lavc: set AVCodecContext.hwaccel in ff_get_format()
[libav.git] / libavcodec / h264_slice.c
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
48 static const uint8_t rem6[QP_MAX_NUM + 1] = {
49 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
50 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
51 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
52 };
53
54 static 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
60 static 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
67 static 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
86 static 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)]
106 static 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
125 static 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
134 static 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
138 static 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
147 static 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
164 static 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
182 static 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
195 static 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
216 static 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
243 static 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;
294 fail:
295 ff_h264_unref_picture(h, pic);
296 return (ret < 0) ? ret : AVERROR(ENOMEM);
297 }
298
299 static 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
308 static 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
328 static 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
355 static 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
381 void 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 */
401 static 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
430 static 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
445 static 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
469 static int h264_slice_header_init(H264Context *h, int reinit);
470
471 int 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
667 static 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 /* We mark the current picture as non-reference after allocating it, so
734 * that if we break out due to an error it can be released automatically
735 * in the next ff_MPV_frame_start().
736 */
737 h->cur_pic_ptr->reference = 0;
738
739 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
740
741 h->next_output_pic = NULL;
742
743 assert(h->cur_pic_ptr->long_ref == 0);
744
745 return 0;
746 }
747
748 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
749 uint8_t *src_cb, uint8_t *src_cr,
750 int linesize, int uvlinesize,
751 int simple)
752 {
753 uint8_t *top_border;
754 int top_idx = 1;
755 const int pixel_shift = h->pixel_shift;
756 int chroma444 = CHROMA444(h);
757 int chroma422 = CHROMA422(h);
758
759 src_y -= linesize;
760 src_cb -= uvlinesize;
761 src_cr -= uvlinesize;
762
763 if (!simple && FRAME_MBAFF(h)) {
764 if (h->mb_y & 1) {
765 if (!MB_MBAFF(h)) {
766 top_border = h->top_borders[0][h->mb_x];
767 AV_COPY128(top_border, src_y + 15 * linesize);
768 if (pixel_shift)
769 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
770 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
771 if (chroma444) {
772 if (pixel_shift) {
773 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
774 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
775 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
776 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
777 } else {
778 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
779 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
780 }
781 } else if (chroma422) {
782 if (pixel_shift) {
783 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
784 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
785 } else {
786 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
787 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
788 }
789 } else {
790 if (pixel_shift) {
791 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
792 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
793 } else {
794 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
795 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
796 }
797 }
798 }
799 }
800 } else if (MB_MBAFF(h)) {
801 top_idx = 0;
802 } else
803 return;
804 }
805
806 top_border = h->top_borders[top_idx][h->mb_x];
807 /* There are two lines saved, the line above the top macroblock
808 * of a pair, and the line above the bottom macroblock. */
809 AV_COPY128(top_border, src_y + 16 * linesize);
810 if (pixel_shift)
811 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
812
813 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
814 if (chroma444) {
815 if (pixel_shift) {
816 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
817 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
818 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
819 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
820 } else {
821 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
822 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
823 }
824 } else if (chroma422) {
825 if (pixel_shift) {
826 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
827 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
828 } else {
829 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
830 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
831 }
832 } else {
833 if (pixel_shift) {
834 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
835 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
836 } else {
837 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
838 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
839 }
840 }
841 }
842 }
843
844 /**
845 * Initialize implicit_weight table.
846 * @param field 0/1 initialize the weight for interlaced MBAFF
847 * -1 initializes the rest
848 */
849 static void implicit_weight_table(H264Context *h, int field)
850 {
851 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
852
853 for (i = 0; i < 2; i++) {
854 h->luma_weight_flag[i] = 0;
855 h->chroma_weight_flag[i] = 0;
856 }
857
858 if (field < 0) {
859 if (h->picture_structure == PICT_FRAME) {
860 cur_poc = h->cur_pic_ptr->poc;
861 } else {
862 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
863 }
864 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
865 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
866 h->use_weight = 0;
867 h->use_weight_chroma = 0;
868 return;
869 }
870 ref_start = 0;
871 ref_count0 = h->ref_count[0];
872 ref_count1 = h->ref_count[1];
873 } else {
874 cur_poc = h->cur_pic_ptr->field_poc[field];
875 ref_start = 16;
876 ref_count0 = 16 + 2 * h->ref_count[0];
877 ref_count1 = 16 + 2 * h->ref_count[1];
878 }
879
880 h->use_weight = 2;
881 h->use_weight_chroma = 2;
882 h->luma_log2_weight_denom = 5;
883 h->chroma_log2_weight_denom = 5;
884
885 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
886 int poc0 = h->ref_list[0][ref0].poc;
887 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
888 int w = 32;
889 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
890 int poc1 = h->ref_list[1][ref1].poc;
891 int td = av_clip(poc1 - poc0, -128, 127);
892 if (td) {
893 int tb = av_clip(cur_poc - poc0, -128, 127);
894 int tx = (16384 + (FFABS(td) >> 1)) / td;
895 int dist_scale_factor = (tb * tx + 32) >> 8;
896 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
897 w = 64 - dist_scale_factor;
898 }
899 }
900 if (field < 0) {
901 h->implicit_weight[ref0][ref1][0] =
902 h->implicit_weight[ref0][ref1][1] = w;
903 } else {
904 h->implicit_weight[ref0][ref1][field] = w;
905 }
906 }
907 }
908 }
909
910 /**
911 * initialize scan tables
912 */
913 static void init_scan_tables(H264Context *h)
914 {
915 int i;
916 for (i = 0; i < 16; i++) {
917 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
918 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
919 h->field_scan[i] = TRANSPOSE(field_scan[i]);
920 #undef TRANSPOSE
921 }
922 for (i = 0; i < 64; i++) {
923 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
924 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
925 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
926 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
927 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
928 #undef TRANSPOSE
929 }
930 if (h->sps.transform_bypass) { // FIXME same ugly
931 h->zigzag_scan_q0 = zigzag_scan;
932 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
933 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
934 h->field_scan_q0 = field_scan;
935 h->field_scan8x8_q0 = field_scan8x8;
936 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
937 } else {
938 h->zigzag_scan_q0 = h->zigzag_scan;
939 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
940 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
941 h->field_scan_q0 = h->field_scan;
942 h->field_scan8x8_q0 = h->field_scan8x8;
943 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
944 }
945 }
946
947 /**
948 * Replicate H264 "master" context to thread contexts.
949 */
950 static int clone_slice(H264Context *dst, H264Context *src)
951 {
952 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
953 dst->cur_pic_ptr = src->cur_pic_ptr;
954 dst->cur_pic = src->cur_pic;
955 dst->linesize = src->linesize;
956 dst->uvlinesize = src->uvlinesize;
957 dst->first_field = src->first_field;
958
959 dst->prev_poc_msb = src->prev_poc_msb;
960 dst->prev_poc_lsb = src->prev_poc_lsb;
961 dst->prev_frame_num_offset = src->prev_frame_num_offset;
962 dst->prev_frame_num = src->prev_frame_num;
963 dst->short_ref_count = src->short_ref_count;
964
965 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
966 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
967 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
968
969 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
970 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
971
972 return 0;
973 }
974
975 static enum AVPixelFormat get_pixel_format(H264Context *h)
976 {
977 switch (h->sps.bit_depth_luma) {
978 case 9:
979 if (CHROMA444(h)) {
980 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
981 return AV_PIX_FMT_GBRP9;
982 } else
983 return AV_PIX_FMT_YUV444P9;
984 } else if (CHROMA422(h))
985 return AV_PIX_FMT_YUV422P9;
986 else
987 return AV_PIX_FMT_YUV420P9;
988 break;
989 case 10:
990 if (CHROMA444(h)) {
991 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
992 return AV_PIX_FMT_GBRP10;
993 } else
994 return AV_PIX_FMT_YUV444P10;
995 } else if (CHROMA422(h))
996 return AV_PIX_FMT_YUV422P10;
997 else
998 return AV_PIX_FMT_YUV420P10;
999 break;
1000 case 8:
1001 if (CHROMA444(h)) {
1002 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1003 return AV_PIX_FMT_GBRP;
1004 } else
1005 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
1006 : AV_PIX_FMT_YUV444P;
1007 } else if (CHROMA422(h)) {
1008 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
1009 : AV_PIX_FMT_YUV422P;
1010 } else {
1011 return ff_get_format(h->avctx, h->avctx->codec->pix_fmts ?
1012 h->avctx->codec->pix_fmts :
1013 h->avctx->color_range == AVCOL_RANGE_JPEG ?
1014 h264_hwaccel_pixfmt_list_jpeg_420 :
1015 h264_hwaccel_pixfmt_list_420);
1016 }
1017 break;
1018 default:
1019 av_log(h->avctx, AV_LOG_ERROR,
1020 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1021 return AVERROR_INVALIDDATA;
1022 }
1023 }
1024
1025 /* export coded and cropped frame dimensions to AVCodecContext */
1026 static int init_dimensions(H264Context *h)
1027 {
1028 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1029 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1030
1031 /* handle container cropping */
1032 if (!h->sps.crop &&
1033 FFALIGN(h->avctx->width, 16) == h->width &&
1034 FFALIGN(h->avctx->height, 16) == h->height) {
1035 width = h->avctx->width;
1036 height = h->avctx->height;
1037 }
1038
1039 if (width <= 0 || height <= 0) {
1040 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1041 width, height);
1042 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1043 return AVERROR_INVALIDDATA;
1044
1045 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1046 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
1047 h->sps.crop = 0;
1048
1049 width = h->width;
1050 height = h->height;
1051 }
1052
1053 h->avctx->coded_width = h->width;
1054 h->avctx->coded_height = h->height;
1055 h->avctx->width = width;
1056 h->avctx->height = height;
1057
1058 return 0;
1059 }
1060
1061 static int h264_slice_header_init(H264Context *h, int reinit)
1062 {
1063 int nb_slices = (HAVE_THREADS &&
1064 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1065 h->avctx->thread_count : 1;
1066 int i, ret;
1067
1068 h->avctx->sample_aspect_ratio = h->sps.sar;
1069 av_assert0(h->avctx->sample_aspect_ratio.den);
1070 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1071 &h->chroma_x_shift, &h->chroma_y_shift);
1072
1073 if (h->sps.timing_info_present_flag) {
1074 int64_t den = h->sps.time_scale;
1075 if (h->x264_build < 44U)
1076 den *= 2;
1077 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
1078 h->sps.num_units_in_tick, den, 1 << 30);
1079 }
1080
1081 if (reinit)
1082 ff_h264_free_tables(h, 0);
1083 h->first_field = 0;
1084 h->prev_interlaced_frame = 1;
1085
1086 init_scan_tables(h);
1087 ret = ff_h264_alloc_tables(h);
1088 if (ret < 0) {
1089 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1090 return ret;
1091 }
1092
1093 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1094 int max_slices;
1095 if (h->mb_height)
1096 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1097 else
1098 max_slices = H264_MAX_THREADS;
1099 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1100 " reducing to %d\n", nb_slices, max_slices);
1101 nb_slices = max_slices;
1102 }
1103 h->slice_context_count = nb_slices;
1104
1105 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1106 ret = ff_h264_context_init(h);
1107 if (ret < 0) {
1108 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1109 return ret;
1110 }
1111 } else {
1112 for (i = 1; i < h->slice_context_count; i++) {
1113 H264Context *c;
1114 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1115 if (!c)
1116 return AVERROR(ENOMEM);
1117 c->avctx = h->avctx;
1118 c->dsp = h->dsp;
1119 c->vdsp = h->vdsp;
1120 c->h264dsp = h->h264dsp;
1121 c->h264qpel = h->h264qpel;
1122 c->h264chroma = h->h264chroma;
1123 c->sps = h->sps;
1124 c->pps = h->pps;
1125 c->pixel_shift = h->pixel_shift;
1126 c->width = h->width;
1127 c->height = h->height;
1128 c->linesize = h->linesize;
1129 c->uvlinesize = h->uvlinesize;
1130 c->chroma_x_shift = h->chroma_x_shift;
1131 c->chroma_y_shift = h->chroma_y_shift;
1132 c->qscale = h->qscale;
1133 c->droppable = h->droppable;
1134 c->data_partitioning = h->data_partitioning;
1135 c->low_delay = h->low_delay;
1136 c->mb_width = h->mb_width;
1137 c->mb_height = h->mb_height;
1138 c->mb_stride = h->mb_stride;
1139 c->mb_num = h->mb_num;
1140 c->flags = h->flags;
1141 c->workaround_bugs = h->workaround_bugs;
1142 c->pict_type = h->pict_type;
1143
1144 init_scan_tables(c);
1145 clone_tables(c, h, i);
1146 c->context_initialized = 1;
1147 }
1148
1149 for (i = 0; i < h->slice_context_count; i++)
1150 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1151 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1152 return ret;
1153 }
1154 }
1155
1156 h->context_initialized = 1;
1157
1158 return 0;
1159 }
1160
1161 /**
1162 * Decode a slice header.
1163 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1164 *
1165 * @param h h264context
1166 * @param h0 h264 master context (differs from 'h' when doing sliced based
1167 * parallel decoding)
1168 *
1169 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1170 */
1171 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1172 {
1173 unsigned int first_mb_in_slice;
1174 unsigned int pps_id;
1175 int ret;
1176 unsigned int slice_type, tmp, i, j;
1177 int default_ref_list_done = 0;
1178 int last_pic_structure, last_pic_droppable;
1179 int needs_reinit = 0;
1180 int field_pic_flag, bottom_field_flag;
1181
1182 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1183 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1184
1185 first_mb_in_slice = get_ue_golomb(&h->gb);
1186
1187 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1188 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1189 ff_h264_field_end(h, 1);
1190 }
1191
1192 h0->current_slice = 0;
1193 if (!h0->first_field) {
1194 if (h->cur_pic_ptr && !h->droppable) {
1195 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1196 h->picture_structure == PICT_BOTTOM_FIELD);
1197 }
1198 h->cur_pic_ptr = NULL;
1199 }
1200 }
1201
1202 slice_type = get_ue_golomb_31(&h->gb);
1203 if (slice_type > 9) {
1204 av_log(h->avctx, AV_LOG_ERROR,
1205 "slice type %d too large at %d %d\n",
1206 slice_type, h->mb_x, h->mb_y);
1207 return AVERROR_INVALIDDATA;
1208 }
1209 if (slice_type > 4) {
1210 slice_type -= 5;
1211 h->slice_type_fixed = 1;
1212 } else
1213 h->slice_type_fixed = 0;
1214
1215 slice_type = golomb_to_pict_type[slice_type];
1216 if (slice_type == AV_PICTURE_TYPE_I ||
1217 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
1218 default_ref_list_done = 1;
1219 }
1220 h->slice_type = slice_type;
1221 h->slice_type_nos = slice_type & 3;
1222
1223 if (h->nal_unit_type == NAL_IDR_SLICE &&
1224 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1225 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1226 return AVERROR_INVALIDDATA;
1227 }
1228
1229 // to make a few old functions happy, it's wrong though
1230 h->pict_type = h->slice_type;
1231
1232 pps_id = get_ue_golomb(&h->gb);
1233 if (pps_id >= MAX_PPS_COUNT) {
1234 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1235 return AVERROR_INVALIDDATA;
1236 }
1237 if (!h0->pps_buffers[pps_id]) {
1238 av_log(h->avctx, AV_LOG_ERROR,
1239 "non-existing PPS %u referenced\n",
1240 pps_id);
1241 return AVERROR_INVALIDDATA;
1242 }
1243 h->pps = *h0->pps_buffers[pps_id];
1244
1245 if (!h0->sps_buffers[h->pps.sps_id]) {
1246 av_log(h->avctx, AV_LOG_ERROR,
1247 "non-existing SPS %u referenced\n",
1248 h->pps.sps_id);
1249 return AVERROR_INVALIDDATA;
1250 }
1251
1252 if (h->pps.sps_id != h->sps.sps_id ||
1253 h0->sps_buffers[h->pps.sps_id]->new) {
1254 h0->sps_buffers[h->pps.sps_id]->new = 0;
1255
1256 h->sps = *h0->sps_buffers[h->pps.sps_id];
1257
1258 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1259 h->chroma_format_idc != h->sps.chroma_format_idc) {
1260 h->bit_depth_luma = h->sps.bit_depth_luma;
1261 h->chroma_format_idc = h->sps.chroma_format_idc;
1262 needs_reinit = 1;
1263 }
1264 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1265 return ret;
1266 }
1267
1268 h->avctx->profile = ff_h264_get_profile(&h->sps);
1269 h->avctx->level = h->sps.level_idc;
1270 h->avctx->refs = h->sps.ref_frame_count;
1271
1272 if (h->mb_width != h->sps.mb_width ||
1273 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1274 needs_reinit = 1;
1275
1276 h->mb_width = h->sps.mb_width;
1277 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1278 h->mb_num = h->mb_width * h->mb_height;
1279 h->mb_stride = h->mb_width + 1;
1280
1281 h->b_stride = h->mb_width * 4;
1282
1283 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1284
1285 h->width = 16 * h->mb_width;
1286 h->height = 16 * h->mb_height;
1287
1288 ret = init_dimensions(h);
1289 if (ret < 0)
1290 return ret;
1291
1292 if (h->sps.video_signal_type_present_flag) {
1293 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1294 : AVCOL_RANGE_MPEG;
1295 if (h->sps.colour_description_present_flag) {
1296 if (h->avctx->colorspace != h->sps.colorspace)
1297 needs_reinit = 1;
1298 h->avctx->color_primaries = h->sps.color_primaries;
1299 h->avctx->color_trc = h->sps.color_trc;
1300 h->avctx->colorspace = h->sps.colorspace;
1301 }
1302 }
1303
1304 if (h->context_initialized && needs_reinit) {
1305 if (h != h0) {
1306 av_log(h->avctx, AV_LOG_ERROR,
1307 "changing width %d -> %d / height %d -> %d on "
1308 "slice %d\n",
1309 h->width, h->avctx->coded_width,
1310 h->height, h->avctx->coded_height,
1311 h0->current_slice + 1);
1312 return AVERROR_INVALIDDATA;
1313 }
1314
1315 ff_h264_flush_change(h);
1316
1317 if ((ret = get_pixel_format(h)) < 0)
1318 return ret;
1319 h->avctx->pix_fmt = ret;
1320
1321 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1322 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1323
1324 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1325 av_log(h->avctx, AV_LOG_ERROR,
1326 "h264_slice_header_init() failed\n");
1327 return ret;
1328 }
1329 }
1330 if (!h->context_initialized) {
1331 if (h != h0) {
1332 av_log(h->avctx, AV_LOG_ERROR,
1333 "Cannot (re-)initialize context during parallel decoding.\n");
1334 return AVERROR_PATCHWELCOME;
1335 }
1336
1337 if ((ret = get_pixel_format(h)) < 0)
1338 return ret;
1339 h->avctx->pix_fmt = ret;
1340
1341 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1342 av_log(h->avctx, AV_LOG_ERROR,
1343 "h264_slice_header_init() failed\n");
1344 return ret;
1345 }
1346 }
1347
1348 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1349 h->dequant_coeff_pps = pps_id;
1350 h264_init_dequant_tables(h);
1351 }
1352
1353 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1354
1355 h->mb_mbaff = 0;
1356 h->mb_aff_frame = 0;
1357 last_pic_structure = h0->picture_structure;
1358 last_pic_droppable = h0->droppable;
1359 h->droppable = h->nal_ref_idc == 0;
1360 if (h->sps.frame_mbs_only_flag) {
1361 h->picture_structure = PICT_FRAME;
1362 } else {
1363 field_pic_flag = get_bits1(&h->gb);
1364 if (field_pic_flag) {
1365 bottom_field_flag = get_bits1(&h->gb);
1366 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1367 } else {
1368 h->picture_structure = PICT_FRAME;
1369 h->mb_aff_frame = h->sps.mb_aff;
1370 }
1371 }
1372 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1373
1374 if (h0->current_slice != 0) {
1375 if (last_pic_structure != h->picture_structure ||
1376 last_pic_droppable != h->droppable) {
1377 av_log(h->avctx, AV_LOG_ERROR,
1378 "Changing field mode (%d -> %d) between slices is not allowed\n",
1379 last_pic_structure, h->picture_structure);
1380 h->picture_structure = last_pic_structure;
1381 h->droppable = last_pic_droppable;
1382 return AVERROR_INVALIDDATA;
1383 } else if (!h0->cur_pic_ptr) {
1384 av_log(h->avctx, AV_LOG_ERROR,
1385 "unset cur_pic_ptr on slice %d\n",
1386 h0->current_slice + 1);
1387 return AVERROR_INVALIDDATA;
1388 }
1389 } else {
1390 /* Shorten frame num gaps so we don't have to allocate reference
1391 * frames just to throw them away */
1392 if (h->frame_num != h->prev_frame_num) {
1393 int unwrap_prev_frame_num = h->prev_frame_num;
1394 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1395
1396 if (unwrap_prev_frame_num > h->frame_num)
1397 unwrap_prev_frame_num -= max_frame_num;
1398
1399 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1400 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1401 if (unwrap_prev_frame_num < 0)
1402 unwrap_prev_frame_num += max_frame_num;
1403
1404 h->prev_frame_num = unwrap_prev_frame_num;
1405 }
1406 }
1407
1408 /* See if we have a decoded first field looking for a pair...
1409 * Here, we're using that to see if we should mark previously
1410 * decode frames as "finished".
1411 * We have to do that before the "dummy" in-between frame allocation,
1412 * since that can modify s->current_picture_ptr. */
1413 if (h0->first_field) {
1414 assert(h0->cur_pic_ptr);
1415 assert(h0->cur_pic_ptr->f.buf[0]);
1416 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1417
1418 /* figure out if we have a complementary field pair */
1419 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1420 /* Previous field is unmatched. Don't display it, but let it
1421 * remain for reference if marked as such. */
1422 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1423 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1424 last_pic_structure == PICT_TOP_FIELD);
1425 }
1426 } else {
1427 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1428 /* This and previous field were reference, but had
1429 * different frame_nums. Consider this field first in
1430 * pair. Throw away previous field except for reference
1431 * purposes. */
1432 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1433 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1434 last_pic_structure == PICT_TOP_FIELD);
1435 }
1436 } else {
1437 /* Second field in complementary pair */
1438 if (!((last_pic_structure == PICT_TOP_FIELD &&
1439 h->picture_structure == PICT_BOTTOM_FIELD) ||
1440 (last_pic_structure == PICT_BOTTOM_FIELD &&
1441 h->picture_structure == PICT_TOP_FIELD))) {
1442 av_log(h->avctx, AV_LOG_ERROR,
1443 "Invalid field mode combination %d/%d\n",
1444 last_pic_structure, h->picture_structure);
1445 h->picture_structure = last_pic_structure;
1446 h->droppable = last_pic_droppable;
1447 return AVERROR_INVALIDDATA;
1448 } else if (last_pic_droppable != h->droppable) {
1449 avpriv_request_sample(h->avctx,
1450 "Found reference and non-reference fields in the same frame, which");
1451 h->picture_structure = last_pic_structure;
1452 h->droppable = last_pic_droppable;
1453 return AVERROR_PATCHWELCOME;
1454 }
1455 }
1456 }
1457 }
1458
1459 while (h->frame_num != h->prev_frame_num &&
1460 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1461 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1462 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1463 h->frame_num, h->prev_frame_num);
1464 ret = h264_frame_start(h);
1465 if (ret < 0) {
1466 h0->first_field = 0;
1467 return ret;
1468 }
1469
1470 h->prev_frame_num++;
1471 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1472 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1473 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1474 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1475 ret = ff_generate_sliding_window_mmcos(h, 1);
1476 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1477 return ret;
1478 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1479 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1480 return ret;
1481 /* Error concealment: If a ref is missing, copy the previous ref
1482 * in its place.
1483 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1484 * many assumptions about there being no actual duplicates.
1485 * FIXME: This does not copy padding for out-of-frame motion
1486 * vectors. Given we are concealing a lost frame, this probably
1487 * is not noticeable by comparison, but it should be fixed. */
1488 if (h->short_ref_count) {
1489 if (prev) {
1490 av_image_copy(h->short_ref[0]->f.data,
1491 h->short_ref[0]->f.linesize,
1492 (const uint8_t **)prev->f.data,
1493 prev->f.linesize,
1494 h->avctx->pix_fmt,
1495 h->mb_width * 16,
1496 h->mb_height * 16);
1497 h->short_ref[0]->poc = prev->poc + 2;
1498 }
1499 h->short_ref[0]->frame_num = h->prev_frame_num;
1500 }
1501 }
1502
1503 /* See if we have a decoded first field looking for a pair...
1504 * We're using that to see whether to continue decoding in that
1505 * frame, or to allocate a new one. */
1506 if (h0->first_field) {
1507 assert(h0->cur_pic_ptr);
1508 assert(h0->cur_pic_ptr->f.buf[0]);
1509 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1510
1511 /* figure out if we have a complementary field pair */
1512 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1513 /* Previous field is unmatched. Don't display it, but let it
1514 * remain for reference if marked as such. */
1515 h0->cur_pic_ptr = NULL;
1516 h0->first_field = FIELD_PICTURE(h);
1517 } else {
1518 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1519 /* This and the previous field had different frame_nums.
1520 * Consider this field first in pair. Throw away previous
1521 * one except for reference purposes. */
1522 h0->first_field = 1;
1523 h0->cur_pic_ptr = NULL;
1524 } else {
1525 /* Second field in complementary pair */
1526 h0->first_field = 0;
1527 }
1528 }
1529 } else {
1530 /* Frame or first field in a potentially complementary pair */
1531 h0->first_field = FIELD_PICTURE(h);
1532 }
1533
1534 if (!FIELD_PICTURE(h) || h0->first_field) {
1535 if (h264_frame_start(h) < 0) {
1536 h0->first_field = 0;
1537 return AVERROR_INVALIDDATA;
1538 }
1539 } else {
1540 release_unused_pictures(h, 0);
1541 }
1542 }
1543 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1544 return ret;
1545
1546 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1547
1548 assert(h->mb_num == h->mb_width * h->mb_height);
1549 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1550 first_mb_in_slice >= h->mb_num) {
1551 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1552 return AVERROR_INVALIDDATA;
1553 }
1554 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1555 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1556 FIELD_OR_MBAFF_PICTURE(h);
1557 if (h->picture_structure == PICT_BOTTOM_FIELD)
1558 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1559 assert(h->mb_y < h->mb_height);
1560
1561 if (h->picture_structure == PICT_FRAME) {
1562 h->curr_pic_num = h->frame_num;
1563 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1564 } else {
1565 h->curr_pic_num = 2 * h->frame_num + 1;
1566 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1567 }
1568
1569 if (h->nal_unit_type == NAL_IDR_SLICE)
1570 get_ue_golomb(&h->gb); /* idr_pic_id */
1571
1572 if (h->sps.poc_type == 0) {
1573 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1574
1575 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1576 h->delta_poc_bottom = get_se_golomb(&h->gb);
1577 }
1578
1579 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1580 h->delta_poc[0] = get_se_golomb(&h->gb);
1581
1582 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1583 h->delta_poc[1] = get_se_golomb(&h->gb);
1584 }
1585
1586 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1587
1588 if (h->pps.redundant_pic_cnt_present)
1589 h->redundant_pic_count = get_ue_golomb(&h->gb);
1590
1591 ret = ff_set_ref_count(h);
1592 if (ret < 0)
1593 return ret;
1594 else if (ret == 1)
1595 default_ref_list_done = 0;
1596
1597 if (!default_ref_list_done)
1598 ff_h264_fill_default_ref_list(h);
1599
1600 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1601 ret = ff_h264_decode_ref_pic_list_reordering(h);
1602 if (ret < 0) {
1603 h->ref_count[1] = h->ref_count[0] = 0;
1604 return ret;
1605 }
1606 }
1607
1608 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1609 (h->pps.weighted_bipred_idc == 1 &&
1610 h->slice_type_nos == AV_PICTURE_TYPE_B))
1611 ff_pred_weight_table(h);
1612 else if (h->pps.weighted_bipred_idc == 2 &&
1613 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1614 implicit_weight_table(h, -1);
1615 } else {
1616 h->use_weight = 0;
1617 for (i = 0; i < 2; i++) {
1618 h->luma_weight_flag[i] = 0;
1619 h->chroma_weight_flag[i] = 0;
1620 }
1621 }
1622
1623 // If frame-mt is enabled, only update mmco tables for the first slice
1624 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1625 // or h->mmco, which will cause ref list mix-ups and decoding errors
1626 // further down the line. This may break decoding if the first slice is
1627 // corrupt, thus we only do this if frame-mt is enabled.
1628 if (h->nal_ref_idc) {
1629 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1630 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1631 h0->current_slice == 0);
1632 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1633 return AVERROR_INVALIDDATA;
1634 }
1635
1636 if (FRAME_MBAFF(h)) {
1637 ff_h264_fill_mbaff_ref_list(h);
1638
1639 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1640 implicit_weight_table(h, 0);
1641 implicit_weight_table(h, 1);
1642 }
1643 }
1644
1645 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1646 ff_h264_direct_dist_scale_factor(h);
1647 ff_h264_direct_ref_list_init(h);
1648
1649 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1650 tmp = get_ue_golomb_31(&h->gb);
1651 if (tmp > 2) {
1652 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1653 return AVERROR_INVALIDDATA;
1654 }
1655 h->cabac_init_idc = tmp;
1656 }
1657
1658 h->last_qscale_diff = 0;
1659 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1660 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1661 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1662 return AVERROR_INVALIDDATA;
1663 }
1664 h->qscale = tmp;
1665 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1666 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1667 // FIXME qscale / qp ... stuff
1668 if (h->slice_type == AV_PICTURE_TYPE_SP)
1669 get_bits1(&h->gb); /* sp_for_switch_flag */
1670 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1671 h->slice_type == AV_PICTURE_TYPE_SI)
1672 get_se_golomb(&h->gb); /* slice_qs_delta */
1673
1674 h->deblocking_filter = 1;
1675 h->slice_alpha_c0_offset = 0;
1676 h->slice_beta_offset = 0;
1677 if (h->pps.deblocking_filter_parameters_present) {
1678 tmp = get_ue_golomb_31(&h->gb);
1679 if (tmp > 2) {
1680 av_log(h->avctx, AV_LOG_ERROR,
1681 "deblocking_filter_idc %u out of range\n", tmp);
1682 return AVERROR_INVALIDDATA;
1683 }
1684 h->deblocking_filter = tmp;
1685 if (h->deblocking_filter < 2)
1686 h->deblocking_filter ^= 1; // 1<->0
1687
1688 if (h->deblocking_filter) {
1689 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1690 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1691 if (h->slice_alpha_c0_offset > 12 ||
1692 h->slice_alpha_c0_offset < -12 ||
1693 h->slice_beta_offset > 12 ||
1694 h->slice_beta_offset < -12) {
1695 av_log(h->avctx, AV_LOG_ERROR,
1696 "deblocking filter parameters %d %d out of range\n",
1697 h->slice_alpha_c0_offset, h->slice_beta_offset);
1698 return AVERROR_INVALIDDATA;
1699 }
1700 }
1701 }
1702
1703 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1704 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1705 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1706 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1707 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1708 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1709 h->nal_ref_idc == 0))
1710 h->deblocking_filter = 0;
1711
1712 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1713 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1714 /* Cheat slightly for speed:
1715 * Do not bother to deblock across slices. */
1716 h->deblocking_filter = 2;
1717 } else {
1718 h0->max_contexts = 1;
1719 if (!h0->single_decode_warning) {
1720 av_log(h->avctx, AV_LOG_INFO,
1721 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1722 h0->single_decode_warning = 1;
1723 }
1724 if (h != h0) {
1725 av_log(h->avctx, AV_LOG_ERROR,
1726 "Deblocking switched inside frame.\n");
1727 return 1;
1728 }
1729 }
1730 }
1731 h->qp_thresh = 15 -
1732 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1733 FFMAX3(0,
1734 h->pps.chroma_qp_index_offset[0],
1735 h->pps.chroma_qp_index_offset[1]) +
1736 6 * (h->sps.bit_depth_luma - 8);
1737
1738 h0->last_slice_type = slice_type;
1739 h->slice_num = ++h0->current_slice;
1740 if (h->slice_num >= MAX_SLICES) {
1741 av_log(h->avctx, AV_LOG_ERROR,
1742 "Too many slices, increase MAX_SLICES and recompile\n");
1743 }
1744
1745 for (j = 0; j < 2; j++) {
1746 int id_list[16];
1747 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1748 for (i = 0; i < 16; i++) {
1749 id_list[i] = 60;
1750 if (j < h->list_count && i < h->ref_count[j] &&
1751 h->ref_list[j][i].f.buf[0]) {
1752 int k;
1753 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1754 for (k = 0; k < h->short_ref_count; k++)
1755 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1756 id_list[i] = k;
1757 break;
1758 }
1759 for (k = 0; k < h->long_ref_count; k++)
1760 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1761 id_list[i] = h->short_ref_count + k;
1762 break;
1763 }
1764 }
1765 }
1766
1767 ref2frm[0] =
1768 ref2frm[1] = -1;
1769 for (i = 0; i < 16; i++)
1770 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1771 ref2frm[18 + 0] =
1772 ref2frm[18 + 1] = -1;
1773 for (i = 16; i < 48; i++)
1774 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1775 (h->ref_list[j][i].reference & 3);
1776 }
1777
1778 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1779 av_log(h->avctx, AV_LOG_DEBUG,
1780 "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",
1781 h->slice_num,
1782 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1783 first_mb_in_slice,
1784 av_get_picture_type_char(h->slice_type),
1785 h->slice_type_fixed ? " fix" : "",
1786 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1787 pps_id, h->frame_num,
1788 h->cur_pic_ptr->field_poc[0],
1789 h->cur_pic_ptr->field_poc[1],
1790 h->ref_count[0], h->ref_count[1],
1791 h->qscale,
1792 h->deblocking_filter,
1793 h->slice_alpha_c0_offset, h->slice_beta_offset,
1794 h->use_weight,
1795 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1796 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1797 }
1798
1799 return 0;
1800 }
1801
1802 int ff_h264_get_slice_type(const H264Context *h)
1803 {
1804 switch (h->slice_type) {
1805 case AV_PICTURE_TYPE_P:
1806 return 0;
1807 case AV_PICTURE_TYPE_B:
1808 return 1;
1809 case AV_PICTURE_TYPE_I:
1810 return 2;
1811 case AV_PICTURE_TYPE_SP:
1812 return 3;
1813 case AV_PICTURE_TYPE_SI:
1814 return 4;
1815 default:
1816 return AVERROR_INVALIDDATA;
1817 }
1818 }
1819
1820 static av_always_inline void fill_filter_caches_inter(H264Context *h,
1821 int mb_type, int top_xy,
1822 int left_xy[LEFT_MBS],
1823 int top_type,
1824 int left_type[LEFT_MBS],
1825 int mb_xy, int list)
1826 {
1827 int b_stride = h->b_stride;
1828 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
1829 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
1830 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1831 if (USES_LIST(top_type, list)) {
1832 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1833 const int b8_xy = 4 * top_xy + 2;
1834 int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1835 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1836 ref_cache[0 - 1 * 8] =
1837 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1838 ref_cache[2 - 1 * 8] =
1839 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1840 } else {
1841 AV_ZERO128(mv_dst - 1 * 8);
1842 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1843 }
1844
1845 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1846 if (USES_LIST(left_type[LTOP], list)) {
1847 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1848 const int b8_xy = 4 * left_xy[LTOP] + 1;
1849 int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1850 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1851 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1852 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1853 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1854 ref_cache[-1 + 0] =
1855 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1856 ref_cache[-1 + 16] =
1857 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1858 } else {
1859 AV_ZERO32(mv_dst - 1 + 0);
1860 AV_ZERO32(mv_dst - 1 + 8);
1861 AV_ZERO32(mv_dst - 1 + 16);
1862 AV_ZERO32(mv_dst - 1 + 24);
1863 ref_cache[-1 + 0] =
1864 ref_cache[-1 + 8] =
1865 ref_cache[-1 + 16] =
1866 ref_cache[-1 + 24] = LIST_NOT_USED;
1867 }
1868 }
1869 }
1870
1871 if (!USES_LIST(mb_type, list)) {
1872 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1873 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1874 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1875 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1876 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1877 return;
1878 }
1879
1880 {
1881 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1882 int (*ref2frm)[64] = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1883 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1884 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1885 AV_WN32A(&ref_cache[0 * 8], ref01);
1886 AV_WN32A(&ref_cache[1 * 8], ref01);
1887 AV_WN32A(&ref_cache[2 * 8], ref23);
1888 AV_WN32A(&ref_cache[3 * 8], ref23);
1889 }
1890
1891 {
1892 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
1893 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1894 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1895 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1896 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1897 }
1898 }
1899
1900 /**
1901 *
1902 * @return non zero if the loop filter can be skipped
1903 */
1904 static int fill_filter_caches(H264Context *h, int mb_type)
1905 {
1906 const int mb_xy = h->mb_xy;
1907 int top_xy, left_xy[LEFT_MBS];
1908 int top_type, left_type[LEFT_MBS];
1909 uint8_t *nnz;
1910 uint8_t *nnz_cache;
1911
1912 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
1913
1914 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1915 * stuff, I can't imagine that these complex rules are worth it. */
1916
1917 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1918 if (FRAME_MBAFF(h)) {
1919 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1920 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1921 if (h->mb_y & 1) {
1922 if (left_mb_field_flag != curr_mb_field_flag)
1923 left_xy[LTOP] -= h->mb_stride;
1924 } else {
1925 if (curr_mb_field_flag)
1926 top_xy += h->mb_stride &
1927 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1928 if (left_mb_field_flag != curr_mb_field_flag)
1929 left_xy[LBOT] += h->mb_stride;
1930 }
1931 }
1932
1933 h->top_mb_xy = top_xy;
1934 h->left_mb_xy[LTOP] = left_xy[LTOP];
1935 h->left_mb_xy[LBOT] = left_xy[LBOT];
1936 {
1937 /* For sufficiently low qp, filtering wouldn't do anything.
1938 * This is a conservative estimate: could also check beta_offset
1939 * and more accurate chroma_qp. */
1940 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1941 int qp = h->cur_pic.qscale_table[mb_xy];
1942 if (qp <= qp_thresh &&
1943 (left_xy[LTOP] < 0 ||
1944 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1945 (top_xy < 0 ||
1946 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1947 if (!FRAME_MBAFF(h))
1948 return 1;
1949 if ((left_xy[LTOP] < 0 ||
1950 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1951 (top_xy < h->mb_stride ||
1952 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1953 return 1;
1954 }
1955 }
1956
1957 top_type = h->cur_pic.mb_type[top_xy];
1958 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1959 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1960 if (h->deblocking_filter == 2) {
1961 if (h->slice_table[top_xy] != h->slice_num)
1962 top_type = 0;
1963 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
1964 left_type[LTOP] = left_type[LBOT] = 0;
1965 } else {
1966 if (h->slice_table[top_xy] == 0xFFFF)
1967 top_type = 0;
1968 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1969 left_type[LTOP] = left_type[LBOT] = 0;
1970 }
1971 h->top_type = top_type;
1972 h->left_type[LTOP] = left_type[LTOP];
1973 h->left_type[LBOT] = left_type[LBOT];
1974
1975 if (IS_INTRA(mb_type))
1976 return 0;
1977
1978 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1979 top_type, left_type, mb_xy, 0);
1980 if (h->list_count == 2)
1981 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1982 top_type, left_type, mb_xy, 1);
1983
1984 nnz = h->non_zero_count[mb_xy];
1985 nnz_cache = h->non_zero_count_cache;
1986 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1987 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1988 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1989 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1990 h->cbp = h->cbp_table[mb_xy];
1991
1992 if (top_type) {
1993 nnz = h->non_zero_count[top_xy];
1994 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1995 }
1996
1997 if (left_type[LTOP]) {
1998 nnz = h->non_zero_count[left_xy[LTOP]];
1999 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2000 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2001 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2002 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2003 }
2004
2005 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2006 * from what the loop filter needs */
2007 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2008 if (IS_8x8DCT(top_type)) {
2009 nnz_cache[4 + 8 * 0] =
2010 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2011 nnz_cache[6 + 8 * 0] =
2012 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2013 }
2014 if (IS_8x8DCT(left_type[LTOP])) {
2015 nnz_cache[3 + 8 * 1] =
2016 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2017 }
2018 if (IS_8x8DCT(left_type[LBOT])) {
2019 nnz_cache[3 + 8 * 3] =
2020 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2021 }
2022
2023 if (IS_8x8DCT(mb_type)) {
2024 nnz_cache[scan8[0]] =
2025 nnz_cache[scan8[1]] =
2026 nnz_cache[scan8[2]] =
2027 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2028
2029 nnz_cache[scan8[0 + 4]] =
2030 nnz_cache[scan8[1 + 4]] =
2031 nnz_cache[scan8[2 + 4]] =
2032 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2033
2034 nnz_cache[scan8[0 + 8]] =
2035 nnz_cache[scan8[1 + 8]] =
2036 nnz_cache[scan8[2 + 8]] =
2037 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2038
2039 nnz_cache[scan8[0 + 12]] =
2040 nnz_cache[scan8[1 + 12]] =
2041 nnz_cache[scan8[2 + 12]] =
2042 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2043 }
2044 }
2045
2046 return 0;
2047 }
2048
2049 static void loop_filter(H264Context *h, int start_x, int end_x)
2050 {
2051 uint8_t *dest_y, *dest_cb, *dest_cr;
2052 int linesize, uvlinesize, mb_x, mb_y;
2053 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2054 const int old_slice_type = h->slice_type;
2055 const int pixel_shift = h->pixel_shift;
2056 const int block_h = 16 >> h->chroma_y_shift;
2057
2058 if (h->deblocking_filter) {
2059 for (mb_x = start_x; mb_x < end_x; mb_x++)
2060 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2061 int mb_xy, mb_type;
2062 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2063 h->slice_num = h->slice_table[mb_xy];
2064 mb_type = h->cur_pic.mb_type[mb_xy];
2065 h->list_count = h->list_counts[mb_xy];
2066
2067 if (FRAME_MBAFF(h))
2068 h->mb_mbaff =
2069 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2070
2071 h->mb_x = mb_x;
2072 h->mb_y = mb_y;
2073 dest_y = h->cur_pic.f.data[0] +
2074 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2075 dest_cb = h->cur_pic.f.data[1] +
2076 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2077 mb_y * h->uvlinesize * block_h;
2078 dest_cr = h->cur_pic.f.data[2] +
2079 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2080 mb_y * h->uvlinesize * block_h;
2081 // FIXME simplify above
2082
2083 if (MB_FIELD(h)) {
2084 linesize = h->mb_linesize = h->linesize * 2;
2085 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2086 if (mb_y & 1) { // FIXME move out of this function?
2087 dest_y -= h->linesize * 15;
2088 dest_cb -= h->uvlinesize * (block_h - 1);
2089 dest_cr -= h->uvlinesize * (block_h - 1);
2090 }
2091 } else {
2092 linesize = h->mb_linesize = h->linesize;
2093 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2094 }
2095 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2096 uvlinesize, 0);
2097 if (fill_filter_caches(h, mb_type))
2098 continue;
2099 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2100 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2101
2102 if (FRAME_MBAFF(h)) {
2103 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2104 linesize, uvlinesize);
2105 } else {
2106 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2107 dest_cr, linesize, uvlinesize);
2108 }
2109 }
2110 }
2111 h->slice_type = old_slice_type;
2112 h->mb_x = end_x;
2113 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2114 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2115 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2116 }
2117
2118 static void predict_field_decoding_flag(H264Context *h)
2119 {
2120 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2121 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2122 h->cur_pic.mb_type[mb_xy - 1] :
2123 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2124 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2125 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2126 }
2127
2128 /**
2129 * Draw edges and report progress for the last MB row.
2130 */
2131 static void decode_finish_row(H264Context *h)
2132 {
2133 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2134 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2135 int height = 16 << FRAME_MBAFF(h);
2136 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2137
2138 if (h->deblocking_filter) {
2139 if ((top + height) >= pic_height)
2140 height += deblock_border;
2141 top -= deblock_border;
2142 }
2143
2144 if (top >= pic_height || (top + height) < 0)
2145 return;
2146
2147 height = FFMIN(height, pic_height - top);
2148 if (top < 0) {
2149 height = top + height;
2150 top = 0;
2151 }
2152
2153 ff_h264_draw_horiz_band(h, top, height);
2154
2155 if (h->droppable)
2156 return;
2157
2158 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2159 h->picture_structure == PICT_BOTTOM_FIELD);
2160 }
2161
2162 static void er_add_slice(H264Context *h, int startx, int starty,
2163 int endx, int endy, int status)
2164 {
2165 #if CONFIG_ERROR_RESILIENCE
2166 ERContext *er = &h->er;
2167
2168 er->ref_count = h->ref_count[0];
2169 ff_er_add_slice(er, startx, starty, endx, endy, status);
2170 #endif
2171 }
2172
2173 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2174 {
2175 H264Context *h = *(void **)arg;
2176 int lf_x_start = h->mb_x;
2177
2178 h->mb_skip_run = -1;
2179
2180 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2181 avctx->codec_id != AV_CODEC_ID_H264 ||
2182 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2183
2184 if (h->pps.cabac) {
2185 /* realign */
2186 align_get_bits(&h->gb);
2187
2188 /* init cabac */
2189 ff_init_cabac_decoder(&h->cabac,
2190 h->gb.buffer + get_bits_count(&h->gb) / 8,
2191 (get_bits_left(&h->gb) + 7) / 8);
2192
2193 ff_h264_init_cabac_states(h);
2194
2195 for (;;) {
2196 // START_TIMER
2197 int ret = ff_h264_decode_mb_cabac(h);
2198 int eos;
2199 // STOP_TIMER("decode_mb_cabac")
2200
2201 if (ret >= 0)
2202 ff_h264_hl_decode_mb(h);
2203
2204 // FIXME optimal? or let mb_decode decode 16x32 ?
2205 if (ret >= 0 && FRAME_MBAFF(h)) {
2206 h->mb_y++;
2207
2208 ret = ff_h264_decode_mb_cabac(h);
2209
2210 if (ret >= 0)
2211 ff_h264_hl_decode_mb(h);
2212 h->mb_y--;
2213 }
2214 eos = get_cabac_terminate(&h->cabac);
2215
2216 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2217 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2218 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2219 h->mb_y, ER_MB_END);
2220 if (h->mb_x >= lf_x_start)
2221 loop_filter(h, lf_x_start, h->mb_x + 1);
2222 return 0;
2223 }
2224 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2225 av_log(h->avctx, AV_LOG_ERROR,
2226 "error while decoding MB %d %d, bytestream %td\n",
2227 h->mb_x, h->mb_y,
2228 h->cabac.bytestream_end - h->cabac.bytestream);
2229 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2230 h->mb_y, ER_MB_ERROR);
2231 return AVERROR_INVALIDDATA;
2232 }
2233
2234 if (++h->mb_x >= h->mb_width) {
2235 loop_filter(h, lf_x_start, h->mb_x);
2236 h->mb_x = lf_x_start = 0;
2237 decode_finish_row(h);
2238 ++h->mb_y;
2239 if (FIELD_OR_MBAFF_PICTURE(h)) {
2240 ++h->mb_y;
2241 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2242 predict_field_decoding_flag(h);
2243 }
2244 }
2245
2246 if (eos || h->mb_y >= h->mb_height) {
2247 tprintf(h->avctx, "slice end %d %d\n",
2248 get_bits_count(&h->gb), h->gb.size_in_bits);
2249 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2250 h->mb_y, ER_MB_END);
2251 if (h->mb_x > lf_x_start)
2252 loop_filter(h, lf_x_start, h->mb_x);
2253 return 0;
2254 }
2255 }
2256 } else {
2257 for (;;) {
2258 int ret = ff_h264_decode_mb_cavlc(h);
2259
2260 if (ret >= 0)
2261 ff_h264_hl_decode_mb(h);
2262
2263 // FIXME optimal? or let mb_decode decode 16x32 ?
2264 if (ret >= 0 && FRAME_MBAFF(h)) {
2265 h->mb_y++;
2266 ret = ff_h264_decode_mb_cavlc(h);
2267
2268 if (ret >= 0)
2269 ff_h264_hl_decode_mb(h);
2270 h->mb_y--;
2271 }
2272
2273 if (ret < 0) {
2274 av_log(h->avctx, AV_LOG_ERROR,
2275 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2276 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2277 h->mb_y, ER_MB_ERROR);
2278 return ret;
2279 }
2280
2281 if (++h->mb_x >= h->mb_width) {
2282 loop_filter(h, lf_x_start, h->mb_x);
2283 h->mb_x = lf_x_start = 0;
2284 decode_finish_row(h);
2285 ++h->mb_y;
2286 if (FIELD_OR_MBAFF_PICTURE(h)) {
2287 ++h->mb_y;
2288 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2289 predict_field_decoding_flag(h);
2290 }
2291 if (h->mb_y >= h->mb_height) {
2292 tprintf(h->avctx, "slice end %d %d\n",
2293 get_bits_count(&h->gb), h->gb.size_in_bits);
2294
2295 if (get_bits_left(&h->gb) == 0) {
2296 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2297 h->mb_x - 1, h->mb_y,
2298 ER_MB_END);
2299
2300 return 0;
2301 } else {
2302 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2303 h->mb_x - 1, h->mb_y,
2304 ER_MB_END);
2305
2306 return AVERROR_INVALIDDATA;
2307 }
2308 }
2309 }
2310
2311 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2312 tprintf(h->avctx, "slice end %d %d\n",
2313 get_bits_count(&h->gb), h->gb.size_in_bits);
2314
2315 if (get_bits_left(&h->gb) == 0) {
2316 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2317 h->mb_x - 1, h->mb_y,
2318 ER_MB_END);
2319 if (h->mb_x > lf_x_start)
2320 loop_filter(h, lf_x_start, h->mb_x);
2321
2322 return 0;
2323 } else {
2324 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2325 h->mb_y, ER_MB_ERROR);
2326
2327 return AVERROR_INVALIDDATA;
2328 }
2329 }
2330 }
2331 }
2332 }
2333
2334 /**
2335 * Call decode_slice() for each context.
2336 *
2337 * @param h h264 master context
2338 * @param context_count number of contexts to execute
2339 */
2340 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2341 {
2342 AVCodecContext *const avctx = h->avctx;
2343 H264Context *hx;
2344 int i;
2345
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;
2350 }
2351
2352 if (h->avctx->hwaccel)
2353 return 0;
2354 if (context_count == 1) {
2355 return decode_slice(avctx, &h);
2356 } else {
2357 for (i = 1; i < context_count; i++) {
2358 hx = h->thread_context[i];
2359 hx->er.error_count = 0;
2360 }
2361
2362 avctx->execute(avctx, decode_slice, h->thread_context,
2363 NULL, context_count, sizeof(void *));
2364
2365 /* pull back stuff from slices to master context */
2366 hx = h->thread_context[context_count - 1];
2367 h->mb_x = hx->mb_x;
2368 h->mb_y = hx->mb_y;
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;
2373 }
2374
2375 return 0;
2376 }