blob: e3de18a510502d6b53dae8c95047e4c5ec39bcbc
1 | /* |
2 | * DV encoder |
3 | * Copyright (c) 2003 Roman Shaposhnik |
4 | * |
5 | * This file is part of FFmpeg. |
6 | * |
7 | * FFmpeg is free software; you can redistribute it and/or |
8 | * modify it under the terms of the GNU Lesser General Public |
9 | * License as published by the Free Software Foundation; either |
10 | * version 2.1 of the License, or (at your option) any later version. |
11 | * |
12 | * FFmpeg is distributed in the hope that it will be useful, |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 | * Lesser General Public License for more details. |
16 | * |
17 | * You should have received a copy of the GNU Lesser General Public |
18 | * License along with FFmpeg; if not, write to the Free Software |
19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
20 | */ |
21 | |
22 | /** |
23 | * @file |
24 | * DV encoder |
25 | */ |
26 | |
27 | #include "config.h" |
28 | |
29 | #include "libavutil/attributes.h" |
30 | #include "libavutil/internal.h" |
31 | #include "libavutil/pixdesc.h" |
32 | |
33 | #include "avcodec.h" |
34 | #include "dv.h" |
35 | #include "dv_profile_internal.h" |
36 | #include "dv_tablegen.h" |
37 | #include "fdctdsp.h" |
38 | #include "internal.h" |
39 | #include "mathops.h" |
40 | #include "me_cmp.h" |
41 | #include "pixblockdsp.h" |
42 | #include "put_bits.h" |
43 | |
44 | static av_cold int dvvideo_encode_init(AVCodecContext *avctx) |
45 | { |
46 | DVVideoContext *s = avctx->priv_data; |
47 | FDCTDSPContext fdsp; |
48 | MECmpContext mecc; |
49 | PixblockDSPContext pdsp; |
50 | int ret; |
51 | |
52 | s->sys = av_dv_codec_profile2(avctx->width, avctx->height, avctx->pix_fmt, avctx->time_base); |
53 | if (!s->sys) { |
54 | av_log(avctx, AV_LOG_ERROR, "Found no DV profile for %ix%i %s video. " |
55 | "Valid DV profiles are:\n", |
56 | avctx->width, avctx->height, av_get_pix_fmt_name(avctx->pix_fmt)); |
57 | ff_dv_print_profiles(avctx, AV_LOG_ERROR); |
58 | return AVERROR(EINVAL); |
59 | } |
60 | if (avctx->height > 576) { |
61 | av_log(avctx, AV_LOG_ERROR, "DVCPRO HD encoding is not supported.\n"); |
62 | return AVERROR_PATCHWELCOME; |
63 | } |
64 | ret = ff_dv_init_dynamic_tables(s, s->sys); |
65 | if (ret < 0) { |
66 | av_log(avctx, AV_LOG_ERROR, "Error initializing work tables.\n"); |
67 | return ret; |
68 | } |
69 | |
70 | dv_vlc_map_tableinit(); |
71 | |
72 | memset(&fdsp,0, sizeof(fdsp)); |
73 | memset(&mecc,0, sizeof(mecc)); |
74 | memset(&pdsp,0, sizeof(pdsp)); |
75 | ff_fdctdsp_init(&fdsp, avctx); |
76 | ff_me_cmp_init(&mecc, avctx); |
77 | ff_pixblockdsp_init(&pdsp, avctx); |
78 | ff_set_cmp(&mecc, mecc.ildct_cmp, avctx->ildct_cmp); |
79 | |
80 | s->get_pixels = pdsp.get_pixels; |
81 | s->ildct_cmp = mecc.ildct_cmp[5]; |
82 | |
83 | s->fdct[0] = fdsp.fdct; |
84 | s->fdct[1] = fdsp.fdct248; |
85 | |
86 | return ff_dvvideo_init(avctx); |
87 | } |
88 | |
89 | /* bit budget for AC only in 5 MBs */ |
90 | static const int vs_total_ac_bits = (100 * 4 + 68 * 2) * 5; |
91 | static const int mb_area_start[5] = { 1, 6, 21, 43, 64 }; |
92 | |
93 | #if CONFIG_SMALL |
94 | /* Convert run and level (where level != 0) pair into VLC, returning bit size */ |
95 | static av_always_inline int dv_rl2vlc(int run, int level, int sign, |
96 | uint32_t *vlc) |
97 | { |
98 | int size; |
99 | if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) { |
100 | *vlc = dv_vlc_map[run][level].vlc | sign; |
101 | size = dv_vlc_map[run][level].size; |
102 | } else { |
103 | if (level < DV_VLC_MAP_LEV_SIZE) { |
104 | *vlc = dv_vlc_map[0][level].vlc | sign; |
105 | size = dv_vlc_map[0][level].size; |
106 | } else { |
107 | *vlc = 0xfe00 | (level << 1) | sign; |
108 | size = 16; |
109 | } |
110 | if (run) { |
111 | *vlc |= ((run < 16) ? dv_vlc_map[run - 1][0].vlc : |
112 | (0x1f80 | (run - 1))) << size; |
113 | size += (run < 16) ? dv_vlc_map[run - 1][0].size : 13; |
114 | } |
115 | } |
116 | |
117 | return size; |
118 | } |
119 | |
120 | static av_always_inline int dv_rl2vlc_size(int run, int level) |
121 | { |
122 | int size; |
123 | |
124 | if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) { |
125 | size = dv_vlc_map[run][level].size; |
126 | } else { |
127 | size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16; |
128 | if (run) |
129 | size += (run < 16) ? dv_vlc_map[run - 1][0].size : 13; |
130 | } |
131 | return size; |
132 | } |
133 | #else |
134 | static av_always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t *vlc) |
135 | { |
136 | *vlc = dv_vlc_map[run][l].vlc | sign; |
137 | return dv_vlc_map[run][l].size; |
138 | } |
139 | |
140 | static av_always_inline int dv_rl2vlc_size(int run, int l) |
141 | { |
142 | return dv_vlc_map[run][l].size; |
143 | } |
144 | #endif |
145 | |
146 | typedef struct EncBlockInfo { |
147 | int area_q[4]; |
148 | int bit_size[4]; |
149 | int prev[5]; |
150 | int cur_ac; |
151 | int cno; |
152 | int dct_mode; |
153 | int16_t mb[64]; |
154 | uint8_t next[64]; |
155 | uint8_t sign[64]; |
156 | uint8_t partial_bit_count; |
157 | uint32_t partial_bit_buffer; /* we can't use uint16_t here */ |
158 | } EncBlockInfo; |
159 | |
160 | static av_always_inline PutBitContext *dv_encode_ac(EncBlockInfo *bi, |
161 | PutBitContext *pb_pool, |
162 | PutBitContext *pb_end) |
163 | { |
164 | int prev, bits_left; |
165 | PutBitContext *pb = pb_pool; |
166 | int size = bi->partial_bit_count; |
167 | uint32_t vlc = bi->partial_bit_buffer; |
168 | |
169 | bi->partial_bit_count = |
170 | bi->partial_bit_buffer = 0; |
171 | for (;;) { |
172 | /* Find suitable storage space */ |
173 | for (; size > (bits_left = put_bits_left(pb)); pb++) { |
174 | if (bits_left) { |
175 | size -= bits_left; |
176 | put_bits(pb, bits_left, vlc >> size); |
177 | vlc = av_mod_uintp2(vlc, size); |
178 | } |
179 | if (pb + 1 >= pb_end) { |
180 | bi->partial_bit_count = size; |
181 | bi->partial_bit_buffer = vlc; |
182 | return pb; |
183 | } |
184 | } |
185 | |
186 | /* Store VLC */ |
187 | put_bits(pb, size, vlc); |
188 | |
189 | if (bi->cur_ac >= 64) |
190 | break; |
191 | |
192 | /* Construct the next VLC */ |
193 | prev = bi->cur_ac; |
194 | bi->cur_ac = bi->next[prev]; |
195 | if (bi->cur_ac < 64) { |
196 | size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], |
197 | bi->sign[bi->cur_ac], &vlc); |
198 | } else { |
199 | size = 4; |
200 | vlc = 6; /* End Of Block stamp */ |
201 | } |
202 | } |
203 | return pb; |
204 | } |
205 | |
206 | static av_always_inline int dv_guess_dct_mode(DVVideoContext *s, uint8_t *data, |
207 | ptrdiff_t linesize) |
208 | { |
209 | if (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) { |
210 | int ps = s->ildct_cmp(NULL, data, NULL, linesize, 8) - 400; |
211 | if (ps > 0) { |
212 | int is = s->ildct_cmp(NULL, data, NULL, linesize << 1, 4) + |
213 | s->ildct_cmp(NULL, data + linesize, NULL, linesize << 1, 4); |
214 | return ps > is; |
215 | } |
216 | } |
217 | |
218 | return 0; |
219 | } |
220 | |
221 | static const int dv_weight_bits = 18; |
222 | static const int dv_weight_88[64] = { |
223 | 131072, 257107, 257107, 242189, 252167, 242189, 235923, 237536, |
224 | 237536, 235923, 229376, 231390, 223754, 231390, 229376, 222935, |
225 | 224969, 217965, 217965, 224969, 222935, 200636, 218652, 211916, |
226 | 212325, 211916, 218652, 200636, 188995, 196781, 205965, 206433, |
227 | 206433, 205965, 196781, 188995, 185364, 185364, 200636, 200704, |
228 | 200636, 185364, 185364, 174609, 180568, 195068, 195068, 180568, |
229 | 174609, 170091, 175557, 189591, 175557, 170091, 165371, 170627, |
230 | 170627, 165371, 160727, 153560, 160727, 144651, 144651, 136258, |
231 | }; |
232 | static const int dv_weight_248[64] = { |
233 | 131072, 262144, 257107, 257107, 242189, 242189, 242189, 242189, |
234 | 237536, 237536, 229376, 229376, 200636, 200636, 224973, 224973, |
235 | 223754, 223754, 235923, 235923, 229376, 229376, 217965, 217965, |
236 | 211916, 211916, 196781, 196781, 185364, 185364, 206433, 206433, |
237 | 211916, 211916, 222935, 222935, 200636, 200636, 205964, 205964, |
238 | 200704, 200704, 180568, 180568, 175557, 175557, 195068, 195068, |
239 | 185364, 185364, 188995, 188995, 174606, 174606, 175557, 175557, |
240 | 170627, 170627, 153560, 153560, 165371, 165371, 144651, 144651, |
241 | }; |
242 | |
243 | static av_always_inline int dv_init_enc_block(EncBlockInfo *bi, uint8_t *data, |
244 | ptrdiff_t linesize, |
245 | DVVideoContext *s, int bias) |
246 | { |
247 | const int *weight; |
248 | const uint8_t *zigzag_scan; |
249 | LOCAL_ALIGNED_16(int16_t, blk, [64]); |
250 | int i, area; |
251 | /* We offer two different methods for class number assignment: the |
252 | * method suggested in SMPTE 314M Table 22, and an improved |
253 | * method. The SMPTE method is very conservative; it assigns class |
254 | * 3 (i.e. severe quantization) to any block where the largest AC |
255 | * component is greater than 36. FFmpeg's DV encoder tracks AC bit |
256 | * consumption precisely, so there is no need to bias most blocks |
257 | * towards strongly lossy compression. Instead, we assign class 2 |
258 | * to most blocks, and use class 3 only when strictly necessary |
259 | * (for blocks whose largest AC component exceeds 255). */ |
260 | |
261 | #if 0 /* SMPTE spec method */ |
262 | static const int classes[] = { 12, 24, 36, 0xffff }; |
263 | #else /* improved FFmpeg method */ |
264 | static const int classes[] = { -1, -1, 255, 0xffff }; |
265 | #endif |
266 | int max = classes[0]; |
267 | int prev = 0; |
268 | |
269 | av_assert2((((int) blk) & 15) == 0); |
270 | |
271 | bi->area_q[0] = |
272 | bi->area_q[1] = |
273 | bi->area_q[2] = |
274 | bi->area_q[3] = 0; |
275 | bi->partial_bit_count = 0; |
276 | bi->partial_bit_buffer = 0; |
277 | bi->cur_ac = 0; |
278 | if (data) { |
279 | bi->dct_mode = dv_guess_dct_mode(s, data, linesize); |
280 | s->get_pixels(blk, data, linesize); |
281 | s->fdct[bi->dct_mode](blk); |
282 | } else { |
283 | /* We rely on the fact that encoding all zeros leads to an immediate |
284 | * EOB, which is precisely what the spec calls for in the "dummy" |
285 | * blocks. */ |
286 | memset(blk, 0, 64 * sizeof(*blk)); |
287 | bi->dct_mode = 0; |
288 | } |
289 | bi->mb[0] = blk[0]; |
290 | |
291 | zigzag_scan = bi->dct_mode ? ff_dv_zigzag248_direct : ff_zigzag_direct; |
292 | weight = bi->dct_mode ? dv_weight_248 : dv_weight_88; |
293 | |
294 | for (area = 0; area < 4; area++) { |
295 | bi->prev[area] = prev; |
296 | bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) |
297 | for (i = mb_area_start[area]; i < mb_area_start[area + 1]; i++) { |
298 | int level = blk[zigzag_scan[i]]; |
299 | |
300 | if (level + 15 > 30U) { |
301 | bi->sign[i] = (level >> 31) & 1; |
302 | /* Weight it and shift down into range, adding for rounding. |
303 | * The extra division by a factor of 2^4 reverses the 8x |
304 | * expansion of the DCT AND the 2x doubling of the weights. */ |
305 | level = (FFABS(level) * weight[i] + (1 << (dv_weight_bits + 3))) >> |
306 | (dv_weight_bits + 4); |
307 | bi->mb[i] = level; |
308 | if (level > max) |
309 | max = level; |
310 | bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, level); |
311 | bi->next[prev] = i; |
312 | prev = i; |
313 | } |
314 | } |
315 | } |
316 | bi->next[prev] = i; |
317 | for (bi->cno = 0; max > classes[bi->cno]; bi->cno++) |
318 | ; |
319 | |
320 | bi->cno += bias; |
321 | |
322 | if (bi->cno >= 3) { |
323 | bi->cno = 3; |
324 | prev = 0; |
325 | i = bi->next[prev]; |
326 | for (area = 0; area < 4; area++) { |
327 | bi->prev[area] = prev; |
328 | bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) |
329 | for (; i < mb_area_start[area + 1]; i = bi->next[i]) { |
330 | bi->mb[i] >>= 1; |
331 | |
332 | if (bi->mb[i]) { |
333 | bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]); |
334 | bi->next[prev] = i; |
335 | prev = i; |
336 | } |
337 | } |
338 | } |
339 | bi->next[prev] = i; |
340 | } |
341 | |
342 | return bi->bit_size[0] + bi->bit_size[1] + |
343 | bi->bit_size[2] + bi->bit_size[3]; |
344 | } |
345 | |
346 | static inline void dv_guess_qnos(EncBlockInfo *blks, int *qnos) |
347 | { |
348 | int size[5]; |
349 | int i, j, k, a, prev, a2; |
350 | EncBlockInfo *b; |
351 | |
352 | size[0] = |
353 | size[1] = |
354 | size[2] = |
355 | size[3] = |
356 | size[4] = 1 << 24; |
357 | do { |
358 | b = blks; |
359 | for (i = 0; i < 5; i++) { |
360 | if (!qnos[i]) |
361 | continue; |
362 | |
363 | qnos[i]--; |
364 | size[i] = 0; |
365 | for (j = 0; j < 6; j++, b++) { |
366 | for (a = 0; a < 4; a++) { |
367 | if (b->area_q[a] != ff_dv_quant_shifts[qnos[i] + ff_dv_quant_offset[b->cno]][a]) { |
368 | b->bit_size[a] = 1; // 4 areas 4 bits for EOB :) |
369 | b->area_q[a]++; |
370 | prev = b->prev[a]; |
371 | av_assert2(b->next[prev] >= mb_area_start[a + 1] || b->mb[prev]); |
372 | for (k = b->next[prev]; k < mb_area_start[a + 1]; k = b->next[k]) { |
373 | b->mb[k] >>= 1; |
374 | if (b->mb[k]) { |
375 | b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]); |
376 | prev = k; |
377 | } else { |
378 | if (b->next[k] >= mb_area_start[a + 1] && b->next[k] < 64) { |
379 | for (a2 = a + 1; b->next[k] >= mb_area_start[a2 + 1]; a2++) |
380 | b->prev[a2] = prev; |
381 | av_assert2(a2 < 4); |
382 | av_assert2(b->mb[b->next[k]]); |
383 | b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]]) - |
384 | dv_rl2vlc_size(b->next[k] - k - 1, b->mb[b->next[k]]); |
385 | av_assert2(b->prev[a2] == k && (a2 + 1 >= 4 || b->prev[a2 + 1] != k)); |
386 | b->prev[a2] = prev; |
387 | } |
388 | b->next[prev] = b->next[k]; |
389 | } |
390 | } |
391 | b->prev[a + 1] = prev; |
392 | } |
393 | size[i] += b->bit_size[a]; |
394 | } |
395 | } |
396 | if (vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4]) |
397 | return; |
398 | } |
399 | } while (qnos[0] | qnos[1] | qnos[2] | qnos[3] | qnos[4]); |
400 | |
401 | for (a = 2; a == 2 || vs_total_ac_bits < size[0]; a += a) { |
402 | b = blks; |
403 | size[0] = 5 * 6 * 4; // EOB |
404 | for (j = 0; j < 6 * 5; j++, b++) { |
405 | prev = b->prev[0]; |
406 | for (k = b->next[prev]; k < 64; k = b->next[k]) { |
407 | if (b->mb[k] < a && b->mb[k] > -a) { |
408 | b->next[prev] = b->next[k]; |
409 | } else { |
410 | size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]); |
411 | prev = k; |
412 | } |
413 | } |
414 | } |
415 | } |
416 | } |
417 | |
418 | static int dv_encode_video_segment(AVCodecContext *avctx, void *arg) |
419 | { |
420 | DVVideoContext *s = avctx->priv_data; |
421 | DVwork_chunk *work_chunk = arg; |
422 | int mb_index, i, j; |
423 | int mb_x, mb_y, c_offset; |
424 | ptrdiff_t linesize, y_stride; |
425 | uint8_t *y_ptr; |
426 | uint8_t *dif; |
427 | LOCAL_ALIGNED_8(uint8_t, scratch, [128]); |
428 | EncBlockInfo enc_blks[5 * DV_MAX_BPM]; |
429 | PutBitContext pbs[5 * DV_MAX_BPM]; |
430 | PutBitContext *pb; |
431 | EncBlockInfo *enc_blk; |
432 | int vs_bit_size = 0; |
433 | int qnos[5] = { 15, 15, 15, 15, 15 }; /* No quantization */ |
434 | int *qnosp = &qnos[0]; |
435 | |
436 | dif = &s->buf[work_chunk->buf_offset * 80]; |
437 | enc_blk = &enc_blks[0]; |
438 | for (mb_index = 0; mb_index < 5; mb_index++) { |
439 | dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y); |
440 | |
441 | /* initializing luminance blocks */ |
442 | if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) || |
443 | (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) || |
444 | (s->sys->height >= 720 && mb_y != 134)) { |
445 | y_stride = s->frame->linesize[0] << 3; |
446 | } else { |
447 | y_stride = 16; |
448 | } |
449 | y_ptr = s->frame->data[0] + |
450 | ((mb_y * s->frame->linesize[0] + mb_x) << 3); |
451 | linesize = s->frame->linesize[0]; |
452 | |
453 | if (s->sys->video_stype == 4) { /* SD 422 */ |
454 | vs_bit_size += |
455 | dv_init_enc_block(enc_blk + 0, y_ptr, linesize, s, 0) + |
456 | dv_init_enc_block(enc_blk + 1, NULL, linesize, s, 0) + |
457 | dv_init_enc_block(enc_blk + 2, y_ptr + 8, linesize, s, 0) + |
458 | dv_init_enc_block(enc_blk + 3, NULL, linesize, s, 0); |
459 | } else { |
460 | vs_bit_size += |
461 | dv_init_enc_block(enc_blk + 0, y_ptr, linesize, s, 0) + |
462 | dv_init_enc_block(enc_blk + 1, y_ptr + 8, linesize, s, 0) + |
463 | dv_init_enc_block(enc_blk + 2, y_ptr + y_stride, linesize, s, 0) + |
464 | dv_init_enc_block(enc_blk + 3, y_ptr + 8 + y_stride, linesize, s, 0); |
465 | } |
466 | enc_blk += 4; |
467 | |
468 | /* initializing chrominance blocks */ |
469 | c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] + |
470 | (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << 3); |
471 | for (j = 2; j; j--) { |
472 | uint8_t *c_ptr = s->frame->data[j] + c_offset; |
473 | linesize = s->frame->linesize[j]; |
474 | y_stride = (mb_y == 134) ? 8 : (s->frame->linesize[j] << 3); |
475 | if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) { |
476 | uint8_t *d; |
477 | uint8_t *b = scratch; |
478 | for (i = 0; i < 8; i++) { |
479 | d = c_ptr + (linesize << 3); |
480 | b[0] = c_ptr[0]; |
481 | b[1] = c_ptr[1]; |
482 | b[2] = c_ptr[2]; |
483 | b[3] = c_ptr[3]; |
484 | b[4] = d[0]; |
485 | b[5] = d[1]; |
486 | b[6] = d[2]; |
487 | b[7] = d[3]; |
488 | c_ptr += linesize; |
489 | b += 16; |
490 | } |
491 | c_ptr = scratch; |
492 | linesize = 16; |
493 | } |
494 | |
495 | vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr, linesize, s, 1); |
496 | if (s->sys->bpm == 8) |
497 | vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr + y_stride, |
498 | linesize, s, 1); |
499 | } |
500 | } |
501 | |
502 | if (vs_total_ac_bits < vs_bit_size) |
503 | dv_guess_qnos(&enc_blks[0], qnosp); |
504 | |
505 | /* DIF encoding process */ |
506 | for (j = 0; j < 5 * s->sys->bpm;) { |
507 | int start_mb = j; |
508 | |
509 | dif[3] = *qnosp++; |
510 | dif += 4; |
511 | |
512 | /* First pass over individual cells only */ |
513 | for (i = 0; i < s->sys->bpm; i++, j++) { |
514 | int sz = s->sys->block_sizes[i] >> 3; |
515 | |
516 | init_put_bits(&pbs[j], dif, sz); |
517 | put_sbits(&pbs[j], 9, ((enc_blks[j].mb[0] >> 3) - 1024 + 2) >> 2); |
518 | put_bits(&pbs[j], 1, enc_blks[j].dct_mode); |
519 | put_bits(&pbs[j], 2, enc_blks[j].cno); |
520 | |
521 | dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j + 1]); |
522 | dif += sz; |
523 | } |
524 | |
525 | /* Second pass over each MB space */ |
526 | pb = &pbs[start_mb]; |
527 | for (i = 0; i < s->sys->bpm; i++) |
528 | if (enc_blks[start_mb + i].partial_bit_count) |
529 | pb = dv_encode_ac(&enc_blks[start_mb + i], pb, |
530 | &pbs[start_mb + s->sys->bpm]); |
531 | } |
532 | |
533 | /* Third and final pass over the whole video segment space */ |
534 | pb = &pbs[0]; |
535 | for (j = 0; j < 5 * s->sys->bpm; j++) { |
536 | if (enc_blks[j].partial_bit_count) |
537 | pb = dv_encode_ac(&enc_blks[j], pb, &pbs[s->sys->bpm * 5]); |
538 | if (enc_blks[j].partial_bit_count) |
539 | av_log(avctx, AV_LOG_ERROR, "ac bitstream overflow\n"); |
540 | } |
541 | |
542 | for (j = 0; j < 5 * s->sys->bpm; j++) { |
543 | int pos; |
544 | int size = pbs[j].size_in_bits >> 3; |
545 | flush_put_bits(&pbs[j]); |
546 | pos = put_bits_count(&pbs[j]) >> 3; |
547 | if (pos > size) { |
548 | av_log(avctx, AV_LOG_ERROR, |
549 | "bitstream written beyond buffer size\n"); |
550 | return -1; |
551 | } |
552 | memset(pbs[j].buf + pos, 0xff, size - pos); |
553 | } |
554 | |
555 | return 0; |
556 | } |
557 | |
558 | static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, |
559 | uint8_t *buf) |
560 | { |
561 | /* |
562 | * Here's what SMPTE314M says about these two: |
563 | * (page 6) APTn, AP1n, AP2n, AP3n: These data shall be identical |
564 | * as track application IDs (APTn = 001, AP1n = |
565 | * 001, AP2n = 001, AP3n = 001), if the source signal |
566 | * comes from a digital VCR. If the signal source is |
567 | * unknown, all bits for these data shall be set to 1. |
568 | * (page 12) STYPE: STYPE defines a signal type of video signal |
569 | * 00000b = 4:1:1 compression |
570 | * 00100b = 4:2:2 compression |
571 | * XXXXXX = Reserved |
572 | * Now, I've got two problems with these statements: |
573 | * 1. it looks like APT == 111b should be a safe bet, but it isn't. |
574 | * It seems that for PAL as defined in IEC 61834 we have to set |
575 | * APT to 000 and for SMPTE314M to 001. |
576 | * 2. It is not at all clear what STYPE is used for 4:2:0 PAL |
577 | * compression scheme (if any). |
578 | */ |
579 | int apt = (c->sys->pix_fmt == AV_PIX_FMT_YUV420P ? 0 : 1); |
580 | int fs = c->frame->top_field_first ? 0x00 : 0x40; |
581 | |
582 | uint8_t aspect = 0; |
583 | if ((int) (av_q2d(c->avctx->sample_aspect_ratio) * |
584 | c->avctx->width / c->avctx->height * 10) >= 17) /* 16:9 */ |
585 | aspect = 0x02; |
586 | |
587 | buf[0] = (uint8_t) pack_id; |
588 | switch (pack_id) { |
589 | case dv_header525: /* I can't imagine why these two weren't defined as real */ |
590 | case dv_header625: /* packs in SMPTE314M -- they definitely look like ones */ |
591 | buf[1] = 0xf8 | /* reserved -- always 1 */ |
592 | (apt & 0x07); /* APT: Track application ID */ |
593 | buf[2] = (0 << 7) | /* TF1: audio data is 0 - valid; 1 - invalid */ |
594 | (0x0f << 3) | /* reserved -- always 1 */ |
595 | (apt & 0x07); /* AP1: Audio application ID */ |
596 | buf[3] = (0 << 7) | /* TF2: video data is 0 - valid; 1 - invalid */ |
597 | (0x0f << 3) | /* reserved -- always 1 */ |
598 | (apt & 0x07); /* AP2: Video application ID */ |
599 | buf[4] = (0 << 7) | /* TF3: subcode(SSYB) is 0 - valid; 1 - invalid */ |
600 | (0x0f << 3) | /* reserved -- always 1 */ |
601 | (apt & 0x07); /* AP3: Subcode application ID */ |
602 | break; |
603 | case dv_video_source: |
604 | buf[1] = 0xff; /* reserved -- always 1 */ |
605 | buf[2] = (1 << 7) | /* B/W: 0 - b/w, 1 - color */ |
606 | (1 << 6) | /* following CLF is valid - 0, invalid - 1 */ |
607 | (3 << 4) | /* CLF: color frames ID (see ITU-R BT.470-4) */ |
608 | 0xf; /* reserved -- always 1 */ |
609 | buf[3] = (3 << 6) | /* reserved -- always 1 */ |
610 | (c->sys->dsf << 5) | /* system: 60fields/50fields */ |
611 | c->sys->video_stype; /* signal type video compression */ |
612 | buf[4] = 0xff; /* VISC: 0xff -- no information */ |
613 | break; |
614 | case dv_video_control: |
615 | buf[1] = (0 << 6) | /* Copy generation management (CGMS) 0 -- free */ |
616 | 0x3f; /* reserved -- always 1 */ |
617 | buf[2] = 0xc8 | /* reserved -- always b11001xxx */ |
618 | aspect; |
619 | buf[3] = (1 << 7) | /* frame/field flag 1 -- frame, 0 -- field */ |
620 | fs | /* first/second field flag 0 -- field 2, 1 -- field 1 */ |
621 | (1 << 5) | /* frame change flag 0 -- same picture as before, 1 -- different */ |
622 | (1 << 4) | /* 1 - interlaced, 0 - noninterlaced */ |
623 | 0xc; /* reserved -- always b1100 */ |
624 | buf[4] = 0xff; /* reserved -- always 1 */ |
625 | break; |
626 | default: |
627 | buf[1] = |
628 | buf[2] = |
629 | buf[3] = |
630 | buf[4] = 0xff; |
631 | } |
632 | return 5; |
633 | } |
634 | |
635 | static inline int dv_write_dif_id(enum dv_section_type t, uint8_t chan_num, |
636 | uint8_t seq_num, uint8_t dif_num, |
637 | uint8_t *buf) |
638 | { |
639 | buf[0] = (uint8_t) t; /* Section type */ |
640 | buf[1] = (seq_num << 4) | /* DIF seq number 0-9 for 525/60; 0-11 for 625/50 */ |
641 | (chan_num << 3) | /* FSC: for 50Mb/s 0 - first channel; 1 - second */ |
642 | 7; /* reserved -- always 1 */ |
643 | buf[2] = dif_num; /* DIF block number Video: 0-134, Audio: 0-8 */ |
644 | return 3; |
645 | } |
646 | |
647 | static inline int dv_write_ssyb_id(uint8_t syb_num, uint8_t fr, uint8_t *buf) |
648 | { |
649 | if (syb_num == 0 || syb_num == 6) { |
650 | buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ |
651 | (0 << 4) | /* AP3 (Subcode application ID) */ |
652 | 0x0f; /* reserved -- always 1 */ |
653 | } else if (syb_num == 11) { |
654 | buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ |
655 | 0x7f; /* reserved -- always 1 */ |
656 | } else { |
657 | buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ |
658 | (0 << 4) | /* APT (Track application ID) */ |
659 | 0x0f; /* reserved -- always 1 */ |
660 | } |
661 | buf[1] = 0xf0 | /* reserved -- always 1 */ |
662 | (syb_num & 0x0f); /* SSYB number 0 - 11 */ |
663 | buf[2] = 0xff; /* reserved -- always 1 */ |
664 | return 3; |
665 | } |
666 | |
667 | static void dv_format_frame(DVVideoContext *c, uint8_t *buf) |
668 | { |
669 | int chan, i, j, k; |
670 | |
671 | for (chan = 0; chan < c->sys->n_difchan; chan++) { |
672 | for (i = 0; i < c->sys->difseg_size; i++) { |
673 | memset(buf, 0xff, 80 * 6); /* first 6 DIF blocks are for control data */ |
674 | |
675 | /* DV header: 1DIF */ |
676 | buf += dv_write_dif_id(dv_sect_header, chan, i, 0, buf); |
677 | buf += dv_write_pack((c->sys->dsf ? dv_header625 : dv_header525), |
678 | c, buf); |
679 | buf += 72; /* unused bytes */ |
680 | |
681 | /* DV subcode: 2DIFs */ |
682 | for (j = 0; j < 2; j++) { |
683 | buf += dv_write_dif_id(dv_sect_subcode, chan, i, j, buf); |
684 | for (k = 0; k < 6; k++) |
685 | buf += dv_write_ssyb_id(k, (i < c->sys->difseg_size / 2), buf) + 5; |
686 | buf += 29; /* unused bytes */ |
687 | } |
688 | |
689 | /* DV VAUX: 3DIFS */ |
690 | for (j = 0; j < 3; j++) { |
691 | buf += dv_write_dif_id(dv_sect_vaux, chan, i, j, buf); |
692 | buf += dv_write_pack(dv_video_source, c, buf); |
693 | buf += dv_write_pack(dv_video_control, c, buf); |
694 | buf += 7 * 5; |
695 | buf += dv_write_pack(dv_video_source, c, buf); |
696 | buf += dv_write_pack(dv_video_control, c, buf); |
697 | buf += 4 * 5 + 2; /* unused bytes */ |
698 | } |
699 | |
700 | /* DV Audio/Video: 135 Video DIFs + 9 Audio DIFs */ |
701 | for (j = 0; j < 135; j++) { |
702 | if (j % 15 == 0) { |
703 | memset(buf, 0xff, 80); |
704 | buf += dv_write_dif_id(dv_sect_audio, chan, i, j / 15, buf); |
705 | buf += 77; /* audio control & shuffled PCM audio */ |
706 | } |
707 | buf += dv_write_dif_id(dv_sect_video, chan, i, j, buf); |
708 | buf += 77; /* 1 video macroblock: 1 bytes control |
709 | * 4 * 14 bytes Y 8x8 data |
710 | * 10 bytes Cr 8x8 data |
711 | * 10 bytes Cb 8x8 data */ |
712 | } |
713 | } |
714 | } |
715 | } |
716 | |
717 | static int dvvideo_encode_frame(AVCodecContext *c, AVPacket *pkt, |
718 | const AVFrame *frame, int *got_packet) |
719 | { |
720 | DVVideoContext *s = c->priv_data; |
721 | int ret; |
722 | |
723 | if ((ret = ff_alloc_packet2(c, pkt, s->sys->frame_size, 0)) < 0) |
724 | return ret; |
725 | |
726 | c->pix_fmt = s->sys->pix_fmt; |
727 | s->frame = frame; |
728 | #if FF_API_CODED_FRAME |
729 | FF_DISABLE_DEPRECATION_WARNINGS |
730 | c->coded_frame->key_frame = 1; |
731 | c->coded_frame->pict_type = AV_PICTURE_TYPE_I; |
732 | FF_ENABLE_DEPRECATION_WARNINGS |
733 | #endif |
734 | |
735 | s->buf = pkt->data; |
736 | c->execute(c, dv_encode_video_segment, s->work_chunks, NULL, |
737 | dv_work_pool_size(s->sys), sizeof(DVwork_chunk)); |
738 | |
739 | emms_c(); |
740 | |
741 | dv_format_frame(s, pkt->data); |
742 | |
743 | pkt->flags |= AV_PKT_FLAG_KEY; |
744 | *got_packet = 1; |
745 | |
746 | return 0; |
747 | } |
748 | |
749 | AVCodec ff_dvvideo_encoder = { |
750 | .name = "dvvideo", |
751 | .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"), |
752 | .type = AVMEDIA_TYPE_VIDEO, |
753 | .id = AV_CODEC_ID_DVVIDEO, |
754 | .priv_data_size = sizeof(DVVideoContext), |
755 | .init = dvvideo_encode_init, |
756 | .encode2 = dvvideo_encode_frame, |
757 | .capabilities = AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_INTRA_ONLY, |
758 | .pix_fmts = (const enum AVPixelFormat[]) { |
759 | AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV422P, |
760 | AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE |
761 | }, |
762 | }; |
763 |