, lf_x_start, h->mb_x);
2291 h->mb_x = lf_x_start = 0;
2292 decode_finish_row(h);
2293 ++h->mb_y;
2294 if (FIELD_OR_MBAFF_PICTURE(h)) {
2295 ++h->mb_y;
2296 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2297 predict_field_decoding_flag(h);
2298 }
2299 if (h->mb_y >= h->mb_height) {
2300 tprintf(h->avctx, "slice end %d %d\n",
2301 get_bits_count(&h->gb), h->gb.size_in_bits);
2302
2303 if (get_bits_left(&h->gb) == 0) {
2304 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2305 h->mb_x - 1, h->mb_y,
2306 ER_MB_END);
2307
2308 return 0;
2309 } else {
2310 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2311 h->mb_x - 1, h->mb_y,
2312 ER_MB_END);
2313
2314 return AVERROR_INVALIDDATA;
2315 }
2316 }
2317 }
2318
2319 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2320 tprintf(h->avctx, "slice end %d %d\n",
2321 get_bits_count(&h->gb), h->gb.size_in_bits);
2322
2323 if (get_bits_left(&h->gb) == 0) {
2324 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2325 h->mb_x - 1, h->mb_y,
2326 ER_MB_END);
2327 if (h->mb_x > lf_x_start)
2328 loop_filter(h, lf_x_start, h->mb_x);
2329
2330 return 0;
2331 } else {
2332 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2333 h->mb_y, ER_MB_ERROR);
2334
2335 return AVERROR_INVALIDDATA;
2336 }
2337 }
2338 }
2339 }
2340}
2341
2342/**
2343 * Call decode_slice() for each context.
2344 *
2345 * @param h h264 master context
2346 * @param context_count number of contexts to execute
2347 */
2348int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2349{
2350 AVCodecContext *const avctx = h->avctx;
2351 H264Context *hx;
2352 int i;
2353
2354 if (h->mb_y >= h->mb_height) {
2355 av_log(h->avctx, AV_LOG_ERROR,
2356 "Input contains more MB rows than the frame height.\n");
2357 return AVERROR_INVALIDDATA;
2358 }
2359
2360 if (h->avctx->hwaccel)
2361 return 0;
2362 if (context_count == 1) {
2363 return decode_slice(avctx, &h);
2364 } else {
2365 for (i = 1; i < context_count; i++) {
2366 hx = h->thread_context[i];
2367 hx->er.error_count = 0;
2368 }
2369
2370 avctx->execute(avctx, decode_slice, h->thread_context,
2371 NULL, context_count, sizeof(void *));
2372
2373 /* pull back stuff from slices to master context */
2374 hx = h->thread_context[context_count - 1];
2375 h->mb_x = hx->mb_x;
2376 h->mb_y = hx->mb_y;
2377 h->droppable = hx->droppable;
2378 h->picture_structure = hx->picture_structure;
2379 for (i = 1; i < context_count; i++)
2380 h->er.error_count += h->thread_context[i]->er.error_count;
2381 }
2382
2383 return 0;
2384}