| 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/pixdesc.h" |
| 31 | |
| 32 | #include "avcodec.h" |
| 33 | #include "dv.h" |
| 34 | #include "dv_profile_internal.h" |
| 35 | #include "dv_tablegen.h" |
| 36 | #include "fdctdsp.h" |
| 37 | #include "internal.h" |
| 38 | #include "me_cmp.h" |
| 39 | #include "pixblockdsp.h" |
| 40 | #include "put_bits.h" |
| 41 | |
| 42 | static av_cold int dvvideo_encode_init(AVCodecContext *avctx) |
| 43 | { |
| 44 | DVVideoContext *s = avctx->priv_data; |
| 45 | FDCTDSPContext fdsp; |
| 46 | MECmpContext mecc; |
| 47 | PixblockDSPContext pdsp; |
| 48 | int ret; |
| 49 | |
| 50 | s->sys = av_dv_codec_profile(avctx->width, avctx->height, avctx->pix_fmt); |
| 51 | if (!s->sys) { |
| 52 | av_log(avctx, AV_LOG_ERROR, "Found no DV profile for %ix%i %s video. " |
| 53 | "Valid DV profiles are:\n", |
| 54 | avctx->width, avctx->height, av_get_pix_fmt_name(avctx->pix_fmt)); |
| 55 | ff_dv_print_profiles(avctx, AV_LOG_ERROR); |
| 56 | return AVERROR(EINVAL); |
| 57 | } |
| 58 | if (avctx->height > 576) { |
| 59 | av_log(avctx, AV_LOG_ERROR, "DVCPRO HD encoding is not supported.\n"); |
| 60 | return AVERROR_PATCHWELCOME; |
| 61 | } |
| 62 | ret = ff_dv_init_dynamic_tables(s, s->sys); |
| 63 | if (ret < 0) { |
| 64 | av_log(avctx, AV_LOG_ERROR, "Error initializing work tables.\n"); |
| 65 | return ret; |
| 66 | } |
| 67 | |
| 68 | avctx->coded_frame = av_frame_alloc(); |
| 69 | if (!avctx->coded_frame) |
| 70 | return AVERROR(ENOMEM); |
| 71 | |
| 72 | dv_vlc_map_tableinit(); |
| 73 | |
| 74 | memset(&fdsp,0, sizeof(fdsp)); |
| 75 | memset(&mecc,0, sizeof(mecc)); |
| 76 | memset(&pdsp,0, sizeof(pdsp)); |
| 77 | ff_fdctdsp_init(&fdsp, avctx); |
| 78 | ff_me_cmp_init(&mecc, avctx); |
| 79 | ff_pixblockdsp_init(&pdsp, avctx); |
| 80 | ff_set_cmp(&mecc, mecc.ildct_cmp, avctx->ildct_cmp); |
| 81 | |
| 82 | s->get_pixels = pdsp.get_pixels; |
| 83 | s->ildct_cmp = mecc.ildct_cmp[5]; |
| 84 | |
| 85 | s->fdct[0] = fdsp.fdct; |
| 86 | s->fdct[1] = fdsp.fdct248; |
| 87 | |
| 88 | return ff_dvvideo_init(avctx); |
| 89 | } |
| 90 | |
| 91 | /* bit budget for AC only in 5 MBs */ |
| 92 | static const int vs_total_ac_bits = (100 * 4 + 68 * 2) * 5; |
| 93 | static const int mb_area_start[5] = { 1, 6, 21, 43, 64 }; |
| 94 | |
| 95 | #if CONFIG_SMALL |
| 96 | /* Convert run and level (where level != 0) pair into VLC, returning bit size */ |
| 97 | static av_always_inline int dv_rl2vlc(int run, int level, int sign, |
| 98 | uint32_t *vlc) |
| 99 | { |
| 100 | int size; |
| 101 | if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) { |
| 102 | *vlc = dv_vlc_map[run][level].vlc | sign; |
| 103 | size = dv_vlc_map[run][level].size; |
| 104 | } else { |
| 105 | if (level < DV_VLC_MAP_LEV_SIZE) { |
| 106 | *vlc = dv_vlc_map[0][level].vlc | sign; |
| 107 | size = dv_vlc_map[0][level].size; |
| 108 | } else { |
| 109 | *vlc = 0xfe00 | (level << 1) | sign; |
| 110 | size = 16; |
| 111 | } |
| 112 | if (run) { |
| 113 | *vlc |= ((run < 16) ? dv_vlc_map[run - 1][0].vlc : |
| 114 | (0x1f80 | (run - 1))) << size; |
| 115 | size += (run < 16) ? dv_vlc_map[run - 1][0].size : 13; |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | return size; |
| 120 | } |
| 121 | |
| 122 | static av_always_inline int dv_rl2vlc_size(int run, int level) |
| 123 | { |
| 124 | int size; |
| 125 | |
| 126 | if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) { |
| 127 | size = dv_vlc_map[run][level].size; |
| 128 | } else { |
| 129 | size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16; |
| 130 | if (run) |
| 131 | size += (run < 16) ? dv_vlc_map[run - 1][0].size : 13; |
| 132 | } |
| 133 | return size; |
| 134 | } |
| 135 | #else |
| 136 | static av_always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t *vlc) |
| 137 | { |
| 138 | *vlc = dv_vlc_map[run][l].vlc | sign; |
| 139 | return dv_vlc_map[run][l].size; |
| 140 | } |
| 141 | |
| 142 | static av_always_inline int dv_rl2vlc_size(int run, int l) |
| 143 | { |
| 144 | return dv_vlc_map[run][l].size; |
| 145 | } |
| 146 | #endif |
| 147 | |
| 148 | typedef struct EncBlockInfo { |
| 149 | int area_q[4]; |
| 150 | int bit_size[4]; |
| 151 | int prev[5]; |
| 152 | int cur_ac; |
| 153 | int cno; |
| 154 | int dct_mode; |
| 155 | int16_t mb[64]; |
| 156 | uint8_t next[64]; |
| 157 | uint8_t sign[64]; |
| 158 | uint8_t partial_bit_count; |
| 159 | uint32_t partial_bit_buffer; /* we can't use uint16_t here */ |
| 160 | } EncBlockInfo; |
| 161 | |
| 162 | static av_always_inline PutBitContext *dv_encode_ac(EncBlockInfo *bi, |
| 163 | PutBitContext *pb_pool, |
| 164 | PutBitContext *pb_end) |
| 165 | { |
| 166 | int prev, bits_left; |
| 167 | PutBitContext *pb = pb_pool; |
| 168 | int size = bi->partial_bit_count; |
| 169 | uint32_t vlc = bi->partial_bit_buffer; |
| 170 | |
| 171 | bi->partial_bit_count = |
| 172 | bi->partial_bit_buffer = 0; |
| 173 | for (;;) { |
| 174 | /* Find suitable storage space */ |
| 175 | for (; size > (bits_left = put_bits_left(pb)); pb++) { |
| 176 | if (bits_left) { |
| 177 | size -= bits_left; |
| 178 | put_bits(pb, bits_left, vlc >> size); |
| 179 | vlc = vlc & ((1 << size) - 1); |
| 180 | } |
| 181 | if (pb + 1 >= pb_end) { |
| 182 | bi->partial_bit_count = size; |
| 183 | bi->partial_bit_buffer = vlc; |
| 184 | return pb; |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | /* Store VLC */ |
| 189 | put_bits(pb, size, vlc); |
| 190 | |
| 191 | if (bi->cur_ac >= 64) |
| 192 | break; |
| 193 | |
| 194 | /* Construct the next VLC */ |
| 195 | prev = bi->cur_ac; |
| 196 | bi->cur_ac = bi->next[prev]; |
| 197 | if (bi->cur_ac < 64) { |
| 198 | size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], |
| 199 | bi->sign[bi->cur_ac], &vlc); |
| 200 | } else { |
| 201 | size = 4; |
| 202 | vlc = 6; /* End Of Block stamp */ |
| 203 | } |
| 204 | } |
| 205 | return pb; |
| 206 | } |
| 207 | |
| 208 | static av_always_inline int dv_guess_dct_mode(DVVideoContext *s, uint8_t *data, |
| 209 | int linesize) |
| 210 | { |
| 211 | if (s->avctx->flags & CODEC_FLAG_INTERLACED_DCT) { |
| 212 | int ps = s->ildct_cmp(NULL, data, NULL, linesize, 8) - 400; |
| 213 | if (ps > 0) { |
| 214 | int is = s->ildct_cmp(NULL, data, NULL, linesize << 1, 4) + |
| 215 | s->ildct_cmp(NULL, data + linesize, NULL, linesize << 1, 4); |
| 216 | return ps > is; |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | return 0; |
| 221 | } |
| 222 | |
| 223 | static const int dv_weight_bits = 18; |
| 224 | static const int dv_weight_88[64] = { |
| 225 | 131072, 257107, 257107, 242189, 252167, 242189, 235923, 237536, |
| 226 | 237536, 235923, 229376, 231390, 223754, 231390, 229376, 222935, |
| 227 | 224969, 217965, 217965, 224969, 222935, 200636, 218652, 211916, |
| 228 | 212325, 211916, 218652, 200636, 188995, 196781, 205965, 206433, |
| 229 | 206433, 205965, 196781, 188995, 185364, 185364, 200636, 200704, |
| 230 | 200636, 185364, 185364, 174609, 180568, 195068, 195068, 180568, |
| 231 | 174609, 170091, 175557, 189591, 175557, 170091, 165371, 170627, |
| 232 | 170627, 165371, 160727, 153560, 160727, 144651, 144651, 136258, |
| 233 | }; |
| 234 | static const int dv_weight_248[64] = { |
| 235 | 131072, 242189, 257107, 237536, 229376, 200636, 242189, 223754, |
| 236 | 224969, 196781, 262144, 242189, 229376, 200636, 257107, 237536, |
| 237 | 211916, 185364, 235923, 217965, 229376, 211916, 206433, 180568, |
| 238 | 242189, 223754, 224969, 196781, 211916, 185364, 235923, 217965, |
| 239 | 200704, 175557, 222935, 205965, 200636, 185364, 195068, 170627, |
| 240 | 229376, 211916, 206433, 180568, 200704, 175557, 222935, 205965, |
| 241 | 175557, 153560, 188995, 174609, 165371, 144651, 200636, 185364, |
| 242 | 195068, 170627, 175557, 153560, 188995, 174609, 165371, 144651, |
| 243 | }; |
| 244 | |
| 245 | static av_always_inline int dv_init_enc_block(EncBlockInfo *bi, uint8_t *data, |
| 246 | int linesize, DVVideoContext *s, |
| 247 | int bias) |
| 248 | { |
| 249 | const int *weight; |
| 250 | const uint8_t *zigzag_scan; |
| 251 | LOCAL_ALIGNED_16(int16_t, blk, [64]); |
| 252 | int i, area; |
| 253 | /* We offer two different methods for class number assignment: the |
| 254 | * method suggested in SMPTE 314M Table 22, and an improved |
| 255 | * method. The SMPTE method is very conservative; it assigns class |
| 256 | * 3 (i.e. severe quantization) to any block where the largest AC |
| 257 | * component is greater than 36. FFmpeg's DV encoder tracks AC bit |
| 258 | * consumption precisely, so there is no need to bias most blocks |
| 259 | * towards strongly lossy compression. Instead, we assign class 2 |
| 260 | * to most blocks, and use class 3 only when strictly necessary |
| 261 | * (for blocks whose largest AC component exceeds 255). */ |
| 262 | |
| 263 | #if 0 /* SMPTE spec method */ |
| 264 | static const int classes[] = { 12, 24, 36, 0xffff }; |
| 265 | #else /* improved FFmpeg method */ |
| 266 | static const int classes[] = { -1, -1, 255, 0xffff }; |
| 267 | #endif |
| 268 | int max = classes[0]; |
| 269 | int prev = 0; |
| 270 | |
| 271 | av_assert2((((int) blk) & 15) == 0); |
| 272 | |
| 273 | bi->area_q[0] = |
| 274 | bi->area_q[1] = |
| 275 | bi->area_q[2] = |
| 276 | bi->area_q[3] = 0; |
| 277 | bi->partial_bit_count = 0; |
| 278 | bi->partial_bit_buffer = 0; |
| 279 | bi->cur_ac = 0; |
| 280 | if (data) { |
| 281 | bi->dct_mode = dv_guess_dct_mode(s, data, linesize); |
| 282 | s->get_pixels(blk, data, linesize); |
| 283 | s->fdct[bi->dct_mode](blk); |
| 284 | } else { |
| 285 | /* We rely on the fact that encoding all zeros leads to an immediate |
| 286 | * EOB, which is precisely what the spec calls for in the "dummy" |
| 287 | * blocks. */ |
| 288 | memset(blk, 0, 64 * sizeof(*blk)); |
| 289 | bi->dct_mode = 0; |
| 290 | } |
| 291 | bi->mb[0] = blk[0]; |
| 292 | |
| 293 | zigzag_scan = bi->dct_mode ? ff_dv_zigzag248_direct : ff_zigzag_direct; |
| 294 | weight = bi->dct_mode ? dv_weight_248 : dv_weight_88; |
| 295 | |
| 296 | for (area = 0; area < 4; area++) { |
| 297 | bi->prev[area] = prev; |
| 298 | bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) |
| 299 | for (i = mb_area_start[area]; i < mb_area_start[area + 1]; i++) { |
| 300 | int level = blk[zigzag_scan[i]]; |
| 301 | |
| 302 | if (level + 15 > 30U) { |
| 303 | bi->sign[i] = (level >> 31) & 1; |
| 304 | /* Weight it and shift down into range, adding for rounding. |
| 305 | * The extra division by a factor of 2^4 reverses the 8x |
| 306 | * expansion of the DCT AND the 2x doubling of the weights. */ |
| 307 | level = (FFABS(level) * weight[i] + (1 << (dv_weight_bits + 3))) >> |
| 308 | (dv_weight_bits + 4); |
| 309 | bi->mb[i] = level; |
| 310 | if (level > max) |
| 311 | max = level; |
| 312 | bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, level); |
| 313 | bi->next[prev] = i; |
| 314 | prev = i; |
| 315 | } |
| 316 | } |
| 317 | } |
| 318 | bi->next[prev] = i; |
| 319 | for (bi->cno = 0; max > classes[bi->cno]; bi->cno++) |
| 320 | ; |
| 321 | |
| 322 | bi->cno += bias; |
| 323 | |
| 324 | if (bi->cno >= 3) { |
| 325 | bi->cno = 3; |
| 326 | prev = 0; |
| 327 | i = bi->next[prev]; |
| 328 | for (area = 0; area < 4; area++) { |
| 329 | bi->prev[area] = prev; |
| 330 | bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :) |
| 331 | for (; i < mb_area_start[area + 1]; i = bi->next[i]) { |
| 332 | bi->mb[i] >>= 1; |
| 333 | |
| 334 | if (bi->mb[i]) { |
| 335 | bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]); |
| 336 | bi->next[prev] = i; |
| 337 | prev = i; |
| 338 | } |
| 339 | } |
| 340 | } |
| 341 | bi->next[prev] = i; |
| 342 | } |
| 343 | |
| 344 | return bi->bit_size[0] + bi->bit_size[1] + |
| 345 | bi->bit_size[2] + bi->bit_size[3]; |
| 346 | } |
| 347 | |
| 348 | static inline void dv_guess_qnos(EncBlockInfo *blks, int *qnos) |
| 349 | { |
| 350 | int size[5]; |
| 351 | int i, j, k, a, prev, a2; |
| 352 | EncBlockInfo *b; |
| 353 | |
| 354 | size[0] = |
| 355 | size[1] = |
| 356 | size[2] = |
| 357 | size[3] = |
| 358 | size[4] = 1 << 24; |
| 359 | do { |
| 360 | b = blks; |
| 361 | for (i = 0; i < 5; i++) { |
| 362 | if (!qnos[i]) |
| 363 | continue; |
| 364 | |
| 365 | qnos[i]--; |
| 366 | size[i] = 0; |
| 367 | for (j = 0; j < 6; j++, b++) { |
| 368 | for (a = 0; a < 4; a++) { |
| 369 | if (b->area_q[a] != ff_dv_quant_shifts[qnos[i] + ff_dv_quant_offset[b->cno]][a]) { |
| 370 | b->bit_size[a] = 1; // 4 areas 4 bits for EOB :) |
| 371 | b->area_q[a]++; |
| 372 | prev = b->prev[a]; |
| 373 | av_assert2(b->next[prev] >= mb_area_start[a + 1] || b->mb[prev]); |
| 374 | for (k = b->next[prev]; k < mb_area_start[a + 1]; k = b->next[k]) { |
| 375 | b->mb[k] >>= 1; |
| 376 | if (b->mb[k]) { |
| 377 | b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]); |
| 378 | prev = k; |
| 379 | } else { |
| 380 | if (b->next[k] >= mb_area_start[a + 1] && b->next[k] < 64) { |
| 381 | for (a2 = a + 1; b->next[k] >= mb_area_start[a2 + 1]; a2++) |
| 382 | b->prev[a2] = prev; |
| 383 | av_assert2(a2 < 4); |
| 384 | av_assert2(b->mb[b->next[k]]); |
| 385 | b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]]) - |
| 386 | dv_rl2vlc_size(b->next[k] - k - 1, b->mb[b->next[k]]); |
| 387 | av_assert2(b->prev[a2] == k && (a2 + 1 >= 4 || b->prev[a2 + 1] != k)); |
| 388 | b->prev[a2] = prev; |
| 389 | } |
| 390 | b->next[prev] = b->next[k]; |
| 391 | } |
| 392 | } |
| 393 | b->prev[a + 1] = prev; |
| 394 | } |
| 395 | size[i] += b->bit_size[a]; |
| 396 | } |
| 397 | } |
| 398 | if (vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4]) |
| 399 | return; |
| 400 | } |
| 401 | } while (qnos[0] | qnos[1] | qnos[2] | qnos[3] | qnos[4]); |
| 402 | |
| 403 | for (a = 2; a == 2 || vs_total_ac_bits < size[0]; a += a) { |
| 404 | b = blks; |
| 405 | size[0] = 5 * 6 * 4; // EOB |
| 406 | for (j = 0; j < 6 * 5; j++, b++) { |
| 407 | prev = b->prev[0]; |
| 408 | for (k = b->next[prev]; k < 64; k = b->next[k]) { |
| 409 | if (b->mb[k] < a && b->mb[k] > -a) { |
| 410 | b->next[prev] = b->next[k]; |
| 411 | } else { |
| 412 | size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]); |
| 413 | prev = k; |
| 414 | } |
| 415 | } |
| 416 | } |
| 417 | } |
| 418 | } |
| 419 | |
| 420 | static int dv_encode_video_segment(AVCodecContext *avctx, void *arg) |
| 421 | { |
| 422 | DVVideoContext *s = avctx->priv_data; |
| 423 | DVwork_chunk *work_chunk = arg; |
| 424 | int mb_index, i, j; |
| 425 | int mb_x, mb_y, c_offset, linesize, y_stride; |
| 426 | uint8_t *y_ptr; |
| 427 | uint8_t *dif; |
| 428 | LOCAL_ALIGNED_8(uint8_t, scratch, [128]); |
| 429 | EncBlockInfo enc_blks[5 * DV_MAX_BPM]; |
| 430 | PutBitContext pbs[5 * DV_MAX_BPM]; |
| 431 | PutBitContext *pb; |
| 432 | EncBlockInfo *enc_blk; |
| 433 | int vs_bit_size = 0; |
| 434 | int qnos[5] = { 15, 15, 15, 15, 15 }; /* No quantization */ |
| 435 | int *qnosp = &qnos[0]; |
| 436 | |
| 437 | dif = &s->buf[work_chunk->buf_offset * 80]; |
| 438 | enc_blk = &enc_blks[0]; |
| 439 | for (mb_index = 0; mb_index < 5; mb_index++) { |
| 440 | dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y); |
| 441 | |
| 442 | /* initializing luminance blocks */ |
| 443 | if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) || |
| 444 | (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) || |
| 445 | (s->sys->height >= 720 && mb_y != 134)) { |
| 446 | y_stride = s->frame->linesize[0] << 3; |
| 447 | } else { |
| 448 | y_stride = 16; |
| 449 | } |
| 450 | y_ptr = s->frame->data[0] + |
| 451 | ((mb_y * s->frame->linesize[0] + mb_x) << 3); |
| 452 | linesize = s->frame->linesize[0]; |
| 453 | |
| 454 | if (s->sys->video_stype == 4) { /* SD 422 */ |
| 455 | vs_bit_size += |
| 456 | dv_init_enc_block(enc_blk + 0, y_ptr, linesize, s, 0) + |
| 457 | dv_init_enc_block(enc_blk + 1, NULL, linesize, s, 0) + |
| 458 | dv_init_enc_block(enc_blk + 2, y_ptr + 8, linesize, s, 0) + |
| 459 | dv_init_enc_block(enc_blk + 3, NULL, linesize, s, 0); |
| 460 | } else { |
| 461 | vs_bit_size += |
| 462 | dv_init_enc_block(enc_blk + 0, y_ptr, linesize, s, 0) + |
| 463 | dv_init_enc_block(enc_blk + 1, y_ptr + 8, linesize, s, 0) + |
| 464 | dv_init_enc_block(enc_blk + 2, y_ptr + y_stride, linesize, s, 0) + |
| 465 | dv_init_enc_block(enc_blk + 3, y_ptr + 8 + y_stride, linesize, s, 0); |
| 466 | } |
| 467 | enc_blk += 4; |
| 468 | |
| 469 | /* initializing chrominance blocks */ |
| 470 | c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] + |
| 471 | (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << 3); |
| 472 | for (j = 2; j; j--) { |
| 473 | uint8_t *c_ptr = s->frame->data[j] + c_offset; |
| 474 | linesize = s->frame->linesize[j]; |
| 475 | y_stride = (mb_y == 134) ? 8 : (s->frame->linesize[j] << 3); |
| 476 | if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) { |
| 477 | uint8_t *d; |
| 478 | uint8_t *b = scratch; |
| 479 | for (i = 0; i < 8; i++) { |
| 480 | d = c_ptr + (linesize << 3); |
| 481 | b[0] = c_ptr[0]; |
| 482 | b[1] = c_ptr[1]; |
| 483 | b[2] = c_ptr[2]; |
| 484 | b[3] = c_ptr[3]; |
| 485 | b[4] = d[0]; |
| 486 | b[5] = d[1]; |
| 487 | b[6] = d[2]; |
| 488 | b[7] = d[3]; |
| 489 | c_ptr += linesize; |
| 490 | b += 16; |
| 491 | } |
| 492 | c_ptr = scratch; |
| 493 | linesize = 16; |
| 494 | } |
| 495 | |
| 496 | vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr, linesize, s, 1); |
| 497 | if (s->sys->bpm == 8) |
| 498 | vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr + y_stride, |
| 499 | linesize, s, 1); |
| 500 | } |
| 501 | } |
| 502 | |
| 503 | if (vs_total_ac_bits < vs_bit_size) |
| 504 | dv_guess_qnos(&enc_blks[0], qnosp); |
| 505 | |
| 506 | /* DIF encoding process */ |
| 507 | for (j = 0; j < 5 * s->sys->bpm;) { |
| 508 | int start_mb = j; |
| 509 | |
| 510 | dif[3] = *qnosp++; |
| 511 | dif += 4; |
| 512 | |
| 513 | /* First pass over individual cells only */ |
| 514 | for (i = 0; i < s->sys->bpm; i++, j++) { |
| 515 | int sz = s->sys->block_sizes[i] >> 3; |
| 516 | |
| 517 | init_put_bits(&pbs[j], dif, sz); |
| 518 | put_sbits(&pbs[j], 9, ((enc_blks[j].mb[0] >> 3) - 1024 + 2) >> 2); |
| 519 | put_bits(&pbs[j], 1, enc_blks[j].dct_mode); |
| 520 | put_bits(&pbs[j], 2, enc_blks[j].cno); |
| 521 | |
| 522 | dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j + 1]); |
| 523 | dif += sz; |
| 524 | } |
| 525 | |
| 526 | /* Second pass over each MB space */ |
| 527 | pb = &pbs[start_mb]; |
| 528 | for (i = 0; i < s->sys->bpm; i++) |
| 529 | if (enc_blks[start_mb + i].partial_bit_count) |
| 530 | pb = dv_encode_ac(&enc_blks[start_mb + i], pb, |
| 531 | &pbs[start_mb + s->sys->bpm]); |
| 532 | } |
| 533 | |
| 534 | /* Third and final pass over the whole video segment space */ |
| 535 | pb = &pbs[0]; |
| 536 | for (j = 0; j < 5 * s->sys->bpm; j++) { |
| 537 | if (enc_blks[j].partial_bit_count) |
| 538 | pb = dv_encode_ac(&enc_blks[j], pb, &pbs[s->sys->bpm * 5]); |
| 539 | if (enc_blks[j].partial_bit_count) |
| 540 | av_log(avctx, AV_LOG_ERROR, "ac bitstream overflow\n"); |
| 541 | } |
| 542 | |
| 543 | for (j = 0; j < 5 * s->sys->bpm; j++) { |
| 544 | int pos; |
| 545 | int size = pbs[j].size_in_bits >> 3; |
| 546 | flush_put_bits(&pbs[j]); |
| 547 | pos = put_bits_count(&pbs[j]) >> 3; |
| 548 | if (pos > size) { |
| 549 | av_log(avctx, AV_LOG_ERROR, |
| 550 | "bitstream written beyond buffer size\n"); |
| 551 | return -1; |
| 552 | } |
| 553 | memset(pbs[j].buf + pos, 0xff, size - pos); |
| 554 | } |
| 555 | |
| 556 | return 0; |
| 557 | } |
| 558 | |
| 559 | static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c, |
| 560 | uint8_t *buf) |
| 561 | { |
| 562 | /* |
| 563 | * Here's what SMPTE314M says about these two: |
| 564 | * (page 6) APTn, AP1n, AP2n, AP3n: These data shall be identical |
| 565 | * as track application IDs (APTn = 001, AP1n = |
| 566 | * 001, AP2n = 001, AP3n = 001), if the source signal |
| 567 | * comes from a digital VCR. If the signal source is |
| 568 | * unknown, all bits for these data shall be set to 1. |
| 569 | * (page 12) STYPE: STYPE defines a signal type of video signal |
| 570 | * 00000b = 4:1:1 compression |
| 571 | * 00100b = 4:2:2 compression |
| 572 | * XXXXXX = Reserved |
| 573 | * Now, I've got two problems with these statements: |
| 574 | * 1. it looks like APT == 111b should be a safe bet, but it isn't. |
| 575 | * It seems that for PAL as defined in IEC 61834 we have to set |
| 576 | * APT to 000 and for SMPTE314M to 001. |
| 577 | * 2. It is not at all clear what STYPE is used for 4:2:0 PAL |
| 578 | * compression scheme (if any). |
| 579 | */ |
| 580 | int apt = (c->sys->pix_fmt == AV_PIX_FMT_YUV420P ? 0 : 1); |
| 581 | int fs = c->frame->top_field_first ? 0x00 : 0x40; |
| 582 | |
| 583 | uint8_t aspect = 0; |
| 584 | if ((int) (av_q2d(c->avctx->sample_aspect_ratio) * |
| 585 | c->avctx->width / c->avctx->height * 10) >= 17) /* 16:9 */ |
| 586 | aspect = 0x02; |
| 587 | |
| 588 | buf[0] = (uint8_t) pack_id; |
| 589 | switch (pack_id) { |
| 590 | case dv_header525: /* I can't imagine why these two weren't defined as real */ |
| 591 | case dv_header625: /* packs in SMPTE314M -- they definitely look like ones */ |
| 592 | buf[1] = 0xf8 | /* reserved -- always 1 */ |
| 593 | (apt & 0x07); /* APT: Track application ID */ |
| 594 | buf[2] = (0 << 7) | /* TF1: audio data is 0 - valid; 1 - invalid */ |
| 595 | (0x0f << 3) | /* reserved -- always 1 */ |
| 596 | (apt & 0x07); /* AP1: Audio application ID */ |
| 597 | buf[3] = (0 << 7) | /* TF2: video data is 0 - valid; 1 - invalid */ |
| 598 | (0x0f << 3) | /* reserved -- always 1 */ |
| 599 | (apt & 0x07); /* AP2: Video application ID */ |
| 600 | buf[4] = (0 << 7) | /* TF3: subcode(SSYB) is 0 - valid; 1 - invalid */ |
| 601 | (0x0f << 3) | /* reserved -- always 1 */ |
| 602 | (apt & 0x07); /* AP3: Subcode application ID */ |
| 603 | break; |
| 604 | case dv_video_source: |
| 605 | buf[1] = 0xff; /* reserved -- always 1 */ |
| 606 | buf[2] = (1 << 7) | /* B/W: 0 - b/w, 1 - color */ |
| 607 | (1 << 6) | /* following CLF is valid - 0, invalid - 1 */ |
| 608 | (3 << 4) | /* CLF: color frames ID (see ITU-R BT.470-4) */ |
| 609 | 0xf; /* reserved -- always 1 */ |
| 610 | buf[3] = (3 << 6) | /* reserved -- always 1 */ |
| 611 | (c->sys->dsf << 5) | /* system: 60fields/50fields */ |
| 612 | c->sys->video_stype; /* signal type video compression */ |
| 613 | buf[4] = 0xff; /* VISC: 0xff -- no information */ |
| 614 | break; |
| 615 | case dv_video_control: |
| 616 | buf[1] = (0 << 6) | /* Copy generation management (CGMS) 0 -- free */ |
| 617 | 0x3f; /* reserved -- always 1 */ |
| 618 | buf[2] = 0xc8 | /* reserved -- always b11001xxx */ |
| 619 | aspect; |
| 620 | buf[3] = (1 << 7) | /* frame/field flag 1 -- frame, 0 -- field */ |
| 621 | fs | /* first/second field flag 0 -- field 2, 1 -- field 1 */ |
| 622 | (1 << 5) | /* frame change flag 0 -- same picture as before, 1 -- different */ |
| 623 | (1 << 4) | /* 1 - interlaced, 0 - noninterlaced */ |
| 624 | 0xc; /* reserved -- always b1100 */ |
| 625 | buf[4] = 0xff; /* reserved -- always 1 */ |
| 626 | break; |
| 627 | default: |
| 628 | buf[1] = |
| 629 | buf[2] = |
| 630 | buf[3] = |
| 631 | buf[4] = 0xff; |
| 632 | } |
| 633 | return 5; |
| 634 | } |
| 635 | |
| 636 | static inline int dv_write_dif_id(enum dv_section_type t, uint8_t chan_num, |
| 637 | uint8_t seq_num, uint8_t dif_num, |
| 638 | uint8_t *buf) |
| 639 | { |
| 640 | buf[0] = (uint8_t) t; /* Section type */ |
| 641 | buf[1] = (seq_num << 4) | /* DIF seq number 0-9 for 525/60; 0-11 for 625/50 */ |
| 642 | (chan_num << 3) | /* FSC: for 50Mb/s 0 - first channel; 1 - second */ |
| 643 | 7; /* reserved -- always 1 */ |
| 644 | buf[2] = dif_num; /* DIF block number Video: 0-134, Audio: 0-8 */ |
| 645 | return 3; |
| 646 | } |
| 647 | |
| 648 | static inline int dv_write_ssyb_id(uint8_t syb_num, uint8_t fr, uint8_t *buf) |
| 649 | { |
| 650 | if (syb_num == 0 || syb_num == 6) { |
| 651 | buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ |
| 652 | (0 << 4) | /* AP3 (Subcode application ID) */ |
| 653 | 0x0f; /* reserved -- always 1 */ |
| 654 | } else if (syb_num == 11) { |
| 655 | buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ |
| 656 | 0x7f; /* reserved -- always 1 */ |
| 657 | } else { |
| 658 | buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */ |
| 659 | (0 << 4) | /* APT (Track application ID) */ |
| 660 | 0x0f; /* reserved -- always 1 */ |
| 661 | } |
| 662 | buf[1] = 0xf0 | /* reserved -- always 1 */ |
| 663 | (syb_num & 0x0f); /* SSYB number 0 - 11 */ |
| 664 | buf[2] = 0xff; /* reserved -- always 1 */ |
| 665 | return 3; |
| 666 | } |
| 667 | |
| 668 | static void dv_format_frame(DVVideoContext *c, uint8_t *buf) |
| 669 | { |
| 670 | int chan, i, j, k; |
| 671 | |
| 672 | for (chan = 0; chan < c->sys->n_difchan; chan++) { |
| 673 | for (i = 0; i < c->sys->difseg_size; i++) { |
| 674 | memset(buf, 0xff, 80 * 6); /* first 6 DIF blocks are for control data */ |
| 675 | |
| 676 | /* DV header: 1DIF */ |
| 677 | buf += dv_write_dif_id(dv_sect_header, chan, i, 0, buf); |
| 678 | buf += dv_write_pack((c->sys->dsf ? dv_header625 : dv_header525), |
| 679 | c, buf); |
| 680 | buf += 72; /* unused bytes */ |
| 681 | |
| 682 | /* DV subcode: 2DIFs */ |
| 683 | for (j = 0; j < 2; j++) { |
| 684 | buf += dv_write_dif_id(dv_sect_subcode, chan, i, j, buf); |
| 685 | for (k = 0; k < 6; k++) |
| 686 | buf += dv_write_ssyb_id(k, (i < c->sys->difseg_size / 2), buf) + 5; |
| 687 | buf += 29; /* unused bytes */ |
| 688 | } |
| 689 | |
| 690 | /* DV VAUX: 3DIFS */ |
| 691 | for (j = 0; j < 3; j++) { |
| 692 | buf += dv_write_dif_id(dv_sect_vaux, chan, i, j, buf); |
| 693 | buf += dv_write_pack(dv_video_source, c, buf); |
| 694 | buf += dv_write_pack(dv_video_control, c, buf); |
| 695 | buf += 7 * 5; |
| 696 | buf += dv_write_pack(dv_video_source, c, buf); |
| 697 | buf += dv_write_pack(dv_video_control, c, buf); |
| 698 | buf += 4 * 5 + 2; /* unused bytes */ |
| 699 | } |
| 700 | |
| 701 | /* DV Audio/Video: 135 Video DIFs + 9 Audio DIFs */ |
| 702 | for (j = 0; j < 135; j++) { |
| 703 | if (j % 15 == 0) { |
| 704 | memset(buf, 0xff, 80); |
| 705 | buf += dv_write_dif_id(dv_sect_audio, chan, i, j / 15, buf); |
| 706 | buf += 77; /* audio control & shuffled PCM audio */ |
| 707 | } |
| 708 | buf += dv_write_dif_id(dv_sect_video, chan, i, j, buf); |
| 709 | buf += 77; /* 1 video macroblock: 1 bytes control |
| 710 | * 4 * 14 bytes Y 8x8 data |
| 711 | * 10 bytes Cr 8x8 data |
| 712 | * 10 bytes Cb 8x8 data */ |
| 713 | } |
| 714 | } |
| 715 | } |
| 716 | } |
| 717 | |
| 718 | static int dvvideo_encode_frame(AVCodecContext *c, AVPacket *pkt, |
| 719 | const AVFrame *frame, int *got_packet) |
| 720 | { |
| 721 | DVVideoContext *s = c->priv_data; |
| 722 | int ret; |
| 723 | |
| 724 | if ((ret = ff_alloc_packet2(c, pkt, s->sys->frame_size)) < 0) |
| 725 | return ret; |
| 726 | |
| 727 | c->pix_fmt = s->sys->pix_fmt; |
| 728 | s->frame = frame; |
| 729 | c->coded_frame->key_frame = 1; |
| 730 | c->coded_frame->pict_type = AV_PICTURE_TYPE_I; |
| 731 | |
| 732 | s->buf = pkt->data; |
| 733 | c->execute(c, dv_encode_video_segment, s->work_chunks, NULL, |
| 734 | dv_work_pool_size(s->sys), sizeof(DVwork_chunk)); |
| 735 | |
| 736 | emms_c(); |
| 737 | |
| 738 | dv_format_frame(s, pkt->data); |
| 739 | |
| 740 | pkt->flags |= AV_PKT_FLAG_KEY; |
| 741 | *got_packet = 1; |
| 742 | |
| 743 | return 0; |
| 744 | } |
| 745 | |
| 746 | static int dvvideo_encode_close(AVCodecContext *avctx) |
| 747 | { |
| 748 | av_frame_free(&avctx->coded_frame); |
| 749 | return 0; |
| 750 | } |
| 751 | |
| 752 | AVCodec ff_dvvideo_encoder = { |
| 753 | .name = "dvvideo", |
| 754 | .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"), |
| 755 | .type = AVMEDIA_TYPE_VIDEO, |
| 756 | .id = AV_CODEC_ID_DVVIDEO, |
| 757 | .priv_data_size = sizeof(DVVideoContext), |
| 758 | .init = dvvideo_encode_init, |
| 759 | .encode2 = dvvideo_encode_frame, |
| 760 | .close = dvvideo_encode_close, |
| 761 | .capabilities = CODEC_CAP_SLICE_THREADS, |
| 762 | .pix_fmts = (const enum AVPixelFormat[]) { |
| 763 | AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV422P, |
| 764 | AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE |
| 765 | }, |
| 766 | }; |