| 1 | /* |
| 2 | * DV decoder |
| 3 | * Copyright (c) 2002 Fabrice Bellard |
| 4 | * Copyright (c) 2004 Roman Shaposhnik |
| 5 | * |
| 6 | * 50 Mbps (DVCPRO50) support |
| 7 | * Copyright (c) 2006 Daniel Maas <dmaas@maasdigital.com> |
| 8 | * |
| 9 | * 100 Mbps (DVCPRO HD) support |
| 10 | * Initial code by Daniel Maas <dmaas@maasdigital.com> (funded by BBC R&D) |
| 11 | * Final code by Roman Shaposhnik |
| 12 | * |
| 13 | * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth |
| 14 | * of DV technical info. |
| 15 | * |
| 16 | * This file is part of FFmpeg. |
| 17 | * |
| 18 | * FFmpeg is free software; you can redistribute it and/or |
| 19 | * modify it under the terms of the GNU Lesser General Public |
| 20 | * License as published by the Free Software Foundation; either |
| 21 | * version 2.1 of the License, or (at your option) any later version. |
| 22 | * |
| 23 | * FFmpeg is distributed in the hope that it will be useful, |
| 24 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 25 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 26 | * Lesser General Public License for more details. |
| 27 | * |
| 28 | * You should have received a copy of the GNU Lesser General Public |
| 29 | * License along with FFmpeg; if not, write to the Free Software |
| 30 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 31 | */ |
| 32 | |
| 33 | /** |
| 34 | * @file |
| 35 | * DV decoder |
| 36 | */ |
| 37 | |
| 38 | #include "libavutil/avassert.h" |
| 39 | #include "libavutil/imgutils.h" |
| 40 | #include "libavutil/internal.h" |
| 41 | #include "libavutil/pixdesc.h" |
| 42 | |
| 43 | #include "avcodec.h" |
| 44 | #include "dv.h" |
| 45 | #include "dvdata.h" |
| 46 | #include "get_bits.h" |
| 47 | #include "idctdsp.h" |
| 48 | #include "internal.h" |
| 49 | #include "put_bits.h" |
| 50 | #include "simple_idct.h" |
| 51 | |
| 52 | typedef struct BlockInfo { |
| 53 | const uint32_t *factor_table; |
| 54 | const uint8_t *scan_table; |
| 55 | uint8_t pos; /* position in block */ |
| 56 | void (*idct_put)(uint8_t *dest, int line_size, int16_t *block); |
| 57 | uint8_t partial_bit_count; |
| 58 | uint32_t partial_bit_buffer; |
| 59 | int shift_offset; |
| 60 | } BlockInfo; |
| 61 | |
| 62 | static const int dv_iweight_bits = 14; |
| 63 | |
| 64 | static av_cold int dvvideo_decode_init(AVCodecContext *avctx) |
| 65 | { |
| 66 | DVVideoContext *s = avctx->priv_data; |
| 67 | IDCTDSPContext idsp; |
| 68 | int i; |
| 69 | |
| 70 | memset(&idsp,0, sizeof(idsp)); |
| 71 | ff_idctdsp_init(&idsp, avctx); |
| 72 | |
| 73 | for (i = 0; i < 64; i++) |
| 74 | s->dv_zigzag[0][i] = idsp.idct_permutation[ff_zigzag_direct[i]]; |
| 75 | |
| 76 | if (avctx->lowres){ |
| 77 | for (i = 0; i < 64; i++){ |
| 78 | int j = ff_dv_zigzag248_direct[i]; |
| 79 | s->dv_zigzag[1][i] = idsp.idct_permutation[(j & 7) + (j & 8) * 4 + (j & 48) / 2]; |
| 80 | } |
| 81 | }else |
| 82 | memcpy(s->dv_zigzag[1], ff_dv_zigzag248_direct, sizeof(s->dv_zigzag[1])); |
| 83 | |
| 84 | s->idct_put[0] = idsp.idct_put; |
| 85 | s->idct_put[1] = ff_simple_idct248_put; |
| 86 | |
| 87 | return ff_dvvideo_init(avctx); |
| 88 | } |
| 89 | |
| 90 | /* decode AC coefficients */ |
| 91 | static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, int16_t *block) |
| 92 | { |
| 93 | int last_index = gb->size_in_bits; |
| 94 | const uint8_t *scan_table = mb->scan_table; |
| 95 | const uint32_t *factor_table = mb->factor_table; |
| 96 | int pos = mb->pos; |
| 97 | int partial_bit_count = mb->partial_bit_count; |
| 98 | int level, run, vlc_len, index; |
| 99 | |
| 100 | OPEN_READER_NOSIZE(re, gb); |
| 101 | UPDATE_CACHE(re, gb); |
| 102 | |
| 103 | /* if we must parse a partial VLC, we do it here */ |
| 104 | if (partial_bit_count > 0) { |
| 105 | re_cache = re_cache >> partial_bit_count | |
| 106 | mb->partial_bit_buffer; |
| 107 | re_index -= partial_bit_count; |
| 108 | mb->partial_bit_count = 0; |
| 109 | } |
| 110 | |
| 111 | /* get the AC coefficients until last_index is reached */ |
| 112 | for (;;) { |
| 113 | av_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16), |
| 114 | re_index); |
| 115 | /* our own optimized GET_RL_VLC */ |
| 116 | index = NEG_USR32(re_cache, TEX_VLC_BITS); |
| 117 | vlc_len = ff_dv_rl_vlc[index].len; |
| 118 | if (vlc_len < 0) { |
| 119 | index = NEG_USR32((unsigned) re_cache << TEX_VLC_BITS, -vlc_len) + |
| 120 | ff_dv_rl_vlc[index].level; |
| 121 | vlc_len = TEX_VLC_BITS - vlc_len; |
| 122 | } |
| 123 | level = ff_dv_rl_vlc[index].level; |
| 124 | run = ff_dv_rl_vlc[index].run; |
| 125 | |
| 126 | /* gotta check if we're still within gb boundaries */ |
| 127 | if (re_index + vlc_len > last_index) { |
| 128 | /* should be < 16 bits otherwise a codeword could have been parsed */ |
| 129 | mb->partial_bit_count = last_index - re_index; |
| 130 | mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count); |
| 131 | re_index = last_index; |
| 132 | break; |
| 133 | } |
| 134 | re_index += vlc_len; |
| 135 | |
| 136 | av_dlog(NULL, "run=%d level=%d\n", run, level); |
| 137 | pos += run; |
| 138 | if (pos >= 64) |
| 139 | break; |
| 140 | |
| 141 | level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >> |
| 142 | dv_iweight_bits; |
| 143 | block[scan_table[pos]] = level; |
| 144 | |
| 145 | UPDATE_CACHE(re, gb); |
| 146 | } |
| 147 | CLOSE_READER(re, gb); |
| 148 | mb->pos = pos; |
| 149 | } |
| 150 | |
| 151 | static inline void bit_copy(PutBitContext *pb, GetBitContext *gb) |
| 152 | { |
| 153 | int bits_left = get_bits_left(gb); |
| 154 | while (bits_left >= MIN_CACHE_BITS) { |
| 155 | put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS)); |
| 156 | bits_left -= MIN_CACHE_BITS; |
| 157 | } |
| 158 | if (bits_left > 0) |
| 159 | put_bits(pb, bits_left, get_bits(gb, bits_left)); |
| 160 | } |
| 161 | |
| 162 | /* mb_x and mb_y are in units of 8 pixels */ |
| 163 | static int dv_decode_video_segment(AVCodecContext *avctx, void *arg) |
| 164 | { |
| 165 | DVVideoContext *s = avctx->priv_data; |
| 166 | DVwork_chunk *work_chunk = arg; |
| 167 | int quant, dc, dct_mode, class1, j; |
| 168 | int mb_index, mb_x, mb_y, last_index; |
| 169 | int y_stride, linesize; |
| 170 | int16_t *block, *block1; |
| 171 | int c_offset; |
| 172 | uint8_t *y_ptr; |
| 173 | const uint8_t *buf_ptr; |
| 174 | PutBitContext pb, vs_pb; |
| 175 | GetBitContext gb; |
| 176 | BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1; |
| 177 | LOCAL_ALIGNED_16(int16_t, sblock, [5 * DV_MAX_BPM], [64]); |
| 178 | LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */ |
| 179 | LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [80 * 5 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */ |
| 180 | const int log2_blocksize = 3-s->avctx->lowres; |
| 181 | int is_field_mode[5]; |
| 182 | |
| 183 | av_assert1((((int) mb_bit_buffer) & 7) == 0); |
| 184 | av_assert1((((int) vs_bit_buffer) & 7) == 0); |
| 185 | |
| 186 | memset(sblock, 0, 5 * DV_MAX_BPM * sizeof(*sblock)); |
| 187 | |
| 188 | /* pass 1: read DC and AC coefficients in blocks */ |
| 189 | buf_ptr = &s->buf[work_chunk->buf_offset * 80]; |
| 190 | block1 = &sblock[0][0]; |
| 191 | mb1 = mb_data; |
| 192 | init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80); |
| 193 | for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) { |
| 194 | /* skip header */ |
| 195 | quant = buf_ptr[3] & 0x0f; |
| 196 | buf_ptr += 4; |
| 197 | init_put_bits(&pb, mb_bit_buffer, 80); |
| 198 | mb = mb1; |
| 199 | block = block1; |
| 200 | is_field_mode[mb_index] = 0; |
| 201 | for (j = 0; j < s->sys->bpm; j++) { |
| 202 | last_index = s->sys->block_sizes[j]; |
| 203 | init_get_bits(&gb, buf_ptr, last_index); |
| 204 | |
| 205 | /* get the DC */ |
| 206 | dc = get_sbits(&gb, 9); |
| 207 | dct_mode = get_bits1(&gb); |
| 208 | class1 = get_bits(&gb, 2); |
| 209 | if (DV_PROFILE_IS_HD(s->sys)) { |
| 210 | mb->idct_put = s->idct_put[0]; |
| 211 | mb->scan_table = s->dv_zigzag[0]; |
| 212 | mb->factor_table = &s->idct_factor[(j >= 4) * 4 * 16 * 64 + |
| 213 | class1 * 16 * 64 + |
| 214 | quant * 64]; |
| 215 | is_field_mode[mb_index] |= !j && dct_mode; |
| 216 | } else { |
| 217 | mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3]; |
| 218 | mb->scan_table = s->dv_zigzag[dct_mode]; |
| 219 | mb->factor_table = |
| 220 | &s->idct_factor[(class1 == 3) * 2 * 22 * 64 + |
| 221 | dct_mode * 22 * 64 + |
| 222 | (quant + ff_dv_quant_offset[class1]) * 64]; |
| 223 | } |
| 224 | dc = dc << 2; |
| 225 | /* convert to unsigned because 128 is not added in the |
| 226 | * standard IDCT */ |
| 227 | dc += 1024; |
| 228 | block[0] = dc; |
| 229 | buf_ptr += last_index >> 3; |
| 230 | mb->pos = 0; |
| 231 | mb->partial_bit_count = 0; |
| 232 | |
| 233 | av_dlog(avctx, "MB block: %d, %d ", mb_index, j); |
| 234 | dv_decode_ac(&gb, mb, block); |
| 235 | |
| 236 | /* write the remaining bits in a new buffer only if the |
| 237 | * block is finished */ |
| 238 | if (mb->pos >= 64) |
| 239 | bit_copy(&pb, &gb); |
| 240 | |
| 241 | block += 64; |
| 242 | mb++; |
| 243 | } |
| 244 | |
| 245 | /* pass 2: we can do it just after */ |
| 246 | av_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index); |
| 247 | block = block1; |
| 248 | mb = mb1; |
| 249 | init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb)); |
| 250 | put_bits32(&pb, 0); // padding must be zeroed |
| 251 | flush_put_bits(&pb); |
| 252 | for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) { |
| 253 | if (mb->pos < 64 && get_bits_left(&gb) > 0) { |
| 254 | dv_decode_ac(&gb, mb, block); |
| 255 | /* if still not finished, no need to parse other blocks */ |
| 256 | if (mb->pos < 64) |
| 257 | break; |
| 258 | } |
| 259 | } |
| 260 | /* all blocks are finished, so the extra bytes can be used at |
| 261 | * the video segment level */ |
| 262 | if (j >= s->sys->bpm) |
| 263 | bit_copy(&vs_pb, &gb); |
| 264 | } |
| 265 | |
| 266 | /* we need a pass over the whole video segment */ |
| 267 | av_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb)); |
| 268 | block = &sblock[0][0]; |
| 269 | mb = mb_data; |
| 270 | init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb)); |
| 271 | put_bits32(&vs_pb, 0); // padding must be zeroed |
| 272 | flush_put_bits(&vs_pb); |
| 273 | for (mb_index = 0; mb_index < 5; mb_index++) { |
| 274 | for (j = 0; j < s->sys->bpm; j++) { |
| 275 | if (mb->pos < 64 && get_bits_left(&gb) > 0) { |
| 276 | av_dlog(avctx, "start %d:%d\n", mb_index, j); |
| 277 | dv_decode_ac(&gb, mb, block); |
| 278 | } |
| 279 | if (mb->pos >= 64 && mb->pos < 127) |
| 280 | av_log(avctx, AV_LOG_ERROR, |
| 281 | "AC EOB marker is absent pos=%d\n", mb->pos); |
| 282 | block += 64; |
| 283 | mb++; |
| 284 | } |
| 285 | } |
| 286 | |
| 287 | /* compute idct and place blocks */ |
| 288 | block = &sblock[0][0]; |
| 289 | mb = mb_data; |
| 290 | for (mb_index = 0; mb_index < 5; mb_index++) { |
| 291 | dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y); |
| 292 | |
| 293 | /* idct_put'ting luminance */ |
| 294 | if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) || |
| 295 | (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) || |
| 296 | (s->sys->height >= 720 && mb_y != 134)) { |
| 297 | y_stride = (s->frame->linesize[0] << |
| 298 | ((!is_field_mode[mb_index]) * log2_blocksize)); |
| 299 | } else { |
| 300 | y_stride = (2 << log2_blocksize); |
| 301 | } |
| 302 | y_ptr = s->frame->data[0] + |
| 303 | ((mb_y * s->frame->linesize[0] + mb_x) << log2_blocksize); |
| 304 | linesize = s->frame->linesize[0] << is_field_mode[mb_index]; |
| 305 | mb[0].idct_put(y_ptr, linesize, block + 0 * 64); |
| 306 | if (s->sys->video_stype == 4) { /* SD 422 */ |
| 307 | mb[2].idct_put(y_ptr + (1 << log2_blocksize), linesize, block + 2 * 64); |
| 308 | } else { |
| 309 | mb[1].idct_put(y_ptr + (1 << log2_blocksize), linesize, block + 1 * 64); |
| 310 | mb[2].idct_put(y_ptr + y_stride, linesize, block + 2 * 64); |
| 311 | mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3 * 64); |
| 312 | } |
| 313 | mb += 4; |
| 314 | block += 4 * 64; |
| 315 | |
| 316 | /* idct_put'ting chrominance */ |
| 317 | c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] + |
| 318 | (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize); |
| 319 | for (j = 2; j; j--) { |
| 320 | uint8_t *c_ptr = s->frame->data[j] + c_offset; |
| 321 | if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) { |
| 322 | uint64_t aligned_pixels[64 / 8]; |
| 323 | uint8_t *pixels = (uint8_t *) aligned_pixels; |
| 324 | uint8_t *c_ptr1, *ptr1; |
| 325 | int x, y; |
| 326 | mb->idct_put(pixels, 8, block); |
| 327 | for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->frame->linesize[j], pixels += 8) { |
| 328 | ptr1 = pixels + ((1 << (log2_blocksize))>>1); |
| 329 | c_ptr1 = c_ptr + (s->frame->linesize[j] << log2_blocksize); |
| 330 | for (x = 0; x < (1 << FFMAX(log2_blocksize - 1, 0)); x++) { |
| 331 | c_ptr[x] = pixels[x]; |
| 332 | c_ptr1[x] = ptr1[x]; |
| 333 | } |
| 334 | } |
| 335 | block += 64; |
| 336 | mb++; |
| 337 | } else { |
| 338 | y_stride = (mb_y == 134) ? (1 << log2_blocksize) : |
| 339 | s->frame->linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize); |
| 340 | linesize = s->frame->linesize[j] << is_field_mode[mb_index]; |
| 341 | (mb++)->idct_put(c_ptr, linesize, block); |
| 342 | block += 64; |
| 343 | if (s->sys->bpm == 8) { |
| 344 | (mb++)->idct_put(c_ptr + y_stride, linesize, block); |
| 345 | block += 64; |
| 346 | } |
| 347 | } |
| 348 | } |
| 349 | } |
| 350 | return 0; |
| 351 | } |
| 352 | |
| 353 | /* NOTE: exactly one frame must be given (120000 bytes for NTSC, |
| 354 | * 144000 bytes for PAL - or twice those for 50Mbps) */ |
| 355 | static int dvvideo_decode_frame(AVCodecContext *avctx, void *data, |
| 356 | int *got_frame, AVPacket *avpkt) |
| 357 | { |
| 358 | uint8_t *buf = avpkt->data; |
| 359 | int buf_size = avpkt->size; |
| 360 | DVVideoContext *s = avctx->priv_data; |
| 361 | const uint8_t *vsc_pack; |
| 362 | int apt, is16_9, ret; |
| 363 | const AVDVProfile *sys; |
| 364 | |
| 365 | sys = avpriv_dv_frame_profile2(avctx, s->sys, buf, buf_size); |
| 366 | if (!sys || buf_size < sys->frame_size) { |
| 367 | av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n"); |
| 368 | return -1; /* NOTE: we only accept several full frames */ |
| 369 | } |
| 370 | |
| 371 | if (sys != s->sys) { |
| 372 | ret = ff_dv_init_dynamic_tables(s, sys); |
| 373 | if (ret < 0) { |
| 374 | av_log(avctx, AV_LOG_ERROR, "Error initializing the work tables.\n"); |
| 375 | return ret; |
| 376 | } |
| 377 | s->sys = sys; |
| 378 | } |
| 379 | |
| 380 | s->frame = data; |
| 381 | s->frame->key_frame = 1; |
| 382 | s->frame->pict_type = AV_PICTURE_TYPE_I; |
| 383 | avctx->pix_fmt = s->sys->pix_fmt; |
| 384 | avctx->time_base = s->sys->time_base; |
| 385 | |
| 386 | ret = ff_set_dimensions(avctx, s->sys->width, s->sys->height); |
| 387 | if (ret < 0) |
| 388 | return ret; |
| 389 | |
| 390 | /* Determine the codec's sample_aspect ratio from the packet */ |
| 391 | vsc_pack = buf + 80 * 5 + 48 + 5; |
| 392 | if (*vsc_pack == dv_video_control) { |
| 393 | apt = buf[4] & 0x07; |
| 394 | is16_9 = (vsc_pack[2] & 0x07) == 0x02 || |
| 395 | (!apt && (vsc_pack[2] & 0x07) == 0x07); |
| 396 | ff_set_sar(avctx, s->sys->sar[is16_9]); |
| 397 | } |
| 398 | |
| 399 | if ((ret = ff_get_buffer(avctx, s->frame, 0)) < 0) |
| 400 | return ret; |
| 401 | s->frame->interlaced_frame = 1; |
| 402 | s->frame->top_field_first = 0; |
| 403 | |
| 404 | /* Determine the codec's field order from the packet */ |
| 405 | if ( *vsc_pack == dv_video_control ) { |
| 406 | s->frame->top_field_first = !(vsc_pack[3] & 0x40); |
| 407 | } |
| 408 | |
| 409 | s->buf = buf; |
| 410 | avctx->execute(avctx, dv_decode_video_segment, s->work_chunks, NULL, |
| 411 | dv_work_pool_size(s->sys), sizeof(DVwork_chunk)); |
| 412 | |
| 413 | emms_c(); |
| 414 | |
| 415 | /* return image */ |
| 416 | *got_frame = 1; |
| 417 | |
| 418 | return s->sys->frame_size; |
| 419 | } |
| 420 | |
| 421 | AVCodec ff_dvvideo_decoder = { |
| 422 | .name = "dvvideo", |
| 423 | .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"), |
| 424 | .type = AVMEDIA_TYPE_VIDEO, |
| 425 | .id = AV_CODEC_ID_DVVIDEO, |
| 426 | .priv_data_size = sizeof(DVVideoContext), |
| 427 | .init = dvvideo_decode_init, |
| 428 | .decode = dvvideo_decode_frame, |
| 429 | .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS, |
| 430 | .max_lowres = 3, |
| 431 | }; |