| 1 | /* |
| 2 | * Windows Media Audio Lossless decoder |
| 3 | * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion |
| 4 | * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson |
| 5 | * Copyright (c) 2011 Andreas Ă–man |
| 6 | * Copyright (c) 2011 - 2012 Mashiat Sarker Shakkhar |
| 7 | * |
| 8 | * This file is part of FFmpeg. |
| 9 | * |
| 10 | * FFmpeg is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU Lesser General Public |
| 12 | * License as published by the Free Software Foundation; either |
| 13 | * version 2.1 of the License, or (at your option) any later version. |
| 14 | * |
| 15 | * FFmpeg is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * Lesser General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU Lesser General Public |
| 21 | * License along with FFmpeg; if not, write to the Free Software |
| 22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 23 | */ |
| 24 | |
| 25 | #include <inttypes.h> |
| 26 | |
| 27 | #include "libavutil/attributes.h" |
| 28 | #include "libavutil/avassert.h" |
| 29 | |
| 30 | #include "avcodec.h" |
| 31 | #include "internal.h" |
| 32 | #include "get_bits.h" |
| 33 | #include "put_bits.h" |
| 34 | #include "lossless_audiodsp.h" |
| 35 | #include "wma.h" |
| 36 | #include "wma_common.h" |
| 37 | |
| 38 | /** current decoder limitations */ |
| 39 | #define WMALL_MAX_CHANNELS 8 ///< max number of handled channels |
| 40 | #define MAX_SUBFRAMES 32 ///< max number of subframes per channel |
| 41 | #define MAX_BANDS 29 ///< max number of scale factor bands |
| 42 | #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size |
| 43 | #define MAX_ORDER 256 |
| 44 | |
| 45 | #define WMALL_BLOCK_MIN_BITS 6 ///< log2 of min block size |
| 46 | #define WMALL_BLOCK_MAX_BITS 14 ///< log2 of max block size |
| 47 | #define WMALL_BLOCK_MAX_SIZE (1 << WMALL_BLOCK_MAX_BITS) ///< maximum block size |
| 48 | #define WMALL_BLOCK_SIZES (WMALL_BLOCK_MAX_BITS - WMALL_BLOCK_MIN_BITS + 1) ///< possible block sizes |
| 49 | |
| 50 | #define WMALL_COEFF_PAD_SIZE 16 ///< pad coef buffers with 0 for use with SIMD |
| 51 | |
| 52 | /** |
| 53 | * @brief frame-specific decoder context for a single channel |
| 54 | */ |
| 55 | typedef struct { |
| 56 | int16_t prev_block_len; ///< length of the previous block |
| 57 | uint8_t transmit_coefs; |
| 58 | uint8_t num_subframes; |
| 59 | uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples |
| 60 | uint16_t subframe_offsets[MAX_SUBFRAMES]; ///< subframe positions in the current frame |
| 61 | uint8_t cur_subframe; ///< current subframe number |
| 62 | uint16_t decoded_samples; ///< number of already processed samples |
| 63 | int quant_step; ///< quantization step for the current subframe |
| 64 | int transient_counter; ///< number of transient samples from the beginning of the transient zone |
| 65 | } WmallChannelCtx; |
| 66 | |
| 67 | /** |
| 68 | * @brief main decoder context |
| 69 | */ |
| 70 | typedef struct WmallDecodeCtx { |
| 71 | /* generic decoder variables */ |
| 72 | AVCodecContext *avctx; |
| 73 | AVFrame *frame; |
| 74 | LLAudDSPContext dsp; ///< accelerated DSP functions |
| 75 | uint8_t frame_data[MAX_FRAMESIZE + FF_INPUT_BUFFER_PADDING_SIZE]; ///< compressed frame data |
| 76 | PutBitContext pb; ///< context for filling the frame_data buffer |
| 77 | |
| 78 | /* frame size dependent frame information (set during initialization) */ |
| 79 | uint32_t decode_flags; ///< used compression features |
| 80 | int len_prefix; ///< frame is prefixed with its length |
| 81 | int dynamic_range_compression; ///< frame contains DRC data |
| 82 | uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0]) |
| 83 | uint16_t samples_per_frame; ///< number of samples to output |
| 84 | uint16_t log2_frame_size; |
| 85 | int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels) |
| 86 | int8_t lfe_channel; ///< lfe channel index |
| 87 | uint8_t max_num_subframes; |
| 88 | uint8_t subframe_len_bits; ///< number of bits used for the subframe length |
| 89 | uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1 |
| 90 | uint16_t min_samples_per_subframe; |
| 91 | |
| 92 | /* packet decode state */ |
| 93 | GetBitContext pgb; ///< bitstream reader context for the packet |
| 94 | int next_packet_start; ///< start offset of the next WMA packet in the demuxer packet |
| 95 | uint8_t packet_offset; ///< offset to the frame in the packet |
| 96 | uint8_t packet_sequence_number; ///< current packet number |
| 97 | int num_saved_bits; ///< saved number of bits |
| 98 | int frame_offset; ///< frame offset in the bit reservoir |
| 99 | int subframe_offset; ///< subframe offset in the bit reservoir |
| 100 | uint8_t packet_loss; ///< set in case of bitstream error |
| 101 | uint8_t packet_done; ///< set when a packet is fully decoded |
| 102 | |
| 103 | /* frame decode state */ |
| 104 | uint32_t frame_num; ///< current frame number (not used for decoding) |
| 105 | GetBitContext gb; ///< bitstream reader context |
| 106 | int buf_bit_size; ///< buffer size in bits |
| 107 | int16_t *samples_16[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (16-bit) |
| 108 | int32_t *samples_32[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (24-bit) |
| 109 | uint8_t drc_gain; ///< gain for the DRC tool |
| 110 | int8_t skip_frame; ///< skip output step |
| 111 | int8_t parsed_all_subframes; ///< all subframes decoded? |
| 112 | |
| 113 | /* subframe/block decode state */ |
| 114 | int16_t subframe_len; ///< current subframe length |
| 115 | int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe |
| 116 | int8_t channel_indexes_for_cur_subframe[WMALL_MAX_CHANNELS]; |
| 117 | |
| 118 | WmallChannelCtx channel[WMALL_MAX_CHANNELS]; ///< per channel data |
| 119 | |
| 120 | // WMA Lossless-specific |
| 121 | |
| 122 | uint8_t do_arith_coding; |
| 123 | uint8_t do_ac_filter; |
| 124 | uint8_t do_inter_ch_decorr; |
| 125 | uint8_t do_mclms; |
| 126 | uint8_t do_lpc; |
| 127 | |
| 128 | int8_t acfilter_order; |
| 129 | int8_t acfilter_scaling; |
| 130 | int64_t acfilter_coeffs[16]; |
| 131 | int acfilter_prevvalues[WMALL_MAX_CHANNELS][16]; |
| 132 | |
| 133 | int8_t mclms_order; |
| 134 | int8_t mclms_scaling; |
| 135 | int16_t mclms_coeffs[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS * 32]; |
| 136 | int16_t mclms_coeffs_cur[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS]; |
| 137 | int16_t mclms_prevvalues[WMALL_MAX_CHANNELS * 2 * 32]; |
| 138 | int16_t mclms_updates[WMALL_MAX_CHANNELS * 2 * 32]; |
| 139 | int mclms_recent; |
| 140 | |
| 141 | int movave_scaling; |
| 142 | int quant_stepsize; |
| 143 | |
| 144 | struct { |
| 145 | int order; |
| 146 | int scaling; |
| 147 | int coefsend; |
| 148 | int bitsend; |
| 149 | DECLARE_ALIGNED(16, int16_t, coefs)[MAX_ORDER + WMALL_COEFF_PAD_SIZE/sizeof(int16_t)]; |
| 150 | DECLARE_ALIGNED(16, int16_t, lms_prevvalues)[MAX_ORDER * 2]; |
| 151 | DECLARE_ALIGNED(16, int16_t, lms_updates)[MAX_ORDER * 2]; |
| 152 | int recent; |
| 153 | } cdlms[WMALL_MAX_CHANNELS][9]; |
| 154 | |
| 155 | int cdlms_ttl[WMALL_MAX_CHANNELS]; |
| 156 | |
| 157 | int bV3RTM; |
| 158 | |
| 159 | int is_channel_coded[WMALL_MAX_CHANNELS]; |
| 160 | int update_speed[WMALL_MAX_CHANNELS]; |
| 161 | |
| 162 | int transient[WMALL_MAX_CHANNELS]; |
| 163 | int transient_pos[WMALL_MAX_CHANNELS]; |
| 164 | int seekable_tile; |
| 165 | |
| 166 | int ave_sum[WMALL_MAX_CHANNELS]; |
| 167 | |
| 168 | int channel_residues[WMALL_MAX_CHANNELS][WMALL_BLOCK_MAX_SIZE]; |
| 169 | |
| 170 | int lpc_coefs[WMALL_MAX_CHANNELS][40]; |
| 171 | int lpc_order; |
| 172 | int lpc_scaling; |
| 173 | int lpc_intbits; |
| 174 | |
| 175 | int channel_coeffs[WMALL_MAX_CHANNELS][WMALL_BLOCK_MAX_SIZE]; |
| 176 | } WmallDecodeCtx; |
| 177 | |
| 178 | |
| 179 | static av_cold int decode_init(AVCodecContext *avctx) |
| 180 | { |
| 181 | WmallDecodeCtx *s = avctx->priv_data; |
| 182 | uint8_t *edata_ptr = avctx->extradata; |
| 183 | unsigned int channel_mask; |
| 184 | int i, log2_max_num_subframes; |
| 185 | |
| 186 | if (!avctx->block_align) { |
| 187 | av_log(avctx, AV_LOG_ERROR, "block_align is not set\n"); |
| 188 | return AVERROR(EINVAL); |
| 189 | } |
| 190 | |
| 191 | s->avctx = avctx; |
| 192 | ff_llauddsp_init(&s->dsp); |
| 193 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
| 194 | |
| 195 | if (avctx->extradata_size >= 18) { |
| 196 | s->decode_flags = AV_RL16(edata_ptr + 14); |
| 197 | channel_mask = AV_RL32(edata_ptr + 2); |
| 198 | s->bits_per_sample = AV_RL16(edata_ptr); |
| 199 | if (s->bits_per_sample == 16) |
| 200 | avctx->sample_fmt = AV_SAMPLE_FMT_S16P; |
| 201 | else if (s->bits_per_sample == 24) { |
| 202 | avctx->sample_fmt = AV_SAMPLE_FMT_S32P; |
| 203 | avpriv_report_missing_feature(avctx, "Bit-depth higher than 16"); |
| 204 | return AVERROR_PATCHWELCOME; |
| 205 | } else { |
| 206 | av_log(avctx, AV_LOG_ERROR, "Unknown bit-depth: %"PRIu8"\n", |
| 207 | s->bits_per_sample); |
| 208 | return AVERROR_INVALIDDATA; |
| 209 | } |
| 210 | /* dump the extradata */ |
| 211 | for (i = 0; i < avctx->extradata_size; i++) |
| 212 | av_dlog(avctx, "[%x] ", avctx->extradata[i]); |
| 213 | av_dlog(avctx, "\n"); |
| 214 | |
| 215 | } else { |
| 216 | avpriv_request_sample(avctx, "Unsupported extradata size"); |
| 217 | return AVERROR_PATCHWELCOME; |
| 218 | } |
| 219 | |
| 220 | /* generic init */ |
| 221 | s->log2_frame_size = av_log2(avctx->block_align) + 4; |
| 222 | |
| 223 | /* frame info */ |
| 224 | s->skip_frame = 1; /* skip first frame */ |
| 225 | s->packet_loss = 1; |
| 226 | s->len_prefix = s->decode_flags & 0x40; |
| 227 | |
| 228 | /* get frame len */ |
| 229 | s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate, |
| 230 | 3, s->decode_flags); |
| 231 | av_assert0(s->samples_per_frame <= WMALL_BLOCK_MAX_SIZE); |
| 232 | |
| 233 | /* init previous block len */ |
| 234 | for (i = 0; i < avctx->channels; i++) |
| 235 | s->channel[i].prev_block_len = s->samples_per_frame; |
| 236 | |
| 237 | /* subframe info */ |
| 238 | log2_max_num_subframes = (s->decode_flags & 0x38) >> 3; |
| 239 | s->max_num_subframes = 1 << log2_max_num_subframes; |
| 240 | s->max_subframe_len_bit = 0; |
| 241 | s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1; |
| 242 | |
| 243 | s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes; |
| 244 | s->dynamic_range_compression = s->decode_flags & 0x80; |
| 245 | s->bV3RTM = s->decode_flags & 0x100; |
| 246 | |
| 247 | if (s->max_num_subframes > MAX_SUBFRAMES) { |
| 248 | av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRIu8"\n", |
| 249 | s->max_num_subframes); |
| 250 | return AVERROR_INVALIDDATA; |
| 251 | } |
| 252 | |
| 253 | s->num_channels = avctx->channels; |
| 254 | |
| 255 | /* extract lfe channel position */ |
| 256 | s->lfe_channel = -1; |
| 257 | |
| 258 | if (channel_mask & 8) { |
| 259 | unsigned int mask; |
| 260 | for (mask = 1; mask < 16; mask <<= 1) |
| 261 | if (channel_mask & mask) |
| 262 | ++s->lfe_channel; |
| 263 | } |
| 264 | |
| 265 | if (s->num_channels < 0) { |
| 266 | av_log(avctx, AV_LOG_ERROR, "invalid number of channels %"PRId8"\n", |
| 267 | s->num_channels); |
| 268 | return AVERROR_INVALIDDATA; |
| 269 | } else if (s->num_channels > WMALL_MAX_CHANNELS) { |
| 270 | avpriv_request_sample(avctx, |
| 271 | "More than %d channels", WMALL_MAX_CHANNELS); |
| 272 | return AVERROR_PATCHWELCOME; |
| 273 | } |
| 274 | |
| 275 | s->frame = av_frame_alloc(); |
| 276 | if (!s->frame) |
| 277 | return AVERROR(ENOMEM); |
| 278 | |
| 279 | avctx->channel_layout = channel_mask; |
| 280 | return 0; |
| 281 | } |
| 282 | |
| 283 | /** |
| 284 | * @brief Decode the subframe length. |
| 285 | * @param s context |
| 286 | * @param offset sample offset in the frame |
| 287 | * @return decoded subframe length on success, < 0 in case of an error |
| 288 | */ |
| 289 | static int decode_subframe_length(WmallDecodeCtx *s, int offset) |
| 290 | { |
| 291 | int frame_len_ratio, subframe_len, len; |
| 292 | |
| 293 | /* no need to read from the bitstream when only one length is possible */ |
| 294 | if (offset == s->samples_per_frame - s->min_samples_per_subframe) |
| 295 | return s->min_samples_per_subframe; |
| 296 | |
| 297 | len = av_log2(s->max_num_subframes - 1) + 1; |
| 298 | frame_len_ratio = get_bits(&s->gb, len); |
| 299 | subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1); |
| 300 | |
| 301 | /* sanity check the length */ |
| 302 | if (subframe_len < s->min_samples_per_subframe || |
| 303 | subframe_len > s->samples_per_frame) { |
| 304 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n", |
| 305 | subframe_len); |
| 306 | return AVERROR_INVALIDDATA; |
| 307 | } |
| 308 | return subframe_len; |
| 309 | } |
| 310 | |
| 311 | /** |
| 312 | * @brief Decode how the data in the frame is split into subframes. |
| 313 | * Every WMA frame contains the encoded data for a fixed number of |
| 314 | * samples per channel. The data for every channel might be split |
| 315 | * into several subframes. This function will reconstruct the list of |
| 316 | * subframes for every channel. |
| 317 | * |
| 318 | * If the subframes are not evenly split, the algorithm estimates the |
| 319 | * channels with the lowest number of total samples. |
| 320 | * Afterwards, for each of these channels a bit is read from the |
| 321 | * bitstream that indicates if the channel contains a subframe with the |
| 322 | * next subframe size that is going to be read from the bitstream or not. |
| 323 | * If a channel contains such a subframe, the subframe size gets added to |
| 324 | * the channel's subframe list. |
| 325 | * The algorithm repeats these steps until the frame is properly divided |
| 326 | * between the individual channels. |
| 327 | * |
| 328 | * @param s context |
| 329 | * @return 0 on success, < 0 in case of an error |
| 330 | */ |
| 331 | static int decode_tilehdr(WmallDecodeCtx *s) |
| 332 | { |
| 333 | uint16_t num_samples[WMALL_MAX_CHANNELS] = { 0 }; /* sum of samples for all currently known subframes of a channel */ |
| 334 | uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /* flag indicating if a channel contains the current subframe */ |
| 335 | int channels_for_cur_subframe = s->num_channels; /* number of channels that contain the current subframe */ |
| 336 | int fixed_channel_layout = 0; /* flag indicating that all channels use the same subfra2me offsets and sizes */ |
| 337 | int min_channel_len = 0; /* smallest sum of samples (channels with this length will be processed first) */ |
| 338 | int c, tile_aligned; |
| 339 | |
| 340 | /* reset tiling information */ |
| 341 | for (c = 0; c < s->num_channels; c++) |
| 342 | s->channel[c].num_subframes = 0; |
| 343 | |
| 344 | tile_aligned = get_bits1(&s->gb); |
| 345 | if (s->max_num_subframes == 1 || tile_aligned) |
| 346 | fixed_channel_layout = 1; |
| 347 | |
| 348 | /* loop until the frame data is split between the subframes */ |
| 349 | do { |
| 350 | int subframe_len, in_use = 0; |
| 351 | |
| 352 | /* check which channels contain the subframe */ |
| 353 | for (c = 0; c < s->num_channels; c++) { |
| 354 | if (num_samples[c] == min_channel_len) { |
| 355 | if (fixed_channel_layout || channels_for_cur_subframe == 1 || |
| 356 | (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) { |
| 357 | contains_subframe[c] = 1; |
| 358 | } else { |
| 359 | contains_subframe[c] = get_bits1(&s->gb); |
| 360 | } |
| 361 | in_use |= contains_subframe[c]; |
| 362 | } else |
| 363 | contains_subframe[c] = 0; |
| 364 | } |
| 365 | |
| 366 | if (!in_use) { |
| 367 | av_log(s->avctx, AV_LOG_ERROR, |
| 368 | "Found empty subframe\n"); |
| 369 | return AVERROR_INVALIDDATA; |
| 370 | } |
| 371 | |
| 372 | /* get subframe length, subframe_len == 0 is not allowed */ |
| 373 | if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0) |
| 374 | return AVERROR_INVALIDDATA; |
| 375 | /* add subframes to the individual channels and find new min_channel_len */ |
| 376 | min_channel_len += subframe_len; |
| 377 | for (c = 0; c < s->num_channels; c++) { |
| 378 | WmallChannelCtx *chan = &s->channel[c]; |
| 379 | |
| 380 | if (contains_subframe[c]) { |
| 381 | if (chan->num_subframes >= MAX_SUBFRAMES) { |
| 382 | av_log(s->avctx, AV_LOG_ERROR, |
| 383 | "broken frame: num subframes > 31\n"); |
| 384 | return AVERROR_INVALIDDATA; |
| 385 | } |
| 386 | chan->subframe_len[chan->num_subframes] = subframe_len; |
| 387 | num_samples[c] += subframe_len; |
| 388 | ++chan->num_subframes; |
| 389 | if (num_samples[c] > s->samples_per_frame) { |
| 390 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: " |
| 391 | "channel len(%"PRIu16") > samples_per_frame(%"PRIu16")\n", |
| 392 | num_samples[c], s->samples_per_frame); |
| 393 | return AVERROR_INVALIDDATA; |
| 394 | } |
| 395 | } else if (num_samples[c] <= min_channel_len) { |
| 396 | if (num_samples[c] < min_channel_len) { |
| 397 | channels_for_cur_subframe = 0; |
| 398 | min_channel_len = num_samples[c]; |
| 399 | } |
| 400 | ++channels_for_cur_subframe; |
| 401 | } |
| 402 | } |
| 403 | } while (min_channel_len < s->samples_per_frame); |
| 404 | |
| 405 | for (c = 0; c < s->num_channels; c++) { |
| 406 | int i, offset = 0; |
| 407 | for (i = 0; i < s->channel[c].num_subframes; i++) { |
| 408 | s->channel[c].subframe_offsets[i] = offset; |
| 409 | offset += s->channel[c].subframe_len[i]; |
| 410 | } |
| 411 | } |
| 412 | |
| 413 | return 0; |
| 414 | } |
| 415 | |
| 416 | static void decode_ac_filter(WmallDecodeCtx *s) |
| 417 | { |
| 418 | int i; |
| 419 | s->acfilter_order = get_bits(&s->gb, 4) + 1; |
| 420 | s->acfilter_scaling = get_bits(&s->gb, 4); |
| 421 | |
| 422 | for (i = 0; i < s->acfilter_order; i++) |
| 423 | s->acfilter_coeffs[i] = (s->acfilter_scaling ? |
| 424 | get_bits(&s->gb, s->acfilter_scaling) : 0) + 1; |
| 425 | } |
| 426 | |
| 427 | static void decode_mclms(WmallDecodeCtx *s) |
| 428 | { |
| 429 | s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2; |
| 430 | s->mclms_scaling = get_bits(&s->gb, 4); |
| 431 | if (get_bits1(&s->gb)) { |
| 432 | int i, send_coef_bits; |
| 433 | int cbits = av_log2(s->mclms_scaling + 1); |
| 434 | if (1 << cbits < s->mclms_scaling + 1) |
| 435 | cbits++; |
| 436 | |
| 437 | send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2; |
| 438 | |
| 439 | for (i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++) |
| 440 | s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits); |
| 441 | |
| 442 | for (i = 0; i < s->num_channels; i++) { |
| 443 | int c; |
| 444 | for (c = 0; c < i; c++) |
| 445 | s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits); |
| 446 | } |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | static int decode_cdlms(WmallDecodeCtx *s) |
| 451 | { |
| 452 | int c, i; |
| 453 | int cdlms_send_coef = get_bits1(&s->gb); |
| 454 | |
| 455 | for (c = 0; c < s->num_channels; c++) { |
| 456 | s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1; |
| 457 | for (i = 0; i < s->cdlms_ttl[c]; i++) { |
| 458 | s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8; |
| 459 | if (s->cdlms[c][i].order > MAX_ORDER) { |
| 460 | av_log(s->avctx, AV_LOG_ERROR, |
| 461 | "Order[%d][%d] %d > max (%d), not supported\n", |
| 462 | c, i, s->cdlms[c][i].order, MAX_ORDER); |
| 463 | s->cdlms[0][0].order = 0; |
| 464 | return AVERROR_INVALIDDATA; |
| 465 | } |
| 466 | if(s->cdlms[c][i].order & 8) { |
| 467 | static int warned; |
| 468 | if(!warned) |
| 469 | avpriv_request_sample(s->avctx, "CDLMS of order %d", |
| 470 | s->cdlms[c][i].order); |
| 471 | warned = 1; |
| 472 | } |
| 473 | } |
| 474 | |
| 475 | for (i = 0; i < s->cdlms_ttl[c]; i++) |
| 476 | s->cdlms[c][i].scaling = get_bits(&s->gb, 4); |
| 477 | |
| 478 | if (cdlms_send_coef) { |
| 479 | for (i = 0; i < s->cdlms_ttl[c]; i++) { |
| 480 | int cbits, shift_l, shift_r, j; |
| 481 | cbits = av_log2(s->cdlms[c][i].order); |
| 482 | if ((1 << cbits) < s->cdlms[c][i].order) |
| 483 | cbits++; |
| 484 | s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1; |
| 485 | |
| 486 | cbits = av_log2(s->cdlms[c][i].scaling + 1); |
| 487 | if ((1 << cbits) < s->cdlms[c][i].scaling + 1) |
| 488 | cbits++; |
| 489 | |
| 490 | s->cdlms[c][i].bitsend = get_bits(&s->gb, cbits) + 2; |
| 491 | shift_l = 32 - s->cdlms[c][i].bitsend; |
| 492 | shift_r = 32 - s->cdlms[c][i].scaling - 2; |
| 493 | for (j = 0; j < s->cdlms[c][i].coefsend; j++) |
| 494 | s->cdlms[c][i].coefs[j] = |
| 495 | (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r; |
| 496 | } |
| 497 | } |
| 498 | |
| 499 | for (i = 0; i < s->cdlms_ttl[c]; i++) |
| 500 | memset(s->cdlms[c][i].coefs + s->cdlms[c][i].order, |
| 501 | 0, WMALL_COEFF_PAD_SIZE); |
| 502 | } |
| 503 | |
| 504 | return 0; |
| 505 | } |
| 506 | |
| 507 | static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size) |
| 508 | { |
| 509 | int i = 0; |
| 510 | unsigned int ave_mean; |
| 511 | s->transient[ch] = get_bits1(&s->gb); |
| 512 | if (s->transient[ch]) { |
| 513 | s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size)); |
| 514 | if (s->transient_pos[ch]) |
| 515 | s->transient[ch] = 0; |
| 516 | s->channel[ch].transient_counter = |
| 517 | FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2); |
| 518 | } else if (s->channel[ch].transient_counter) |
| 519 | s->transient[ch] = 1; |
| 520 | |
| 521 | if (s->seekable_tile) { |
| 522 | ave_mean = get_bits(&s->gb, s->bits_per_sample); |
| 523 | s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1); |
| 524 | } |
| 525 | |
| 526 | if (s->seekable_tile) { |
| 527 | if (s->do_inter_ch_decorr) |
| 528 | s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample + 1); |
| 529 | else |
| 530 | s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample); |
| 531 | i++; |
| 532 | } |
| 533 | for (; i < tile_size; i++) { |
| 534 | int quo = 0, rem, rem_bits, residue; |
| 535 | while(get_bits1(&s->gb)) { |
| 536 | quo++; |
| 537 | if (get_bits_left(&s->gb) <= 0) |
| 538 | return -1; |
| 539 | } |
| 540 | if (quo >= 32) |
| 541 | quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1); |
| 542 | |
| 543 | ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1); |
| 544 | if (ave_mean <= 1) |
| 545 | residue = quo; |
| 546 | else { |
| 547 | rem_bits = av_ceil_log2(ave_mean); |
| 548 | rem = get_bits_long(&s->gb, rem_bits); |
| 549 | residue = (quo << rem_bits) + rem; |
| 550 | } |
| 551 | |
| 552 | s->ave_sum[ch] = residue + s->ave_sum[ch] - |
| 553 | (s->ave_sum[ch] >> s->movave_scaling); |
| 554 | |
| 555 | if (residue & 1) |
| 556 | residue = -(residue >> 1) - 1; |
| 557 | else |
| 558 | residue = residue >> 1; |
| 559 | s->channel_residues[ch][i] = residue; |
| 560 | } |
| 561 | |
| 562 | return 0; |
| 563 | |
| 564 | } |
| 565 | |
| 566 | static void decode_lpc(WmallDecodeCtx *s) |
| 567 | { |
| 568 | int ch, i, cbits; |
| 569 | s->lpc_order = get_bits(&s->gb, 5) + 1; |
| 570 | s->lpc_scaling = get_bits(&s->gb, 4); |
| 571 | s->lpc_intbits = get_bits(&s->gb, 3) + 1; |
| 572 | cbits = s->lpc_scaling + s->lpc_intbits; |
| 573 | for (ch = 0; ch < s->num_channels; ch++) |
| 574 | for (i = 0; i < s->lpc_order; i++) |
| 575 | s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits); |
| 576 | } |
| 577 | |
| 578 | static void clear_codec_buffers(WmallDecodeCtx *s) |
| 579 | { |
| 580 | int ich, ilms; |
| 581 | |
| 582 | memset(s->acfilter_coeffs, 0, sizeof(s->acfilter_coeffs)); |
| 583 | memset(s->acfilter_prevvalues, 0, sizeof(s->acfilter_prevvalues)); |
| 584 | memset(s->lpc_coefs, 0, sizeof(s->lpc_coefs)); |
| 585 | |
| 586 | memset(s->mclms_coeffs, 0, sizeof(s->mclms_coeffs)); |
| 587 | memset(s->mclms_coeffs_cur, 0, sizeof(s->mclms_coeffs_cur)); |
| 588 | memset(s->mclms_prevvalues, 0, sizeof(s->mclms_prevvalues)); |
| 589 | memset(s->mclms_updates, 0, sizeof(s->mclms_updates)); |
| 590 | |
| 591 | for (ich = 0; ich < s->num_channels; ich++) { |
| 592 | for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) { |
| 593 | memset(s->cdlms[ich][ilms].coefs, 0, |
| 594 | sizeof(s->cdlms[ich][ilms].coefs)); |
| 595 | memset(s->cdlms[ich][ilms].lms_prevvalues, 0, |
| 596 | sizeof(s->cdlms[ich][ilms].lms_prevvalues)); |
| 597 | memset(s->cdlms[ich][ilms].lms_updates, 0, |
| 598 | sizeof(s->cdlms[ich][ilms].lms_updates)); |
| 599 | } |
| 600 | s->ave_sum[ich] = 0; |
| 601 | } |
| 602 | } |
| 603 | |
| 604 | /** |
| 605 | * @brief Reset filter parameters and transient area at new seekable tile. |
| 606 | */ |
| 607 | static void reset_codec(WmallDecodeCtx *s) |
| 608 | { |
| 609 | int ich, ilms; |
| 610 | s->mclms_recent = s->mclms_order * s->num_channels; |
| 611 | for (ich = 0; ich < s->num_channels; ich++) { |
| 612 | for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) |
| 613 | s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order; |
| 614 | /* first sample of a seekable subframe is considered as the starting of |
| 615 | a transient area which is samples_per_frame samples long */ |
| 616 | s->channel[ich].transient_counter = s->samples_per_frame; |
| 617 | s->transient[ich] = 1; |
| 618 | s->transient_pos[ich] = 0; |
| 619 | } |
| 620 | } |
| 621 | |
| 622 | static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred) |
| 623 | { |
| 624 | int i, j, ich, pred_error; |
| 625 | int order = s->mclms_order; |
| 626 | int num_channels = s->num_channels; |
| 627 | int range = 1 << (s->bits_per_sample - 1); |
| 628 | |
| 629 | for (ich = 0; ich < num_channels; ich++) { |
| 630 | pred_error = s->channel_residues[ich][icoef] - pred[ich]; |
| 631 | if (pred_error > 0) { |
| 632 | for (i = 0; i < order * num_channels; i++) |
| 633 | s->mclms_coeffs[i + ich * order * num_channels] += |
| 634 | s->mclms_updates[s->mclms_recent + i]; |
| 635 | for (j = 0; j < ich; j++) { |
| 636 | if (s->channel_residues[j][icoef] > 0) |
| 637 | s->mclms_coeffs_cur[ich * num_channels + j] += 1; |
| 638 | else if (s->channel_residues[j][icoef] < 0) |
| 639 | s->mclms_coeffs_cur[ich * num_channels + j] -= 1; |
| 640 | } |
| 641 | } else if (pred_error < 0) { |
| 642 | for (i = 0; i < order * num_channels; i++) |
| 643 | s->mclms_coeffs[i + ich * order * num_channels] -= |
| 644 | s->mclms_updates[s->mclms_recent + i]; |
| 645 | for (j = 0; j < ich; j++) { |
| 646 | if (s->channel_residues[j][icoef] > 0) |
| 647 | s->mclms_coeffs_cur[ich * num_channels + j] -= 1; |
| 648 | else if (s->channel_residues[j][icoef] < 0) |
| 649 | s->mclms_coeffs_cur[ich * num_channels + j] += 1; |
| 650 | } |
| 651 | } |
| 652 | } |
| 653 | |
| 654 | for (ich = num_channels - 1; ich >= 0; ich--) { |
| 655 | s->mclms_recent--; |
| 656 | s->mclms_prevvalues[s->mclms_recent] = s->channel_residues[ich][icoef]; |
| 657 | if (s->channel_residues[ich][icoef] > range - 1) |
| 658 | s->mclms_prevvalues[s->mclms_recent] = range - 1; |
| 659 | else if (s->channel_residues[ich][icoef] < -range) |
| 660 | s->mclms_prevvalues[s->mclms_recent] = -range; |
| 661 | |
| 662 | s->mclms_updates[s->mclms_recent] = 0; |
| 663 | if (s->channel_residues[ich][icoef] > 0) |
| 664 | s->mclms_updates[s->mclms_recent] = 1; |
| 665 | else if (s->channel_residues[ich][icoef] < 0) |
| 666 | s->mclms_updates[s->mclms_recent] = -1; |
| 667 | } |
| 668 | |
| 669 | if (s->mclms_recent == 0) { |
| 670 | memcpy(&s->mclms_prevvalues[order * num_channels], |
| 671 | s->mclms_prevvalues, |
| 672 | sizeof(int16_t) * order * num_channels); |
| 673 | memcpy(&s->mclms_updates[order * num_channels], |
| 674 | s->mclms_updates, |
| 675 | sizeof(int16_t) * order * num_channels); |
| 676 | s->mclms_recent = num_channels * order; |
| 677 | } |
| 678 | } |
| 679 | |
| 680 | static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred) |
| 681 | { |
| 682 | int ich, i; |
| 683 | int order = s->mclms_order; |
| 684 | int num_channels = s->num_channels; |
| 685 | |
| 686 | for (ich = 0; ich < num_channels; ich++) { |
| 687 | pred[ich] = 0; |
| 688 | if (!s->is_channel_coded[ich]) |
| 689 | continue; |
| 690 | for (i = 0; i < order * num_channels; i++) |
| 691 | pred[ich] += s->mclms_prevvalues[i + s->mclms_recent] * |
| 692 | s->mclms_coeffs[i + order * num_channels * ich]; |
| 693 | for (i = 0; i < ich; i++) |
| 694 | pred[ich] += s->channel_residues[i][icoef] * |
| 695 | s->mclms_coeffs_cur[i + num_channels * ich]; |
| 696 | pred[ich] += 1 << s->mclms_scaling - 1; |
| 697 | pred[ich] >>= s->mclms_scaling; |
| 698 | s->channel_residues[ich][icoef] += pred[ich]; |
| 699 | } |
| 700 | } |
| 701 | |
| 702 | static void revert_mclms(WmallDecodeCtx *s, int tile_size) |
| 703 | { |
| 704 | int icoef, pred[WMALL_MAX_CHANNELS] = { 0 }; |
| 705 | for (icoef = 0; icoef < tile_size; icoef++) { |
| 706 | mclms_predict(s, icoef, pred); |
| 707 | mclms_update(s, icoef, pred); |
| 708 | } |
| 709 | } |
| 710 | |
| 711 | static void lms_update(WmallDecodeCtx *s, int ich, int ilms, int input) |
| 712 | { |
| 713 | int recent = s->cdlms[ich][ilms].recent; |
| 714 | int range = 1 << s->bits_per_sample - 1; |
| 715 | |
| 716 | if (recent) |
| 717 | recent--; |
| 718 | else { |
| 719 | memcpy(&s->cdlms[ich][ilms].lms_prevvalues[s->cdlms[ich][ilms].order], |
| 720 | s->cdlms[ich][ilms].lms_prevvalues, |
| 721 | 2 * s->cdlms[ich][ilms].order); |
| 722 | memcpy(&s->cdlms[ich][ilms].lms_updates[s->cdlms[ich][ilms].order], |
| 723 | s->cdlms[ich][ilms].lms_updates, |
| 724 | 2 * s->cdlms[ich][ilms].order); |
| 725 | recent = s->cdlms[ich][ilms].order - 1; |
| 726 | } |
| 727 | |
| 728 | s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1); |
| 729 | if (!input) |
| 730 | s->cdlms[ich][ilms].lms_updates[recent] = 0; |
| 731 | else if (input < 0) |
| 732 | s->cdlms[ich][ilms].lms_updates[recent] = -s->update_speed[ich]; |
| 733 | else |
| 734 | s->cdlms[ich][ilms].lms_updates[recent] = s->update_speed[ich]; |
| 735 | |
| 736 | s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 4)] >>= 2; |
| 737 | s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 3)] >>= 1; |
| 738 | s->cdlms[ich][ilms].recent = recent; |
| 739 | } |
| 740 | |
| 741 | static void use_high_update_speed(WmallDecodeCtx *s, int ich) |
| 742 | { |
| 743 | int ilms, recent, icoef; |
| 744 | for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) { |
| 745 | recent = s->cdlms[ich][ilms].recent; |
| 746 | if (s->update_speed[ich] == 16) |
| 747 | continue; |
| 748 | if (s->bV3RTM) { |
| 749 | for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++) |
| 750 | s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2; |
| 751 | } else { |
| 752 | for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++) |
| 753 | s->cdlms[ich][ilms].lms_updates[icoef] *= 2; |
| 754 | } |
| 755 | } |
| 756 | s->update_speed[ich] = 16; |
| 757 | } |
| 758 | |
| 759 | static void use_normal_update_speed(WmallDecodeCtx *s, int ich) |
| 760 | { |
| 761 | int ilms, recent, icoef; |
| 762 | for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) { |
| 763 | recent = s->cdlms[ich][ilms].recent; |
| 764 | if (s->update_speed[ich] == 8) |
| 765 | continue; |
| 766 | if (s->bV3RTM) |
| 767 | for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++) |
| 768 | s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2; |
| 769 | else |
| 770 | for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++) |
| 771 | s->cdlms[ich][ilms].lms_updates[icoef] /= 2; |
| 772 | } |
| 773 | s->update_speed[ich] = 8; |
| 774 | } |
| 775 | |
| 776 | /** Get sign of integer (1 for positive, -1 for negative and 0 for zero) */ |
| 777 | #define WMASIGN(x) ((x > 0) - (x < 0)) |
| 778 | |
| 779 | static void revert_cdlms(WmallDecodeCtx *s, int ch, |
| 780 | int coef_begin, int coef_end) |
| 781 | { |
| 782 | int icoef, pred, ilms, num_lms, residue, input; |
| 783 | |
| 784 | num_lms = s->cdlms_ttl[ch]; |
| 785 | for (ilms = num_lms - 1; ilms >= 0; ilms--) { |
| 786 | for (icoef = coef_begin; icoef < coef_end; icoef++) { |
| 787 | pred = 1 << (s->cdlms[ch][ilms].scaling - 1); |
| 788 | residue = s->channel_residues[ch][icoef]; |
| 789 | pred += s->dsp.scalarproduct_and_madd_int16(s->cdlms[ch][ilms].coefs, |
| 790 | s->cdlms[ch][ilms].lms_prevvalues |
| 791 | + s->cdlms[ch][ilms].recent, |
| 792 | s->cdlms[ch][ilms].lms_updates |
| 793 | + s->cdlms[ch][ilms].recent, |
| 794 | s->cdlms[ch][ilms].order, |
| 795 | WMASIGN(residue)); |
| 796 | input = residue + (pred >> s->cdlms[ch][ilms].scaling); |
| 797 | lms_update(s, ch, ilms, input); |
| 798 | s->channel_residues[ch][icoef] = input; |
| 799 | } |
| 800 | } |
| 801 | emms_c(); |
| 802 | } |
| 803 | |
| 804 | static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size) |
| 805 | { |
| 806 | if (s->num_channels != 2) |
| 807 | return; |
| 808 | else if (s->is_channel_coded[0] || s->is_channel_coded[1]) { |
| 809 | int icoef; |
| 810 | for (icoef = 0; icoef < tile_size; icoef++) { |
| 811 | s->channel_residues[0][icoef] -= s->channel_residues[1][icoef] >> 1; |
| 812 | s->channel_residues[1][icoef] += s->channel_residues[0][icoef]; |
| 813 | } |
| 814 | } |
| 815 | } |
| 816 | |
| 817 | static void revert_acfilter(WmallDecodeCtx *s, int tile_size) |
| 818 | { |
| 819 | int ich, pred, i, j; |
| 820 | int64_t *filter_coeffs = s->acfilter_coeffs; |
| 821 | int scaling = s->acfilter_scaling; |
| 822 | int order = s->acfilter_order; |
| 823 | |
| 824 | for (ich = 0; ich < s->num_channels; ich++) { |
| 825 | int *prevvalues = s->acfilter_prevvalues[ich]; |
| 826 | for (i = 0; i < order; i++) { |
| 827 | pred = 0; |
| 828 | for (j = 0; j < order; j++) { |
| 829 | if (i <= j) |
| 830 | pred += filter_coeffs[j] * prevvalues[j - i]; |
| 831 | else |
| 832 | pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j]; |
| 833 | } |
| 834 | pred >>= scaling; |
| 835 | s->channel_residues[ich][i] += pred; |
| 836 | } |
| 837 | for (i = order; i < tile_size; i++) { |
| 838 | pred = 0; |
| 839 | for (j = 0; j < order; j++) |
| 840 | pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j]; |
| 841 | pred >>= scaling; |
| 842 | s->channel_residues[ich][i] += pred; |
| 843 | } |
| 844 | for (j = 0; j < order; j++) |
| 845 | prevvalues[j] = s->channel_residues[ich][tile_size - j - 1]; |
| 846 | } |
| 847 | } |
| 848 | |
| 849 | static int decode_subframe(WmallDecodeCtx *s) |
| 850 | { |
| 851 | int offset = s->samples_per_frame; |
| 852 | int subframe_len = s->samples_per_frame; |
| 853 | int total_samples = s->samples_per_frame * s->num_channels; |
| 854 | int i, j, rawpcm_tile, padding_zeroes, res; |
| 855 | |
| 856 | s->subframe_offset = get_bits_count(&s->gb); |
| 857 | |
| 858 | /* reset channel context and find the next block offset and size |
| 859 | == the next block of the channel with the smallest number of |
| 860 | decoded samples */ |
| 861 | for (i = 0; i < s->num_channels; i++) { |
| 862 | if (offset > s->channel[i].decoded_samples) { |
| 863 | offset = s->channel[i].decoded_samples; |
| 864 | subframe_len = |
| 865 | s->channel[i].subframe_len[s->channel[i].cur_subframe]; |
| 866 | } |
| 867 | } |
| 868 | |
| 869 | /* get a list of all channels that contain the estimated block */ |
| 870 | s->channels_for_cur_subframe = 0; |
| 871 | for (i = 0; i < s->num_channels; i++) { |
| 872 | const int cur_subframe = s->channel[i].cur_subframe; |
| 873 | /* subtract already processed samples */ |
| 874 | total_samples -= s->channel[i].decoded_samples; |
| 875 | |
| 876 | /* and count if there are multiple subframes that match our profile */ |
| 877 | if (offset == s->channel[i].decoded_samples && |
| 878 | subframe_len == s->channel[i].subframe_len[cur_subframe]) { |
| 879 | total_samples -= s->channel[i].subframe_len[cur_subframe]; |
| 880 | s->channel[i].decoded_samples += |
| 881 | s->channel[i].subframe_len[cur_subframe]; |
| 882 | s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i; |
| 883 | ++s->channels_for_cur_subframe; |
| 884 | } |
| 885 | } |
| 886 | |
| 887 | /* check if the frame will be complete after processing the |
| 888 | estimated block */ |
| 889 | if (!total_samples) |
| 890 | s->parsed_all_subframes = 1; |
| 891 | |
| 892 | |
| 893 | s->seekable_tile = get_bits1(&s->gb); |
| 894 | if (s->seekable_tile) { |
| 895 | clear_codec_buffers(s); |
| 896 | |
| 897 | s->do_arith_coding = get_bits1(&s->gb); |
| 898 | if (s->do_arith_coding) { |
| 899 | avpriv_request_sample(s->avctx, "Arithmetic coding"); |
| 900 | return AVERROR_PATCHWELCOME; |
| 901 | } |
| 902 | s->do_ac_filter = get_bits1(&s->gb); |
| 903 | s->do_inter_ch_decorr = get_bits1(&s->gb); |
| 904 | s->do_mclms = get_bits1(&s->gb); |
| 905 | |
| 906 | if (s->do_ac_filter) |
| 907 | decode_ac_filter(s); |
| 908 | |
| 909 | if (s->do_mclms) |
| 910 | decode_mclms(s); |
| 911 | |
| 912 | if ((res = decode_cdlms(s)) < 0) |
| 913 | return res; |
| 914 | s->movave_scaling = get_bits(&s->gb, 3); |
| 915 | s->quant_stepsize = get_bits(&s->gb, 8) + 1; |
| 916 | |
| 917 | reset_codec(s); |
| 918 | } else if (!s->cdlms[0][0].order) { |
| 919 | av_log(s->avctx, AV_LOG_DEBUG, |
| 920 | "Waiting for seekable tile\n"); |
| 921 | av_frame_unref(s->frame); |
| 922 | return -1; |
| 923 | } |
| 924 | |
| 925 | rawpcm_tile = get_bits1(&s->gb); |
| 926 | |
| 927 | for (i = 0; i < s->num_channels; i++) |
| 928 | s->is_channel_coded[i] = 1; |
| 929 | |
| 930 | if (!rawpcm_tile) { |
| 931 | for (i = 0; i < s->num_channels; i++) |
| 932 | s->is_channel_coded[i] = get_bits1(&s->gb); |
| 933 | |
| 934 | if (s->bV3RTM) { |
| 935 | // LPC |
| 936 | s->do_lpc = get_bits1(&s->gb); |
| 937 | if (s->do_lpc) { |
| 938 | decode_lpc(s); |
| 939 | avpriv_request_sample(s->avctx, "Expect wrong output since " |
| 940 | "inverse LPC filter"); |
| 941 | } |
| 942 | } else |
| 943 | s->do_lpc = 0; |
| 944 | } |
| 945 | |
| 946 | |
| 947 | if (get_bits1(&s->gb)) |
| 948 | padding_zeroes = get_bits(&s->gb, 5); |
| 949 | else |
| 950 | padding_zeroes = 0; |
| 951 | |
| 952 | if (rawpcm_tile) { |
| 953 | int bits = s->bits_per_sample - padding_zeroes; |
| 954 | if (bits <= 0) { |
| 955 | av_log(s->avctx, AV_LOG_ERROR, |
| 956 | "Invalid number of padding bits in raw PCM tile\n"); |
| 957 | return AVERROR_INVALIDDATA; |
| 958 | } |
| 959 | av_dlog(s->avctx, "RAWPCM %d bits per sample. " |
| 960 | "total %d bits, remain=%d\n", bits, |
| 961 | bits * s->num_channels * subframe_len, get_bits_count(&s->gb)); |
| 962 | for (i = 0; i < s->num_channels; i++) |
| 963 | for (j = 0; j < subframe_len; j++) |
| 964 | s->channel_coeffs[i][j] = get_sbits_long(&s->gb, bits); |
| 965 | } else { |
| 966 | for (i = 0; i < s->num_channels; i++) |
| 967 | if (s->is_channel_coded[i]) { |
| 968 | decode_channel_residues(s, i, subframe_len); |
| 969 | if (s->seekable_tile) |
| 970 | use_high_update_speed(s, i); |
| 971 | else |
| 972 | use_normal_update_speed(s, i); |
| 973 | revert_cdlms(s, i, 0, subframe_len); |
| 974 | } else { |
| 975 | memset(s->channel_residues[i], 0, sizeof(**s->channel_residues) * subframe_len); |
| 976 | } |
| 977 | } |
| 978 | if (s->do_mclms) |
| 979 | revert_mclms(s, subframe_len); |
| 980 | if (s->do_inter_ch_decorr) |
| 981 | revert_inter_ch_decorr(s, subframe_len); |
| 982 | if (s->do_ac_filter) |
| 983 | revert_acfilter(s, subframe_len); |
| 984 | |
| 985 | /* Dequantize */ |
| 986 | if (s->quant_stepsize != 1) |
| 987 | for (i = 0; i < s->num_channels; i++) |
| 988 | for (j = 0; j < subframe_len; j++) |
| 989 | s->channel_residues[i][j] *= s->quant_stepsize; |
| 990 | |
| 991 | /* Write to proper output buffer depending on bit-depth */ |
| 992 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 993 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 994 | int subframe_len = s->channel[c].subframe_len[s->channel[c].cur_subframe]; |
| 995 | |
| 996 | for (j = 0; j < subframe_len; j++) { |
| 997 | if (s->bits_per_sample == 16) { |
| 998 | *s->samples_16[c]++ = (int16_t) s->channel_residues[c][j] << padding_zeroes; |
| 999 | } else { |
| 1000 | *s->samples_32[c]++ = s->channel_residues[c][j] << padding_zeroes; |
| 1001 | } |
| 1002 | } |
| 1003 | } |
| 1004 | |
| 1005 | /* handled one subframe */ |
| 1006 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1007 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1008 | if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) { |
| 1009 | av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n"); |
| 1010 | return AVERROR_INVALIDDATA; |
| 1011 | } |
| 1012 | ++s->channel[c].cur_subframe; |
| 1013 | } |
| 1014 | return 0; |
| 1015 | } |
| 1016 | |
| 1017 | /** |
| 1018 | * @brief Decode one WMA frame. |
| 1019 | * @param s codec context |
| 1020 | * @return 0 if the trailer bit indicates that this is the last frame, |
| 1021 | * 1 if there are additional frames |
| 1022 | */ |
| 1023 | static int decode_frame(WmallDecodeCtx *s) |
| 1024 | { |
| 1025 | GetBitContext* gb = &s->gb; |
| 1026 | int more_frames = 0, len = 0, i, ret; |
| 1027 | |
| 1028 | s->frame->nb_samples = s->samples_per_frame; |
| 1029 | if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) { |
| 1030 | /* return an error if no frame could be decoded at all */ |
| 1031 | s->packet_loss = 1; |
| 1032 | return ret; |
| 1033 | } |
| 1034 | for (i = 0; i < s->num_channels; i++) { |
| 1035 | s->samples_16[i] = (int16_t *)s->frame->extended_data[i]; |
| 1036 | s->samples_32[i] = (int32_t *)s->frame->extended_data[i]; |
| 1037 | } |
| 1038 | |
| 1039 | /* get frame length */ |
| 1040 | if (s->len_prefix) |
| 1041 | len = get_bits(gb, s->log2_frame_size); |
| 1042 | |
| 1043 | /* decode tile information */ |
| 1044 | if ((ret = decode_tilehdr(s))) { |
| 1045 | s->packet_loss = 1; |
| 1046 | av_frame_unref(s->frame); |
| 1047 | return ret; |
| 1048 | } |
| 1049 | |
| 1050 | /* read drc info */ |
| 1051 | if (s->dynamic_range_compression) |
| 1052 | s->drc_gain = get_bits(gb, 8); |
| 1053 | |
| 1054 | /* no idea what these are for, might be the number of samples |
| 1055 | that need to be skipped at the beginning or end of a stream */ |
| 1056 | if (get_bits1(gb)) { |
| 1057 | int av_unused skip; |
| 1058 | |
| 1059 | /* usually true for the first frame */ |
| 1060 | if (get_bits1(gb)) { |
| 1061 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); |
| 1062 | av_dlog(s->avctx, "start skip: %i\n", skip); |
| 1063 | } |
| 1064 | |
| 1065 | /* sometimes true for the last frame */ |
| 1066 | if (get_bits1(gb)) { |
| 1067 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); |
| 1068 | av_dlog(s->avctx, "end skip: %i\n", skip); |
| 1069 | } |
| 1070 | |
| 1071 | } |
| 1072 | |
| 1073 | /* reset subframe states */ |
| 1074 | s->parsed_all_subframes = 0; |
| 1075 | for (i = 0; i < s->num_channels; i++) { |
| 1076 | s->channel[i].decoded_samples = 0; |
| 1077 | s->channel[i].cur_subframe = 0; |
| 1078 | } |
| 1079 | |
| 1080 | /* decode all subframes */ |
| 1081 | while (!s->parsed_all_subframes) { |
| 1082 | int decoded_samples = s->channel[0].decoded_samples; |
| 1083 | if (decode_subframe(s) < 0) { |
| 1084 | s->packet_loss = 1; |
| 1085 | if (s->frame->nb_samples) |
| 1086 | s->frame->nb_samples = decoded_samples; |
| 1087 | return 0; |
| 1088 | } |
| 1089 | } |
| 1090 | |
| 1091 | av_dlog(s->avctx, "Frame done\n"); |
| 1092 | |
| 1093 | if (s->skip_frame) |
| 1094 | s->skip_frame = 0; |
| 1095 | |
| 1096 | if (s->len_prefix) { |
| 1097 | if (len != (get_bits_count(gb) - s->frame_offset) + 2) { |
| 1098 | /* FIXME: not sure if this is always an error */ |
| 1099 | av_log(s->avctx, AV_LOG_ERROR, |
| 1100 | "frame[%"PRIu32"] would have to skip %i bits\n", |
| 1101 | s->frame_num, |
| 1102 | len - (get_bits_count(gb) - s->frame_offset) - 1); |
| 1103 | s->packet_loss = 1; |
| 1104 | return 0; |
| 1105 | } |
| 1106 | |
| 1107 | /* skip the rest of the frame data */ |
| 1108 | skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1); |
| 1109 | } |
| 1110 | |
| 1111 | /* decode trailer bit */ |
| 1112 | more_frames = get_bits1(gb); |
| 1113 | ++s->frame_num; |
| 1114 | return more_frames; |
| 1115 | } |
| 1116 | |
| 1117 | /** |
| 1118 | * @brief Calculate remaining input buffer length. |
| 1119 | * @param s codec context |
| 1120 | * @param gb bitstream reader context |
| 1121 | * @return remaining size in bits |
| 1122 | */ |
| 1123 | static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb) |
| 1124 | { |
| 1125 | return s->buf_bit_size - get_bits_count(gb); |
| 1126 | } |
| 1127 | |
| 1128 | /** |
| 1129 | * @brief Fill the bit reservoir with a (partial) frame. |
| 1130 | * @param s codec context |
| 1131 | * @param gb bitstream reader context |
| 1132 | * @param len length of the partial frame |
| 1133 | * @param append decides whether to reset the buffer or not |
| 1134 | */ |
| 1135 | static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len, |
| 1136 | int append) |
| 1137 | { |
| 1138 | int buflen; |
| 1139 | PutBitContext tmp; |
| 1140 | |
| 1141 | /* when the frame data does not need to be concatenated, the input buffer |
| 1142 | is reset and additional bits from the previous frame are copied |
| 1143 | and skipped later so that a fast byte copy is possible */ |
| 1144 | |
| 1145 | if (!append) { |
| 1146 | s->frame_offset = get_bits_count(gb) & 7; |
| 1147 | s->num_saved_bits = s->frame_offset; |
| 1148 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
| 1149 | } |
| 1150 | |
| 1151 | buflen = (s->num_saved_bits + len + 8) >> 3; |
| 1152 | |
| 1153 | if (len <= 0 || buflen > MAX_FRAMESIZE) { |
| 1154 | avpriv_request_sample(s->avctx, "Too small input buffer"); |
| 1155 | s->packet_loss = 1; |
| 1156 | return; |
| 1157 | } |
| 1158 | |
| 1159 | s->num_saved_bits += len; |
| 1160 | if (!append) { |
| 1161 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), |
| 1162 | s->num_saved_bits); |
| 1163 | } else { |
| 1164 | int align = 8 - (get_bits_count(gb) & 7); |
| 1165 | align = FFMIN(align, len); |
| 1166 | put_bits(&s->pb, align, get_bits(gb, align)); |
| 1167 | len -= align; |
| 1168 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len); |
| 1169 | } |
| 1170 | skip_bits_long(gb, len); |
| 1171 | |
| 1172 | tmp = s->pb; |
| 1173 | flush_put_bits(&tmp); |
| 1174 | |
| 1175 | init_get_bits(&s->gb, s->frame_data, s->num_saved_bits); |
| 1176 | skip_bits(&s->gb, s->frame_offset); |
| 1177 | } |
| 1178 | |
| 1179 | static int decode_packet(AVCodecContext *avctx, void *data, int *got_frame_ptr, |
| 1180 | AVPacket* avpkt) |
| 1181 | { |
| 1182 | WmallDecodeCtx *s = avctx->priv_data; |
| 1183 | GetBitContext* gb = &s->pgb; |
| 1184 | const uint8_t* buf = avpkt->data; |
| 1185 | int buf_size = avpkt->size; |
| 1186 | int num_bits_prev_frame, packet_sequence_number, spliced_packet; |
| 1187 | |
| 1188 | s->frame->nb_samples = 0; |
| 1189 | |
| 1190 | if (s->packet_done || s->packet_loss) { |
| 1191 | s->packet_done = 0; |
| 1192 | |
| 1193 | if (!buf_size) |
| 1194 | return 0; |
| 1195 | /* sanity check for the buffer length */ |
| 1196 | if (buf_size < avctx->block_align) { |
| 1197 | av_log(avctx, AV_LOG_ERROR, "buf size %d invalid\n", buf_size); |
| 1198 | return AVERROR_INVALIDDATA; |
| 1199 | } |
| 1200 | |
| 1201 | s->next_packet_start = buf_size - avctx->block_align; |
| 1202 | buf_size = avctx->block_align; |
| 1203 | s->buf_bit_size = buf_size << 3; |
| 1204 | |
| 1205 | /* parse packet header */ |
| 1206 | init_get_bits(gb, buf, s->buf_bit_size); |
| 1207 | packet_sequence_number = get_bits(gb, 4); |
| 1208 | skip_bits(gb, 1); // Skip seekable_frame_in_packet, currently ununused |
| 1209 | spliced_packet = get_bits1(gb); |
| 1210 | if (spliced_packet) |
| 1211 | avpriv_request_sample(avctx, "Bitstream splicing"); |
| 1212 | |
| 1213 | /* get number of bits that need to be added to the previous frame */ |
| 1214 | num_bits_prev_frame = get_bits(gb, s->log2_frame_size); |
| 1215 | |
| 1216 | /* check for packet loss */ |
| 1217 | if (!s->packet_loss && |
| 1218 | ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) { |
| 1219 | s->packet_loss = 1; |
| 1220 | av_log(avctx, AV_LOG_ERROR, |
| 1221 | "Packet loss detected! seq %"PRIx8" vs %x\n", |
| 1222 | s->packet_sequence_number, packet_sequence_number); |
| 1223 | } |
| 1224 | s->packet_sequence_number = packet_sequence_number; |
| 1225 | |
| 1226 | if (num_bits_prev_frame > 0) { |
| 1227 | int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb); |
| 1228 | if (num_bits_prev_frame >= remaining_packet_bits) { |
| 1229 | num_bits_prev_frame = remaining_packet_bits; |
| 1230 | s->packet_done = 1; |
| 1231 | } |
| 1232 | |
| 1233 | /* Append the previous frame data to the remaining data from the |
| 1234 | * previous packet to create a full frame. */ |
| 1235 | save_bits(s, gb, num_bits_prev_frame, 1); |
| 1236 | |
| 1237 | /* decode the cross packet frame if it is valid */ |
| 1238 | if (num_bits_prev_frame < remaining_packet_bits && !s->packet_loss) |
| 1239 | decode_frame(s); |
| 1240 | } else if (s->num_saved_bits - s->frame_offset) { |
| 1241 | av_dlog(avctx, "ignoring %x previously saved bits\n", |
| 1242 | s->num_saved_bits - s->frame_offset); |
| 1243 | } |
| 1244 | |
| 1245 | if (s->packet_loss) { |
| 1246 | /* Reset number of saved bits so that the decoder does not start |
| 1247 | * to decode incomplete frames in the s->len_prefix == 0 case. */ |
| 1248 | s->num_saved_bits = 0; |
| 1249 | s->packet_loss = 0; |
| 1250 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
| 1251 | } |
| 1252 | |
| 1253 | } else { |
| 1254 | int frame_size; |
| 1255 | |
| 1256 | s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3; |
| 1257 | init_get_bits(gb, avpkt->data, s->buf_bit_size); |
| 1258 | skip_bits(gb, s->packet_offset); |
| 1259 | |
| 1260 | if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size && |
| 1261 | (frame_size = show_bits(gb, s->log2_frame_size)) && |
| 1262 | frame_size <= remaining_bits(s, gb)) { |
| 1263 | save_bits(s, gb, frame_size, 0); |
| 1264 | s->packet_done = !decode_frame(s); |
| 1265 | } else if (!s->len_prefix |
| 1266 | && s->num_saved_bits > get_bits_count(&s->gb)) { |
| 1267 | /* when the frames do not have a length prefix, we don't know the |
| 1268 | * compressed length of the individual frames however, we know what |
| 1269 | * part of a new packet belongs to the previous frame therefore we |
| 1270 | * save the incoming packet first, then we append the "previous |
| 1271 | * frame" data from the next packet so that we get a buffer that |
| 1272 | * only contains full frames */ |
| 1273 | s->packet_done = !decode_frame(s); |
| 1274 | } else { |
| 1275 | s->packet_done = 1; |
| 1276 | } |
| 1277 | } |
| 1278 | |
| 1279 | if (s->packet_done && !s->packet_loss && |
| 1280 | remaining_bits(s, gb) > 0) { |
| 1281 | /* save the rest of the data so that it can be decoded |
| 1282 | * with the next packet */ |
| 1283 | save_bits(s, gb, remaining_bits(s, gb), 0); |
| 1284 | } |
| 1285 | |
| 1286 | *got_frame_ptr = s->frame->nb_samples > 0; |
| 1287 | av_frame_move_ref(data, s->frame); |
| 1288 | |
| 1289 | s->packet_offset = get_bits_count(gb) & 7; |
| 1290 | |
| 1291 | return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3; |
| 1292 | } |
| 1293 | |
| 1294 | static void flush(AVCodecContext *avctx) |
| 1295 | { |
| 1296 | WmallDecodeCtx *s = avctx->priv_data; |
| 1297 | s->packet_loss = 1; |
| 1298 | s->packet_done = 0; |
| 1299 | s->num_saved_bits = 0; |
| 1300 | s->frame_offset = 0; |
| 1301 | s->next_packet_start = 0; |
| 1302 | s->cdlms[0][0].order = 0; |
| 1303 | s->frame->nb_samples = 0; |
| 1304 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
| 1305 | } |
| 1306 | |
| 1307 | static av_cold int decode_close(AVCodecContext *avctx) |
| 1308 | { |
| 1309 | WmallDecodeCtx *s = avctx->priv_data; |
| 1310 | |
| 1311 | av_frame_free(&s->frame); |
| 1312 | |
| 1313 | return 0; |
| 1314 | } |
| 1315 | |
| 1316 | AVCodec ff_wmalossless_decoder = { |
| 1317 | .name = "wmalossless", |
| 1318 | .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Lossless"), |
| 1319 | .type = AVMEDIA_TYPE_AUDIO, |
| 1320 | .id = AV_CODEC_ID_WMALOSSLESS, |
| 1321 | .priv_data_size = sizeof(WmallDecodeCtx), |
| 1322 | .init = decode_init, |
| 1323 | .close = decode_close, |
| 1324 | .decode = decode_packet, |
| 1325 | .flush = flush, |
| 1326 | .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1 | CODEC_CAP_DELAY, |
| 1327 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P, |
| 1328 | AV_SAMPLE_FMT_S32P, |
| 1329 | AV_SAMPLE_FMT_NONE }, |
| 1330 | }; |