| 1 | /* |
| 2 | * Wmapro compatible decoder |
| 3 | * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion |
| 4 | * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson |
| 5 | * |
| 6 | * This file is part of FFmpeg. |
| 7 | * |
| 8 | * FFmpeg is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU Lesser General Public |
| 10 | * License as published by the Free Software Foundation; either |
| 11 | * version 2.1 of the License, or (at your option) any later version. |
| 12 | * |
| 13 | * FFmpeg is distributed in the hope that it will be useful, |
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * Lesser General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU Lesser General Public |
| 19 | * License along with FFmpeg; if not, write to the Free Software |
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ |
| 22 | |
| 23 | /** |
| 24 | * @file |
| 25 | * @brief wmapro decoder implementation |
| 26 | * Wmapro is an MDCT based codec comparable to wma standard or AAC. |
| 27 | * The decoding therefore consists of the following steps: |
| 28 | * - bitstream decoding |
| 29 | * - reconstruction of per-channel data |
| 30 | * - rescaling and inverse quantization |
| 31 | * - IMDCT |
| 32 | * - windowing and overlapp-add |
| 33 | * |
| 34 | * The compressed wmapro bitstream is split into individual packets. |
| 35 | * Every such packet contains one or more wma frames. |
| 36 | * The compressed frames may have a variable length and frames may |
| 37 | * cross packet boundaries. |
| 38 | * Common to all wmapro frames is the number of samples that are stored in |
| 39 | * a frame. |
| 40 | * The number of samples and a few other decode flags are stored |
| 41 | * as extradata that has to be passed to the decoder. |
| 42 | * |
| 43 | * The wmapro frames themselves are again split into a variable number of |
| 44 | * subframes. Every subframe contains the data for 2^N time domain samples |
| 45 | * where N varies between 7 and 12. |
| 46 | * |
| 47 | * Example wmapro bitstream (in samples): |
| 48 | * |
| 49 | * || packet 0 || packet 1 || packet 2 packets |
| 50 | * --------------------------------------------------- |
| 51 | * || frame 0 || frame 1 || frame 2 || frames |
| 52 | * --------------------------------------------------- |
| 53 | * || | | || | | | || || subframes of channel 0 |
| 54 | * --------------------------------------------------- |
| 55 | * || | | || | | | || || subframes of channel 1 |
| 56 | * --------------------------------------------------- |
| 57 | * |
| 58 | * The frame layouts for the individual channels of a wma frame does not need |
| 59 | * to be the same. |
| 60 | * |
| 61 | * However, if the offsets and lengths of several subframes of a frame are the |
| 62 | * same, the subframes of the channels can be grouped. |
| 63 | * Every group may then use special coding techniques like M/S stereo coding |
| 64 | * to improve the compression ratio. These channel transformations do not |
| 65 | * need to be applied to a whole subframe. Instead, they can also work on |
| 66 | * individual scale factor bands (see below). |
| 67 | * The coefficients that carry the audio signal in the frequency domain |
| 68 | * are transmitted as huffman-coded vectors with 4, 2 and 1 elements. |
| 69 | * In addition to that, the encoder can switch to a runlevel coding scheme |
| 70 | * by transmitting subframe_length / 128 zero coefficients. |
| 71 | * |
| 72 | * Before the audio signal can be converted to the time domain, the |
| 73 | * coefficients have to be rescaled and inverse quantized. |
| 74 | * A subframe is therefore split into several scale factor bands that get |
| 75 | * scaled individually. |
| 76 | * Scale factors are submitted for every frame but they might be shared |
| 77 | * between the subframes of a channel. Scale factors are initially DPCM-coded. |
| 78 | * Once scale factors are shared, the differences are transmitted as runlevel |
| 79 | * codes. |
| 80 | * Every subframe length and offset combination in the frame layout shares a |
| 81 | * common quantization factor that can be adjusted for every channel by a |
| 82 | * modifier. |
| 83 | * After the inverse quantization, the coefficients get processed by an IMDCT. |
| 84 | * The resulting values are then windowed with a sine window and the first half |
| 85 | * of the values are added to the second half of the output from the previous |
| 86 | * subframe in order to reconstruct the output samples. |
| 87 | */ |
| 88 | |
| 89 | #include <inttypes.h> |
| 90 | |
| 91 | #include "libavutil/float_dsp.h" |
| 92 | #include "libavutil/intfloat.h" |
| 93 | #include "libavutil/intreadwrite.h" |
| 94 | #include "avcodec.h" |
| 95 | #include "internal.h" |
| 96 | #include "get_bits.h" |
| 97 | #include "put_bits.h" |
| 98 | #include "wmaprodata.h" |
| 99 | #include "sinewin.h" |
| 100 | #include "wma.h" |
| 101 | #include "wma_common.h" |
| 102 | |
| 103 | /** current decoder limitations */ |
| 104 | #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels |
| 105 | #define MAX_SUBFRAMES 32 ///< max number of subframes per channel |
| 106 | #define MAX_BANDS 29 ///< max number of scale factor bands |
| 107 | #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size |
| 108 | |
| 109 | #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size |
| 110 | #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size |
| 111 | #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size |
| 112 | #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size |
| 113 | #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes |
| 114 | |
| 115 | |
| 116 | #define VLCBITS 9 |
| 117 | #define SCALEVLCBITS 8 |
| 118 | #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS) |
| 119 | #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS) |
| 120 | #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS) |
| 121 | #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS) |
| 122 | #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS) |
| 123 | |
| 124 | static VLC sf_vlc; ///< scale factor DPCM vlc |
| 125 | static VLC sf_rl_vlc; ///< scale factor run length vlc |
| 126 | static VLC vec4_vlc; ///< 4 coefficients per symbol |
| 127 | static VLC vec2_vlc; ///< 2 coefficients per symbol |
| 128 | static VLC vec1_vlc; ///< 1 coefficient per symbol |
| 129 | static VLC coef_vlc[2]; ///< coefficient run length vlc codes |
| 130 | static float sin64[33]; ///< sine table for decorrelation |
| 131 | |
| 132 | /** |
| 133 | * @brief frame specific decoder context for a single channel |
| 134 | */ |
| 135 | typedef struct { |
| 136 | int16_t prev_block_len; ///< length of the previous block |
| 137 | uint8_t transmit_coefs; |
| 138 | uint8_t num_subframes; |
| 139 | uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples |
| 140 | uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame |
| 141 | uint8_t cur_subframe; ///< current subframe number |
| 142 | uint16_t decoded_samples; ///< number of already processed samples |
| 143 | uint8_t grouped; ///< channel is part of a group |
| 144 | int quant_step; ///< quantization step for the current subframe |
| 145 | int8_t reuse_sf; ///< share scale factors between subframes |
| 146 | int8_t scale_factor_step; ///< scaling step for the current subframe |
| 147 | int max_scale_factor; ///< maximum scale factor for the current subframe |
| 148 | int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values |
| 149 | int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling) |
| 150 | int* scale_factors; ///< pointer to the scale factor values used for decoding |
| 151 | uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block |
| 152 | float* coeffs; ///< pointer to the subframe decode buffer |
| 153 | uint16_t num_vec_coeffs; ///< number of vector coded coefficients |
| 154 | DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer |
| 155 | } WMAProChannelCtx; |
| 156 | |
| 157 | /** |
| 158 | * @brief channel group for channel transformations |
| 159 | */ |
| 160 | typedef struct { |
| 161 | uint8_t num_channels; ///< number of channels in the group |
| 162 | int8_t transform; ///< transform on / off |
| 163 | int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band |
| 164 | float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS]; |
| 165 | float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients |
| 166 | } WMAProChannelGrp; |
| 167 | |
| 168 | /** |
| 169 | * @brief main decoder context |
| 170 | */ |
| 171 | typedef struct WMAProDecodeCtx { |
| 172 | /* generic decoder variables */ |
| 173 | AVCodecContext* avctx; ///< codec context for av_log |
| 174 | AVFloatDSPContext *fdsp; |
| 175 | uint8_t frame_data[MAX_FRAMESIZE + |
| 176 | FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data |
| 177 | PutBitContext pb; ///< context for filling the frame_data buffer |
| 178 | FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size |
| 179 | DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer |
| 180 | const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes |
| 181 | |
| 182 | /* frame size dependent frame information (set during initialization) */ |
| 183 | uint32_t decode_flags; ///< used compression features |
| 184 | uint8_t len_prefix; ///< frame is prefixed with its length |
| 185 | uint8_t dynamic_range_compression; ///< frame contains DRC data |
| 186 | uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0]) |
| 187 | uint16_t samples_per_frame; ///< number of samples to output |
| 188 | uint16_t log2_frame_size; |
| 189 | int8_t lfe_channel; ///< lfe channel index |
| 190 | uint8_t max_num_subframes; |
| 191 | uint8_t subframe_len_bits; ///< number of bits used for the subframe length |
| 192 | uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1 |
| 193 | uint16_t min_samples_per_subframe; |
| 194 | int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size |
| 195 | int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4) |
| 196 | int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix |
| 197 | int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values |
| 198 | |
| 199 | /* packet decode state */ |
| 200 | GetBitContext pgb; ///< bitstream reader context for the packet |
| 201 | int next_packet_start; ///< start offset of the next wma packet in the demuxer packet |
| 202 | uint8_t packet_offset; ///< frame offset in the packet |
| 203 | uint8_t packet_sequence_number; ///< current packet number |
| 204 | int num_saved_bits; ///< saved number of bits |
| 205 | int frame_offset; ///< frame offset in the bit reservoir |
| 206 | int subframe_offset; ///< subframe offset in the bit reservoir |
| 207 | uint8_t packet_loss; ///< set in case of bitstream error |
| 208 | uint8_t packet_done; ///< set when a packet is fully decoded |
| 209 | |
| 210 | /* frame decode state */ |
| 211 | uint32_t frame_num; ///< current frame number (not used for decoding) |
| 212 | GetBitContext gb; ///< bitstream reader context |
| 213 | int buf_bit_size; ///< buffer size in bits |
| 214 | uint8_t drc_gain; ///< gain for the DRC tool |
| 215 | int8_t skip_frame; ///< skip output step |
| 216 | int8_t parsed_all_subframes; ///< all subframes decoded? |
| 217 | |
| 218 | /* subframe/block decode state */ |
| 219 | int16_t subframe_len; ///< current subframe length |
| 220 | int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe |
| 221 | int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS]; |
| 222 | int8_t num_bands; ///< number of scale factor bands |
| 223 | int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream |
| 224 | int16_t* cur_sfb_offsets; ///< sfb offsets for the current block |
| 225 | uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables |
| 226 | int8_t esc_len; ///< length of escaped coefficients |
| 227 | |
| 228 | uint8_t num_chgroups; ///< number of channel groups |
| 229 | WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information |
| 230 | |
| 231 | WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data |
| 232 | } WMAProDecodeCtx; |
| 233 | |
| 234 | |
| 235 | /** |
| 236 | *@brief helper function to print the most important members of the context |
| 237 | *@param s context |
| 238 | */ |
| 239 | static av_cold void dump_context(WMAProDecodeCtx *s) |
| 240 | { |
| 241 | #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b); |
| 242 | #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b); |
| 243 | |
| 244 | PRINT("ed sample bit depth", s->bits_per_sample); |
| 245 | PRINT_HEX("ed decode flags", s->decode_flags); |
| 246 | PRINT("samples per frame", s->samples_per_frame); |
| 247 | PRINT("log2 frame size", s->log2_frame_size); |
| 248 | PRINT("max num subframes", s->max_num_subframes); |
| 249 | PRINT("len prefix", s->len_prefix); |
| 250 | PRINT("num channels", s->avctx->channels); |
| 251 | } |
| 252 | |
| 253 | /** |
| 254 | *@brief Uninitialize the decoder and free all resources. |
| 255 | *@param avctx codec context |
| 256 | *@return 0 on success, < 0 otherwise |
| 257 | */ |
| 258 | static av_cold int decode_end(AVCodecContext *avctx) |
| 259 | { |
| 260 | WMAProDecodeCtx *s = avctx->priv_data; |
| 261 | int i; |
| 262 | |
| 263 | av_freep(&s->fdsp); |
| 264 | |
| 265 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) |
| 266 | ff_mdct_end(&s->mdct_ctx[i]); |
| 267 | |
| 268 | return 0; |
| 269 | } |
| 270 | |
| 271 | /** |
| 272 | *@brief Initialize the decoder. |
| 273 | *@param avctx codec context |
| 274 | *@return 0 on success, -1 otherwise |
| 275 | */ |
| 276 | static av_cold int decode_init(AVCodecContext *avctx) |
| 277 | { |
| 278 | WMAProDecodeCtx *s = avctx->priv_data; |
| 279 | uint8_t *edata_ptr = avctx->extradata; |
| 280 | unsigned int channel_mask; |
| 281 | int i, bits; |
| 282 | int log2_max_num_subframes; |
| 283 | int num_possible_block_sizes; |
| 284 | |
| 285 | if (!avctx->block_align) { |
| 286 | av_log(avctx, AV_LOG_ERROR, "block_align is not set\n"); |
| 287 | return AVERROR(EINVAL); |
| 288 | } |
| 289 | |
| 290 | s->avctx = avctx; |
| 291 | s->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT); |
| 292 | if (!s->fdsp) |
| 293 | return AVERROR(ENOMEM); |
| 294 | |
| 295 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
| 296 | |
| 297 | avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; |
| 298 | |
| 299 | if (avctx->extradata_size >= 18) { |
| 300 | s->decode_flags = AV_RL16(edata_ptr+14); |
| 301 | channel_mask = AV_RL32(edata_ptr+2); |
| 302 | s->bits_per_sample = AV_RL16(edata_ptr); |
| 303 | /** dump the extradata */ |
| 304 | for (i = 0; i < avctx->extradata_size; i++) |
| 305 | av_dlog(avctx, "[%x] ", avctx->extradata[i]); |
| 306 | av_dlog(avctx, "\n"); |
| 307 | |
| 308 | } else { |
| 309 | avpriv_request_sample(avctx, "Unknown extradata size"); |
| 310 | return AVERROR_PATCHWELCOME; |
| 311 | } |
| 312 | |
| 313 | /** generic init */ |
| 314 | s->log2_frame_size = av_log2(avctx->block_align) + 4; |
| 315 | if (s->log2_frame_size > 25) { |
| 316 | avpriv_request_sample(avctx, "Large block align"); |
| 317 | return AVERROR_PATCHWELCOME; |
| 318 | } |
| 319 | |
| 320 | /** frame info */ |
| 321 | s->skip_frame = 1; /* skip first frame */ |
| 322 | s->packet_loss = 1; |
| 323 | s->len_prefix = (s->decode_flags & 0x40); |
| 324 | |
| 325 | /** get frame len */ |
| 326 | bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags); |
| 327 | if (bits > WMAPRO_BLOCK_MAX_BITS) { |
| 328 | avpriv_request_sample(avctx, "14-bit block sizes"); |
| 329 | return AVERROR_PATCHWELCOME; |
| 330 | } |
| 331 | s->samples_per_frame = 1 << bits; |
| 332 | |
| 333 | /** subframe info */ |
| 334 | log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3); |
| 335 | s->max_num_subframes = 1 << log2_max_num_subframes; |
| 336 | if (s->max_num_subframes == 16 || s->max_num_subframes == 4) |
| 337 | s->max_subframe_len_bit = 1; |
| 338 | s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1; |
| 339 | |
| 340 | num_possible_block_sizes = log2_max_num_subframes + 1; |
| 341 | s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes; |
| 342 | s->dynamic_range_compression = (s->decode_flags & 0x80); |
| 343 | |
| 344 | if (s->max_num_subframes > MAX_SUBFRAMES) { |
| 345 | av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n", |
| 346 | s->max_num_subframes); |
| 347 | return AVERROR_INVALIDDATA; |
| 348 | } |
| 349 | |
| 350 | if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) { |
| 351 | av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n", |
| 352 | s->min_samples_per_subframe); |
| 353 | return AVERROR_INVALIDDATA; |
| 354 | } |
| 355 | |
| 356 | if (s->avctx->sample_rate <= 0) { |
| 357 | av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n"); |
| 358 | return AVERROR_INVALIDDATA; |
| 359 | } |
| 360 | |
| 361 | if (avctx->channels < 0) { |
| 362 | av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", |
| 363 | avctx->channels); |
| 364 | return AVERROR_INVALIDDATA; |
| 365 | } else if (avctx->channels > WMAPRO_MAX_CHANNELS) { |
| 366 | avpriv_request_sample(avctx, |
| 367 | "More than %d channels", WMAPRO_MAX_CHANNELS); |
| 368 | return AVERROR_PATCHWELCOME; |
| 369 | } |
| 370 | |
| 371 | /** init previous block len */ |
| 372 | for (i = 0; i < avctx->channels; i++) |
| 373 | s->channel[i].prev_block_len = s->samples_per_frame; |
| 374 | |
| 375 | /** extract lfe channel position */ |
| 376 | s->lfe_channel = -1; |
| 377 | |
| 378 | if (channel_mask & 8) { |
| 379 | unsigned int mask; |
| 380 | for (mask = 1; mask < 16; mask <<= 1) { |
| 381 | if (channel_mask & mask) |
| 382 | ++s->lfe_channel; |
| 383 | } |
| 384 | } |
| 385 | |
| 386 | INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE, |
| 387 | scale_huffbits, 1, 1, |
| 388 | scale_huffcodes, 2, 2, 616); |
| 389 | |
| 390 | INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE, |
| 391 | scale_rl_huffbits, 1, 1, |
| 392 | scale_rl_huffcodes, 4, 4, 1406); |
| 393 | |
| 394 | INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE, |
| 395 | coef0_huffbits, 1, 1, |
| 396 | coef0_huffcodes, 4, 4, 2108); |
| 397 | |
| 398 | INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE, |
| 399 | coef1_huffbits, 1, 1, |
| 400 | coef1_huffcodes, 4, 4, 3912); |
| 401 | |
| 402 | INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE, |
| 403 | vec4_huffbits, 1, 1, |
| 404 | vec4_huffcodes, 2, 2, 604); |
| 405 | |
| 406 | INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE, |
| 407 | vec2_huffbits, 1, 1, |
| 408 | vec2_huffcodes, 2, 2, 562); |
| 409 | |
| 410 | INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE, |
| 411 | vec1_huffbits, 1, 1, |
| 412 | vec1_huffcodes, 2, 2, 562); |
| 413 | |
| 414 | /** calculate number of scale factor bands and their offsets |
| 415 | for every possible block size */ |
| 416 | for (i = 0; i < num_possible_block_sizes; i++) { |
| 417 | int subframe_len = s->samples_per_frame >> i; |
| 418 | int x; |
| 419 | int band = 1; |
| 420 | |
| 421 | s->sfb_offsets[i][0] = 0; |
| 422 | |
| 423 | for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) { |
| 424 | int offset = (subframe_len * 2 * critical_freq[x]) |
| 425 | / s->avctx->sample_rate + 2; |
| 426 | offset &= ~3; |
| 427 | if (offset > s->sfb_offsets[i][band - 1]) |
| 428 | s->sfb_offsets[i][band++] = offset; |
| 429 | |
| 430 | if (offset >= subframe_len) |
| 431 | break; |
| 432 | } |
| 433 | s->sfb_offsets[i][band - 1] = subframe_len; |
| 434 | s->num_sfb[i] = band - 1; |
| 435 | if (s->num_sfb[i] <= 0) { |
| 436 | av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n"); |
| 437 | return AVERROR_INVALIDDATA; |
| 438 | } |
| 439 | } |
| 440 | |
| 441 | |
| 442 | /** Scale factors can be shared between blocks of different size |
| 443 | as every block has a different scale factor band layout. |
| 444 | The matrix sf_offsets is needed to find the correct scale factor. |
| 445 | */ |
| 446 | |
| 447 | for (i = 0; i < num_possible_block_sizes; i++) { |
| 448 | int b; |
| 449 | for (b = 0; b < s->num_sfb[i]; b++) { |
| 450 | int x; |
| 451 | int offset = ((s->sfb_offsets[i][b] |
| 452 | + s->sfb_offsets[i][b + 1] - 1) << i) >> 1; |
| 453 | for (x = 0; x < num_possible_block_sizes; x++) { |
| 454 | int v = 0; |
| 455 | while (s->sfb_offsets[x][v + 1] << x < offset) { |
| 456 | v++; |
| 457 | av_assert0(v < MAX_BANDS); |
| 458 | } |
| 459 | s->sf_offsets[i][x][b] = v; |
| 460 | } |
| 461 | } |
| 462 | } |
| 463 | |
| 464 | /** init MDCT, FIXME: only init needed sizes */ |
| 465 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) |
| 466 | ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1, |
| 467 | 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1)) |
| 468 | / (1 << (s->bits_per_sample - 1))); |
| 469 | |
| 470 | /** init MDCT windows: simple sine window */ |
| 471 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) { |
| 472 | const int win_idx = WMAPRO_BLOCK_MAX_BITS - i; |
| 473 | ff_init_ff_sine_windows(win_idx); |
| 474 | s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx]; |
| 475 | } |
| 476 | |
| 477 | /** calculate subwoofer cutoff values */ |
| 478 | for (i = 0; i < num_possible_block_sizes; i++) { |
| 479 | int block_size = s->samples_per_frame >> i; |
| 480 | int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1) |
| 481 | / s->avctx->sample_rate; |
| 482 | s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size); |
| 483 | } |
| 484 | |
| 485 | /** calculate sine values for the decorrelation matrix */ |
| 486 | for (i = 0; i < 33; i++) |
| 487 | sin64[i] = sin(i*M_PI / 64.0); |
| 488 | |
| 489 | if (avctx->debug & FF_DEBUG_BITSTREAM) |
| 490 | dump_context(s); |
| 491 | |
| 492 | avctx->channel_layout = channel_mask; |
| 493 | |
| 494 | return 0; |
| 495 | } |
| 496 | |
| 497 | /** |
| 498 | *@brief Decode the subframe length. |
| 499 | *@param s context |
| 500 | *@param offset sample offset in the frame |
| 501 | *@return decoded subframe length on success, < 0 in case of an error |
| 502 | */ |
| 503 | static int decode_subframe_length(WMAProDecodeCtx *s, int offset) |
| 504 | { |
| 505 | int frame_len_shift = 0; |
| 506 | int subframe_len; |
| 507 | |
| 508 | /** no need to read from the bitstream when only one length is possible */ |
| 509 | if (offset == s->samples_per_frame - s->min_samples_per_subframe) |
| 510 | return s->min_samples_per_subframe; |
| 511 | |
| 512 | if (get_bits_left(&s->gb) < 1) |
| 513 | return AVERROR_INVALIDDATA; |
| 514 | |
| 515 | /** 1 bit indicates if the subframe is of maximum length */ |
| 516 | if (s->max_subframe_len_bit) { |
| 517 | if (get_bits1(&s->gb)) |
| 518 | frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1); |
| 519 | } else |
| 520 | frame_len_shift = get_bits(&s->gb, s->subframe_len_bits); |
| 521 | |
| 522 | subframe_len = s->samples_per_frame >> frame_len_shift; |
| 523 | |
| 524 | /** sanity check the length */ |
| 525 | if (subframe_len < s->min_samples_per_subframe || |
| 526 | subframe_len > s->samples_per_frame) { |
| 527 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n", |
| 528 | subframe_len); |
| 529 | return AVERROR_INVALIDDATA; |
| 530 | } |
| 531 | return subframe_len; |
| 532 | } |
| 533 | |
| 534 | /** |
| 535 | *@brief Decode how the data in the frame is split into subframes. |
| 536 | * Every WMA frame contains the encoded data for a fixed number of |
| 537 | * samples per channel. The data for every channel might be split |
| 538 | * into several subframes. This function will reconstruct the list of |
| 539 | * subframes for every channel. |
| 540 | * |
| 541 | * If the subframes are not evenly split, the algorithm estimates the |
| 542 | * channels with the lowest number of total samples. |
| 543 | * Afterwards, for each of these channels a bit is read from the |
| 544 | * bitstream that indicates if the channel contains a subframe with the |
| 545 | * next subframe size that is going to be read from the bitstream or not. |
| 546 | * If a channel contains such a subframe, the subframe size gets added to |
| 547 | * the channel's subframe list. |
| 548 | * The algorithm repeats these steps until the frame is properly divided |
| 549 | * between the individual channels. |
| 550 | * |
| 551 | *@param s context |
| 552 | *@return 0 on success, < 0 in case of an error |
| 553 | */ |
| 554 | static int decode_tilehdr(WMAProDecodeCtx *s) |
| 555 | { |
| 556 | uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */ |
| 557 | uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */ |
| 558 | int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */ |
| 559 | int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */ |
| 560 | int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */ |
| 561 | int c; |
| 562 | |
| 563 | /* Should never consume more than 3073 bits (256 iterations for the |
| 564 | * while loop when always the minimum amount of 128 samples is subtracted |
| 565 | * from missing samples in the 8 channel case). |
| 566 | * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4) |
| 567 | */ |
| 568 | |
| 569 | /** reset tiling information */ |
| 570 | for (c = 0; c < s->avctx->channels; c++) |
| 571 | s->channel[c].num_subframes = 0; |
| 572 | |
| 573 | if (s->max_num_subframes == 1 || get_bits1(&s->gb)) |
| 574 | fixed_channel_layout = 1; |
| 575 | |
| 576 | /** loop until the frame data is split between the subframes */ |
| 577 | do { |
| 578 | int subframe_len; |
| 579 | |
| 580 | /** check which channels contain the subframe */ |
| 581 | for (c = 0; c < s->avctx->channels; c++) { |
| 582 | if (num_samples[c] == min_channel_len) { |
| 583 | if (fixed_channel_layout || channels_for_cur_subframe == 1 || |
| 584 | (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) |
| 585 | contains_subframe[c] = 1; |
| 586 | else |
| 587 | contains_subframe[c] = get_bits1(&s->gb); |
| 588 | } else |
| 589 | contains_subframe[c] = 0; |
| 590 | } |
| 591 | |
| 592 | /** get subframe length, subframe_len == 0 is not allowed */ |
| 593 | if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0) |
| 594 | return AVERROR_INVALIDDATA; |
| 595 | |
| 596 | /** add subframes to the individual channels and find new min_channel_len */ |
| 597 | min_channel_len += subframe_len; |
| 598 | for (c = 0; c < s->avctx->channels; c++) { |
| 599 | WMAProChannelCtx* chan = &s->channel[c]; |
| 600 | |
| 601 | if (contains_subframe[c]) { |
| 602 | if (chan->num_subframes >= MAX_SUBFRAMES) { |
| 603 | av_log(s->avctx, AV_LOG_ERROR, |
| 604 | "broken frame: num subframes > 31\n"); |
| 605 | return AVERROR_INVALIDDATA; |
| 606 | } |
| 607 | chan->subframe_len[chan->num_subframes] = subframe_len; |
| 608 | num_samples[c] += subframe_len; |
| 609 | ++chan->num_subframes; |
| 610 | if (num_samples[c] > s->samples_per_frame) { |
| 611 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: " |
| 612 | "channel len > samples_per_frame\n"); |
| 613 | return AVERROR_INVALIDDATA; |
| 614 | } |
| 615 | } else if (num_samples[c] <= min_channel_len) { |
| 616 | if (num_samples[c] < min_channel_len) { |
| 617 | channels_for_cur_subframe = 0; |
| 618 | min_channel_len = num_samples[c]; |
| 619 | } |
| 620 | ++channels_for_cur_subframe; |
| 621 | } |
| 622 | } |
| 623 | } while (min_channel_len < s->samples_per_frame); |
| 624 | |
| 625 | for (c = 0; c < s->avctx->channels; c++) { |
| 626 | int i; |
| 627 | int offset = 0; |
| 628 | for (i = 0; i < s->channel[c].num_subframes; i++) { |
| 629 | av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]" |
| 630 | " len %i\n", s->frame_num, c, i, |
| 631 | s->channel[c].subframe_len[i]); |
| 632 | s->channel[c].subframe_offset[i] = offset; |
| 633 | offset += s->channel[c].subframe_len[i]; |
| 634 | } |
| 635 | } |
| 636 | |
| 637 | return 0; |
| 638 | } |
| 639 | |
| 640 | /** |
| 641 | *@brief Calculate a decorrelation matrix from the bitstream parameters. |
| 642 | *@param s codec context |
| 643 | *@param chgroup channel group for which the matrix needs to be calculated |
| 644 | */ |
| 645 | static void decode_decorrelation_matrix(WMAProDecodeCtx *s, |
| 646 | WMAProChannelGrp *chgroup) |
| 647 | { |
| 648 | int i; |
| 649 | int offset = 0; |
| 650 | int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS]; |
| 651 | memset(chgroup->decorrelation_matrix, 0, s->avctx->channels * |
| 652 | s->avctx->channels * sizeof(*chgroup->decorrelation_matrix)); |
| 653 | |
| 654 | for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++) |
| 655 | rotation_offset[i] = get_bits(&s->gb, 6); |
| 656 | |
| 657 | for (i = 0; i < chgroup->num_channels; i++) |
| 658 | chgroup->decorrelation_matrix[chgroup->num_channels * i + i] = |
| 659 | get_bits1(&s->gb) ? 1.0 : -1.0; |
| 660 | |
| 661 | for (i = 1; i < chgroup->num_channels; i++) { |
| 662 | int x; |
| 663 | for (x = 0; x < i; x++) { |
| 664 | int y; |
| 665 | for (y = 0; y < i + 1; y++) { |
| 666 | float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y]; |
| 667 | float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y]; |
| 668 | int n = rotation_offset[offset + x]; |
| 669 | float sinv; |
| 670 | float cosv; |
| 671 | |
| 672 | if (n < 32) { |
| 673 | sinv = sin64[n]; |
| 674 | cosv = sin64[32 - n]; |
| 675 | } else { |
| 676 | sinv = sin64[64 - n]; |
| 677 | cosv = -sin64[n - 32]; |
| 678 | } |
| 679 | |
| 680 | chgroup->decorrelation_matrix[y + x * chgroup->num_channels] = |
| 681 | (v1 * sinv) - (v2 * cosv); |
| 682 | chgroup->decorrelation_matrix[y + i * chgroup->num_channels] = |
| 683 | (v1 * cosv) + (v2 * sinv); |
| 684 | } |
| 685 | } |
| 686 | offset += i; |
| 687 | } |
| 688 | } |
| 689 | |
| 690 | /** |
| 691 | *@brief Decode channel transformation parameters |
| 692 | *@param s codec context |
| 693 | *@return >= 0 in case of success, < 0 in case of bitstream errors |
| 694 | */ |
| 695 | static int decode_channel_transform(WMAProDecodeCtx* s) |
| 696 | { |
| 697 | int i; |
| 698 | /* should never consume more than 1921 bits for the 8 channel case |
| 699 | * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS |
| 700 | * + MAX_CHANNELS + MAX_BANDS + 1) |
| 701 | */ |
| 702 | |
| 703 | /** in the one channel case channel transforms are pointless */ |
| 704 | s->num_chgroups = 0; |
| 705 | if (s->avctx->channels > 1) { |
| 706 | int remaining_channels = s->channels_for_cur_subframe; |
| 707 | |
| 708 | if (get_bits1(&s->gb)) { |
| 709 | avpriv_request_sample(s->avctx, |
| 710 | "Channel transform bit"); |
| 711 | return AVERROR_PATCHWELCOME; |
| 712 | } |
| 713 | |
| 714 | for (s->num_chgroups = 0; remaining_channels && |
| 715 | s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) { |
| 716 | WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups]; |
| 717 | float** channel_data = chgroup->channel_data; |
| 718 | chgroup->num_channels = 0; |
| 719 | chgroup->transform = 0; |
| 720 | |
| 721 | /** decode channel mask */ |
| 722 | if (remaining_channels > 2) { |
| 723 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 724 | int channel_idx = s->channel_indexes_for_cur_subframe[i]; |
| 725 | if (!s->channel[channel_idx].grouped |
| 726 | && get_bits1(&s->gb)) { |
| 727 | ++chgroup->num_channels; |
| 728 | s->channel[channel_idx].grouped = 1; |
| 729 | *channel_data++ = s->channel[channel_idx].coeffs; |
| 730 | } |
| 731 | } |
| 732 | } else { |
| 733 | chgroup->num_channels = remaining_channels; |
| 734 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 735 | int channel_idx = s->channel_indexes_for_cur_subframe[i]; |
| 736 | if (!s->channel[channel_idx].grouped) |
| 737 | *channel_data++ = s->channel[channel_idx].coeffs; |
| 738 | s->channel[channel_idx].grouped = 1; |
| 739 | } |
| 740 | } |
| 741 | |
| 742 | /** decode transform type */ |
| 743 | if (chgroup->num_channels == 2) { |
| 744 | if (get_bits1(&s->gb)) { |
| 745 | if (get_bits1(&s->gb)) { |
| 746 | avpriv_request_sample(s->avctx, |
| 747 | "Unknown channel transform type"); |
| 748 | return AVERROR_PATCHWELCOME; |
| 749 | } |
| 750 | } else { |
| 751 | chgroup->transform = 1; |
| 752 | if (s->avctx->channels == 2) { |
| 753 | chgroup->decorrelation_matrix[0] = 1.0; |
| 754 | chgroup->decorrelation_matrix[1] = -1.0; |
| 755 | chgroup->decorrelation_matrix[2] = 1.0; |
| 756 | chgroup->decorrelation_matrix[3] = 1.0; |
| 757 | } else { |
| 758 | /** cos(pi/4) */ |
| 759 | chgroup->decorrelation_matrix[0] = 0.70703125; |
| 760 | chgroup->decorrelation_matrix[1] = -0.70703125; |
| 761 | chgroup->decorrelation_matrix[2] = 0.70703125; |
| 762 | chgroup->decorrelation_matrix[3] = 0.70703125; |
| 763 | } |
| 764 | } |
| 765 | } else if (chgroup->num_channels > 2) { |
| 766 | if (get_bits1(&s->gb)) { |
| 767 | chgroup->transform = 1; |
| 768 | if (get_bits1(&s->gb)) { |
| 769 | decode_decorrelation_matrix(s, chgroup); |
| 770 | } else { |
| 771 | /** FIXME: more than 6 coupled channels not supported */ |
| 772 | if (chgroup->num_channels > 6) { |
| 773 | avpriv_request_sample(s->avctx, |
| 774 | "Coupled channels > 6"); |
| 775 | } else { |
| 776 | memcpy(chgroup->decorrelation_matrix, |
| 777 | default_decorrelation[chgroup->num_channels], |
| 778 | chgroup->num_channels * chgroup->num_channels * |
| 779 | sizeof(*chgroup->decorrelation_matrix)); |
| 780 | } |
| 781 | } |
| 782 | } |
| 783 | } |
| 784 | |
| 785 | /** decode transform on / off */ |
| 786 | if (chgroup->transform) { |
| 787 | if (!get_bits1(&s->gb)) { |
| 788 | int i; |
| 789 | /** transform can be enabled for individual bands */ |
| 790 | for (i = 0; i < s->num_bands; i++) { |
| 791 | chgroup->transform_band[i] = get_bits1(&s->gb); |
| 792 | } |
| 793 | } else { |
| 794 | memset(chgroup->transform_band, 1, s->num_bands); |
| 795 | } |
| 796 | } |
| 797 | remaining_channels -= chgroup->num_channels; |
| 798 | } |
| 799 | } |
| 800 | return 0; |
| 801 | } |
| 802 | |
| 803 | /** |
| 804 | *@brief Extract the coefficients from the bitstream. |
| 805 | *@param s codec context |
| 806 | *@param c current channel number |
| 807 | *@return 0 on success, < 0 in case of bitstream errors |
| 808 | */ |
| 809 | static int decode_coeffs(WMAProDecodeCtx *s, int c) |
| 810 | { |
| 811 | /* Integers 0..15 as single-precision floats. The table saves a |
| 812 | costly int to float conversion, and storing the values as |
| 813 | integers allows fast sign-flipping. */ |
| 814 | static const uint32_t fval_tab[16] = { |
| 815 | 0x00000000, 0x3f800000, 0x40000000, 0x40400000, |
| 816 | 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000, |
| 817 | 0x41000000, 0x41100000, 0x41200000, 0x41300000, |
| 818 | 0x41400000, 0x41500000, 0x41600000, 0x41700000, |
| 819 | }; |
| 820 | int vlctable; |
| 821 | VLC* vlc; |
| 822 | WMAProChannelCtx* ci = &s->channel[c]; |
| 823 | int rl_mode = 0; |
| 824 | int cur_coeff = 0; |
| 825 | int num_zeros = 0; |
| 826 | const uint16_t* run; |
| 827 | const float* level; |
| 828 | |
| 829 | av_dlog(s->avctx, "decode coefficients for channel %i\n", c); |
| 830 | |
| 831 | vlctable = get_bits1(&s->gb); |
| 832 | vlc = &coef_vlc[vlctable]; |
| 833 | |
| 834 | if (vlctable) { |
| 835 | run = coef1_run; |
| 836 | level = coef1_level; |
| 837 | } else { |
| 838 | run = coef0_run; |
| 839 | level = coef0_level; |
| 840 | } |
| 841 | |
| 842 | /** decode vector coefficients (consumes up to 167 bits per iteration for |
| 843 | 4 vector coded large values) */ |
| 844 | while ((s->transmit_num_vec_coeffs || !rl_mode) && |
| 845 | (cur_coeff + 3 < ci->num_vec_coeffs)) { |
| 846 | uint32_t vals[4]; |
| 847 | int i; |
| 848 | unsigned int idx; |
| 849 | |
| 850 | idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH); |
| 851 | |
| 852 | if (idx == HUFF_VEC4_SIZE - 1) { |
| 853 | for (i = 0; i < 4; i += 2) { |
| 854 | idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH); |
| 855 | if (idx == HUFF_VEC2_SIZE - 1) { |
| 856 | uint32_t v0, v1; |
| 857 | v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH); |
| 858 | if (v0 == HUFF_VEC1_SIZE - 1) |
| 859 | v0 += ff_wma_get_large_val(&s->gb); |
| 860 | v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH); |
| 861 | if (v1 == HUFF_VEC1_SIZE - 1) |
| 862 | v1 += ff_wma_get_large_val(&s->gb); |
| 863 | vals[i ] = av_float2int(v0); |
| 864 | vals[i+1] = av_float2int(v1); |
| 865 | } else { |
| 866 | vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ]; |
| 867 | vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF]; |
| 868 | } |
| 869 | } |
| 870 | } else { |
| 871 | vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ]; |
| 872 | vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF]; |
| 873 | vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF]; |
| 874 | vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF]; |
| 875 | } |
| 876 | |
| 877 | /** decode sign */ |
| 878 | for (i = 0; i < 4; i++) { |
| 879 | if (vals[i]) { |
| 880 | uint32_t sign = get_bits1(&s->gb) - 1; |
| 881 | AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31); |
| 882 | num_zeros = 0; |
| 883 | } else { |
| 884 | ci->coeffs[cur_coeff] = 0; |
| 885 | /** switch to run level mode when subframe_len / 128 zeros |
| 886 | were found in a row */ |
| 887 | rl_mode |= (++num_zeros > s->subframe_len >> 8); |
| 888 | } |
| 889 | ++cur_coeff; |
| 890 | } |
| 891 | } |
| 892 | |
| 893 | /** decode run level coded coefficients */ |
| 894 | if (cur_coeff < s->subframe_len) { |
| 895 | memset(&ci->coeffs[cur_coeff], 0, |
| 896 | sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff)); |
| 897 | if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc, |
| 898 | level, run, 1, ci->coeffs, |
| 899 | cur_coeff, s->subframe_len, |
| 900 | s->subframe_len, s->esc_len, 0)) |
| 901 | return AVERROR_INVALIDDATA; |
| 902 | } |
| 903 | |
| 904 | return 0; |
| 905 | } |
| 906 | |
| 907 | /** |
| 908 | *@brief Extract scale factors from the bitstream. |
| 909 | *@param s codec context |
| 910 | *@return 0 on success, < 0 in case of bitstream errors |
| 911 | */ |
| 912 | static int decode_scale_factors(WMAProDecodeCtx* s) |
| 913 | { |
| 914 | int i; |
| 915 | |
| 916 | /** should never consume more than 5344 bits |
| 917 | * MAX_CHANNELS * (1 + MAX_BANDS * 23) |
| 918 | */ |
| 919 | |
| 920 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 921 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 922 | int* sf; |
| 923 | int* sf_end; |
| 924 | s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx]; |
| 925 | sf_end = s->channel[c].scale_factors + s->num_bands; |
| 926 | |
| 927 | /** resample scale factors for the new block size |
| 928 | * as the scale factors might need to be resampled several times |
| 929 | * before some new values are transmitted, a backup of the last |
| 930 | * transmitted scale factors is kept in saved_scale_factors |
| 931 | */ |
| 932 | if (s->channel[c].reuse_sf) { |
| 933 | const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx]; |
| 934 | int b; |
| 935 | for (b = 0; b < s->num_bands; b++) |
| 936 | s->channel[c].scale_factors[b] = |
| 937 | s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++]; |
| 938 | } |
| 939 | |
| 940 | if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) { |
| 941 | |
| 942 | if (!s->channel[c].reuse_sf) { |
| 943 | int val; |
| 944 | /** decode DPCM coded scale factors */ |
| 945 | s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1; |
| 946 | val = 45 / s->channel[c].scale_factor_step; |
| 947 | for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) { |
| 948 | val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60; |
| 949 | *sf = val; |
| 950 | } |
| 951 | } else { |
| 952 | int i; |
| 953 | /** run level decode differences to the resampled factors */ |
| 954 | for (i = 0; i < s->num_bands; i++) { |
| 955 | int idx; |
| 956 | int skip; |
| 957 | int val; |
| 958 | int sign; |
| 959 | |
| 960 | idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH); |
| 961 | |
| 962 | if (!idx) { |
| 963 | uint32_t code = get_bits(&s->gb, 14); |
| 964 | val = code >> 6; |
| 965 | sign = (code & 1) - 1; |
| 966 | skip = (code & 0x3f) >> 1; |
| 967 | } else if (idx == 1) { |
| 968 | break; |
| 969 | } else { |
| 970 | skip = scale_rl_run[idx]; |
| 971 | val = scale_rl_level[idx]; |
| 972 | sign = get_bits1(&s->gb)-1; |
| 973 | } |
| 974 | |
| 975 | i += skip; |
| 976 | if (i >= s->num_bands) { |
| 977 | av_log(s->avctx, AV_LOG_ERROR, |
| 978 | "invalid scale factor coding\n"); |
| 979 | return AVERROR_INVALIDDATA; |
| 980 | } |
| 981 | s->channel[c].scale_factors[i] += (val ^ sign) - sign; |
| 982 | } |
| 983 | } |
| 984 | /** swap buffers */ |
| 985 | s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx; |
| 986 | s->channel[c].table_idx = s->table_idx; |
| 987 | s->channel[c].reuse_sf = 1; |
| 988 | } |
| 989 | |
| 990 | /** calculate new scale factor maximum */ |
| 991 | s->channel[c].max_scale_factor = s->channel[c].scale_factors[0]; |
| 992 | for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) { |
| 993 | s->channel[c].max_scale_factor = |
| 994 | FFMAX(s->channel[c].max_scale_factor, *sf); |
| 995 | } |
| 996 | |
| 997 | } |
| 998 | return 0; |
| 999 | } |
| 1000 | |
| 1001 | /** |
| 1002 | *@brief Reconstruct the individual channel data. |
| 1003 | *@param s codec context |
| 1004 | */ |
| 1005 | static void inverse_channel_transform(WMAProDecodeCtx *s) |
| 1006 | { |
| 1007 | int i; |
| 1008 | |
| 1009 | for (i = 0; i < s->num_chgroups; i++) { |
| 1010 | if (s->chgroup[i].transform) { |
| 1011 | float data[WMAPRO_MAX_CHANNELS]; |
| 1012 | const int num_channels = s->chgroup[i].num_channels; |
| 1013 | float** ch_data = s->chgroup[i].channel_data; |
| 1014 | float** ch_end = ch_data + num_channels; |
| 1015 | const int8_t* tb = s->chgroup[i].transform_band; |
| 1016 | int16_t* sfb; |
| 1017 | |
| 1018 | /** multichannel decorrelation */ |
| 1019 | for (sfb = s->cur_sfb_offsets; |
| 1020 | sfb < s->cur_sfb_offsets + s->num_bands; sfb++) { |
| 1021 | int y; |
| 1022 | if (*tb++ == 1) { |
| 1023 | /** multiply values with the decorrelation_matrix */ |
| 1024 | for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) { |
| 1025 | const float* mat = s->chgroup[i].decorrelation_matrix; |
| 1026 | const float* data_end = data + num_channels; |
| 1027 | float* data_ptr = data; |
| 1028 | float** ch; |
| 1029 | |
| 1030 | for (ch = ch_data; ch < ch_end; ch++) |
| 1031 | *data_ptr++ = (*ch)[y]; |
| 1032 | |
| 1033 | for (ch = ch_data; ch < ch_end; ch++) { |
| 1034 | float sum = 0; |
| 1035 | data_ptr = data; |
| 1036 | while (data_ptr < data_end) |
| 1037 | sum += *data_ptr++ * *mat++; |
| 1038 | |
| 1039 | (*ch)[y] = sum; |
| 1040 | } |
| 1041 | } |
| 1042 | } else if (s->avctx->channels == 2) { |
| 1043 | int len = FFMIN(sfb[1], s->subframe_len) - sfb[0]; |
| 1044 | s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0], |
| 1045 | ch_data[0] + sfb[0], |
| 1046 | 181.0 / 128, len); |
| 1047 | s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0], |
| 1048 | ch_data[1] + sfb[0], |
| 1049 | 181.0 / 128, len); |
| 1050 | } |
| 1051 | } |
| 1052 | } |
| 1053 | } |
| 1054 | } |
| 1055 | |
| 1056 | /** |
| 1057 | *@brief Apply sine window and reconstruct the output buffer. |
| 1058 | *@param s codec context |
| 1059 | */ |
| 1060 | static void wmapro_window(WMAProDecodeCtx *s) |
| 1061 | { |
| 1062 | int i; |
| 1063 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1064 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1065 | const float* window; |
| 1066 | int winlen = s->channel[c].prev_block_len; |
| 1067 | float* start = s->channel[c].coeffs - (winlen >> 1); |
| 1068 | |
| 1069 | if (s->subframe_len < winlen) { |
| 1070 | start += (winlen - s->subframe_len) >> 1; |
| 1071 | winlen = s->subframe_len; |
| 1072 | } |
| 1073 | |
| 1074 | window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS]; |
| 1075 | |
| 1076 | winlen >>= 1; |
| 1077 | |
| 1078 | s->fdsp->vector_fmul_window(start, start, start + winlen, |
| 1079 | window, winlen); |
| 1080 | |
| 1081 | s->channel[c].prev_block_len = s->subframe_len; |
| 1082 | } |
| 1083 | } |
| 1084 | |
| 1085 | /** |
| 1086 | *@brief Decode a single subframe (block). |
| 1087 | *@param s codec context |
| 1088 | *@return 0 on success, < 0 when decoding failed |
| 1089 | */ |
| 1090 | static int decode_subframe(WMAProDecodeCtx *s) |
| 1091 | { |
| 1092 | int offset = s->samples_per_frame; |
| 1093 | int subframe_len = s->samples_per_frame; |
| 1094 | int i; |
| 1095 | int total_samples = s->samples_per_frame * s->avctx->channels; |
| 1096 | int transmit_coeffs = 0; |
| 1097 | int cur_subwoofer_cutoff; |
| 1098 | |
| 1099 | s->subframe_offset = get_bits_count(&s->gb); |
| 1100 | |
| 1101 | /** reset channel context and find the next block offset and size |
| 1102 | == the next block of the channel with the smallest number of |
| 1103 | decoded samples |
| 1104 | */ |
| 1105 | for (i = 0; i < s->avctx->channels; i++) { |
| 1106 | s->channel[i].grouped = 0; |
| 1107 | if (offset > s->channel[i].decoded_samples) { |
| 1108 | offset = s->channel[i].decoded_samples; |
| 1109 | subframe_len = |
| 1110 | s->channel[i].subframe_len[s->channel[i].cur_subframe]; |
| 1111 | } |
| 1112 | } |
| 1113 | |
| 1114 | av_dlog(s->avctx, |
| 1115 | "processing subframe with offset %i len %i\n", offset, subframe_len); |
| 1116 | |
| 1117 | /** get a list of all channels that contain the estimated block */ |
| 1118 | s->channels_for_cur_subframe = 0; |
| 1119 | for (i = 0; i < s->avctx->channels; i++) { |
| 1120 | const int cur_subframe = s->channel[i].cur_subframe; |
| 1121 | /** subtract already processed samples */ |
| 1122 | total_samples -= s->channel[i].decoded_samples; |
| 1123 | |
| 1124 | /** and count if there are multiple subframes that match our profile */ |
| 1125 | if (offset == s->channel[i].decoded_samples && |
| 1126 | subframe_len == s->channel[i].subframe_len[cur_subframe]) { |
| 1127 | total_samples -= s->channel[i].subframe_len[cur_subframe]; |
| 1128 | s->channel[i].decoded_samples += |
| 1129 | s->channel[i].subframe_len[cur_subframe]; |
| 1130 | s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i; |
| 1131 | ++s->channels_for_cur_subframe; |
| 1132 | } |
| 1133 | } |
| 1134 | |
| 1135 | /** check if the frame will be complete after processing the |
| 1136 | estimated block */ |
| 1137 | if (!total_samples) |
| 1138 | s->parsed_all_subframes = 1; |
| 1139 | |
| 1140 | |
| 1141 | av_dlog(s->avctx, "subframe is part of %i channels\n", |
| 1142 | s->channels_for_cur_subframe); |
| 1143 | |
| 1144 | /** calculate number of scale factor bands and their offsets */ |
| 1145 | s->table_idx = av_log2(s->samples_per_frame/subframe_len); |
| 1146 | s->num_bands = s->num_sfb[s->table_idx]; |
| 1147 | s->cur_sfb_offsets = s->sfb_offsets[s->table_idx]; |
| 1148 | cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx]; |
| 1149 | |
| 1150 | /** configure the decoder for the current subframe */ |
| 1151 | offset += s->samples_per_frame >> 1; |
| 1152 | |
| 1153 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1154 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1155 | |
| 1156 | s->channel[c].coeffs = &s->channel[c].out[offset]; |
| 1157 | } |
| 1158 | |
| 1159 | s->subframe_len = subframe_len; |
| 1160 | s->esc_len = av_log2(s->subframe_len - 1) + 1; |
| 1161 | |
| 1162 | /** skip extended header if any */ |
| 1163 | if (get_bits1(&s->gb)) { |
| 1164 | int num_fill_bits; |
| 1165 | if (!(num_fill_bits = get_bits(&s->gb, 2))) { |
| 1166 | int len = get_bits(&s->gb, 4); |
| 1167 | num_fill_bits = (len ? get_bits(&s->gb, len) : 0) + 1; |
| 1168 | } |
| 1169 | |
| 1170 | if (num_fill_bits >= 0) { |
| 1171 | if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) { |
| 1172 | av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n"); |
| 1173 | return AVERROR_INVALIDDATA; |
| 1174 | } |
| 1175 | |
| 1176 | skip_bits_long(&s->gb, num_fill_bits); |
| 1177 | } |
| 1178 | } |
| 1179 | |
| 1180 | /** no idea for what the following bit is used */ |
| 1181 | if (get_bits1(&s->gb)) { |
| 1182 | avpriv_request_sample(s->avctx, "Reserved bit"); |
| 1183 | return AVERROR_PATCHWELCOME; |
| 1184 | } |
| 1185 | |
| 1186 | |
| 1187 | if (decode_channel_transform(s) < 0) |
| 1188 | return AVERROR_INVALIDDATA; |
| 1189 | |
| 1190 | |
| 1191 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1192 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1193 | if ((s->channel[c].transmit_coefs = get_bits1(&s->gb))) |
| 1194 | transmit_coeffs = 1; |
| 1195 | } |
| 1196 | |
| 1197 | av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE); |
| 1198 | if (transmit_coeffs) { |
| 1199 | int step; |
| 1200 | int quant_step = 90 * s->bits_per_sample >> 4; |
| 1201 | |
| 1202 | /** decode number of vector coded coefficients */ |
| 1203 | if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) { |
| 1204 | int num_bits = av_log2((s->subframe_len + 3)/4) + 1; |
| 1205 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1206 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1207 | int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2; |
| 1208 | if (num_vec_coeffs > s->subframe_len) { |
| 1209 | av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs); |
| 1210 | return AVERROR_INVALIDDATA; |
| 1211 | } |
| 1212 | av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out)); |
| 1213 | s->channel[c].num_vec_coeffs = num_vec_coeffs; |
| 1214 | } |
| 1215 | } else { |
| 1216 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1217 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1218 | s->channel[c].num_vec_coeffs = s->subframe_len; |
| 1219 | } |
| 1220 | } |
| 1221 | /** decode quantization step */ |
| 1222 | step = get_sbits(&s->gb, 6); |
| 1223 | quant_step += step; |
| 1224 | if (step == -32 || step == 31) { |
| 1225 | const int sign = (step == 31) - 1; |
| 1226 | int quant = 0; |
| 1227 | while (get_bits_count(&s->gb) + 5 < s->num_saved_bits && |
| 1228 | (step = get_bits(&s->gb, 5)) == 31) { |
| 1229 | quant += 31; |
| 1230 | } |
| 1231 | quant_step += ((quant + step) ^ sign) - sign; |
| 1232 | } |
| 1233 | if (quant_step < 0) { |
| 1234 | av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n"); |
| 1235 | } |
| 1236 | |
| 1237 | /** decode quantization step modifiers for every channel */ |
| 1238 | |
| 1239 | if (s->channels_for_cur_subframe == 1) { |
| 1240 | s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step; |
| 1241 | } else { |
| 1242 | int modifier_len = get_bits(&s->gb, 3); |
| 1243 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1244 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1245 | s->channel[c].quant_step = quant_step; |
| 1246 | if (get_bits1(&s->gb)) { |
| 1247 | if (modifier_len) { |
| 1248 | s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1; |
| 1249 | } else |
| 1250 | ++s->channel[c].quant_step; |
| 1251 | } |
| 1252 | } |
| 1253 | } |
| 1254 | |
| 1255 | /** decode scale factors */ |
| 1256 | if (decode_scale_factors(s) < 0) |
| 1257 | return AVERROR_INVALIDDATA; |
| 1258 | } |
| 1259 | |
| 1260 | av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n", |
| 1261 | get_bits_count(&s->gb) - s->subframe_offset); |
| 1262 | |
| 1263 | /** parse coefficients */ |
| 1264 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1265 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1266 | if (s->channel[c].transmit_coefs && |
| 1267 | get_bits_count(&s->gb) < s->num_saved_bits) { |
| 1268 | decode_coeffs(s, c); |
| 1269 | } else |
| 1270 | memset(s->channel[c].coeffs, 0, |
| 1271 | sizeof(*s->channel[c].coeffs) * subframe_len); |
| 1272 | } |
| 1273 | |
| 1274 | av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n", |
| 1275 | get_bits_count(&s->gb) - s->subframe_offset); |
| 1276 | |
| 1277 | if (transmit_coeffs) { |
| 1278 | FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS]; |
| 1279 | /** reconstruct the per channel data */ |
| 1280 | inverse_channel_transform(s); |
| 1281 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1282 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1283 | const int* sf = s->channel[c].scale_factors; |
| 1284 | int b; |
| 1285 | |
| 1286 | if (c == s->lfe_channel) |
| 1287 | memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) * |
| 1288 | (subframe_len - cur_subwoofer_cutoff)); |
| 1289 | |
| 1290 | /** inverse quantization and rescaling */ |
| 1291 | for (b = 0; b < s->num_bands; b++) { |
| 1292 | const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len); |
| 1293 | const int exp = s->channel[c].quant_step - |
| 1294 | (s->channel[c].max_scale_factor - *sf++) * |
| 1295 | s->channel[c].scale_factor_step; |
| 1296 | const float quant = pow(10.0, exp / 20.0); |
| 1297 | int start = s->cur_sfb_offsets[b]; |
| 1298 | s->fdsp->vector_fmul_scalar(s->tmp + start, |
| 1299 | s->channel[c].coeffs + start, |
| 1300 | quant, end - start); |
| 1301 | } |
| 1302 | |
| 1303 | /** apply imdct (imdct_half == DCTIV with reverse) */ |
| 1304 | mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp); |
| 1305 | } |
| 1306 | } |
| 1307 | |
| 1308 | /** window and overlapp-add */ |
| 1309 | wmapro_window(s); |
| 1310 | |
| 1311 | /** handled one subframe */ |
| 1312 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
| 1313 | int c = s->channel_indexes_for_cur_subframe[i]; |
| 1314 | if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) { |
| 1315 | av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n"); |
| 1316 | return AVERROR_INVALIDDATA; |
| 1317 | } |
| 1318 | ++s->channel[c].cur_subframe; |
| 1319 | } |
| 1320 | |
| 1321 | return 0; |
| 1322 | } |
| 1323 | |
| 1324 | /** |
| 1325 | *@brief Decode one WMA frame. |
| 1326 | *@param s codec context |
| 1327 | *@return 0 if the trailer bit indicates that this is the last frame, |
| 1328 | * 1 if there are additional frames |
| 1329 | */ |
| 1330 | static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr) |
| 1331 | { |
| 1332 | AVCodecContext *avctx = s->avctx; |
| 1333 | GetBitContext* gb = &s->gb; |
| 1334 | int more_frames = 0; |
| 1335 | int len = 0; |
| 1336 | int i, ret; |
| 1337 | |
| 1338 | /** get frame length */ |
| 1339 | if (s->len_prefix) |
| 1340 | len = get_bits(gb, s->log2_frame_size); |
| 1341 | |
| 1342 | av_dlog(s->avctx, "decoding frame with length %x\n", len); |
| 1343 | |
| 1344 | /** decode tile information */ |
| 1345 | if (decode_tilehdr(s)) { |
| 1346 | s->packet_loss = 1; |
| 1347 | return 0; |
| 1348 | } |
| 1349 | |
| 1350 | /** read postproc transform */ |
| 1351 | if (s->avctx->channels > 1 && get_bits1(gb)) { |
| 1352 | if (get_bits1(gb)) { |
| 1353 | for (i = 0; i < avctx->channels * avctx->channels; i++) |
| 1354 | skip_bits(gb, 4); |
| 1355 | } |
| 1356 | } |
| 1357 | |
| 1358 | /** read drc info */ |
| 1359 | if (s->dynamic_range_compression) { |
| 1360 | s->drc_gain = get_bits(gb, 8); |
| 1361 | av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain); |
| 1362 | } |
| 1363 | |
| 1364 | /** no idea what these are for, might be the number of samples |
| 1365 | that need to be skipped at the beginning or end of a stream */ |
| 1366 | if (get_bits1(gb)) { |
| 1367 | int av_unused skip; |
| 1368 | |
| 1369 | /** usually true for the first frame */ |
| 1370 | if (get_bits1(gb)) { |
| 1371 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); |
| 1372 | av_dlog(s->avctx, "start skip: %i\n", skip); |
| 1373 | } |
| 1374 | |
| 1375 | /** sometimes true for the last frame */ |
| 1376 | if (get_bits1(gb)) { |
| 1377 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); |
| 1378 | av_dlog(s->avctx, "end skip: %i\n", skip); |
| 1379 | } |
| 1380 | |
| 1381 | } |
| 1382 | |
| 1383 | av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n", |
| 1384 | get_bits_count(gb) - s->frame_offset); |
| 1385 | |
| 1386 | /** reset subframe states */ |
| 1387 | s->parsed_all_subframes = 0; |
| 1388 | for (i = 0; i < avctx->channels; i++) { |
| 1389 | s->channel[i].decoded_samples = 0; |
| 1390 | s->channel[i].cur_subframe = 0; |
| 1391 | s->channel[i].reuse_sf = 0; |
| 1392 | } |
| 1393 | |
| 1394 | /** decode all subframes */ |
| 1395 | while (!s->parsed_all_subframes) { |
| 1396 | if (decode_subframe(s) < 0) { |
| 1397 | s->packet_loss = 1; |
| 1398 | return 0; |
| 1399 | } |
| 1400 | } |
| 1401 | |
| 1402 | /* get output buffer */ |
| 1403 | frame->nb_samples = s->samples_per_frame; |
| 1404 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) { |
| 1405 | s->packet_loss = 1; |
| 1406 | return 0; |
| 1407 | } |
| 1408 | |
| 1409 | /** copy samples to the output buffer */ |
| 1410 | for (i = 0; i < avctx->channels; i++) |
| 1411 | memcpy(frame->extended_data[i], s->channel[i].out, |
| 1412 | s->samples_per_frame * sizeof(*s->channel[i].out)); |
| 1413 | |
| 1414 | for (i = 0; i < avctx->channels; i++) { |
| 1415 | /** reuse second half of the IMDCT output for the next frame */ |
| 1416 | memcpy(&s->channel[i].out[0], |
| 1417 | &s->channel[i].out[s->samples_per_frame], |
| 1418 | s->samples_per_frame * sizeof(*s->channel[i].out) >> 1); |
| 1419 | } |
| 1420 | |
| 1421 | if (s->skip_frame) { |
| 1422 | s->skip_frame = 0; |
| 1423 | *got_frame_ptr = 0; |
| 1424 | av_frame_unref(frame); |
| 1425 | } else { |
| 1426 | *got_frame_ptr = 1; |
| 1427 | } |
| 1428 | |
| 1429 | if (s->len_prefix) { |
| 1430 | if (len != (get_bits_count(gb) - s->frame_offset) + 2) { |
| 1431 | /** FIXME: not sure if this is always an error */ |
| 1432 | av_log(s->avctx, AV_LOG_ERROR, |
| 1433 | "frame[%"PRIu32"] would have to skip %i bits\n", |
| 1434 | s->frame_num, |
| 1435 | len - (get_bits_count(gb) - s->frame_offset) - 1); |
| 1436 | s->packet_loss = 1; |
| 1437 | return 0; |
| 1438 | } |
| 1439 | |
| 1440 | /** skip the rest of the frame data */ |
| 1441 | skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1); |
| 1442 | } else { |
| 1443 | while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) { |
| 1444 | } |
| 1445 | } |
| 1446 | |
| 1447 | /** decode trailer bit */ |
| 1448 | more_frames = get_bits1(gb); |
| 1449 | |
| 1450 | ++s->frame_num; |
| 1451 | return more_frames; |
| 1452 | } |
| 1453 | |
| 1454 | /** |
| 1455 | *@brief Calculate remaining input buffer length. |
| 1456 | *@param s codec context |
| 1457 | *@param gb bitstream reader context |
| 1458 | *@return remaining size in bits |
| 1459 | */ |
| 1460 | static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb) |
| 1461 | { |
| 1462 | return s->buf_bit_size - get_bits_count(gb); |
| 1463 | } |
| 1464 | |
| 1465 | /** |
| 1466 | *@brief Fill the bit reservoir with a (partial) frame. |
| 1467 | *@param s codec context |
| 1468 | *@param gb bitstream reader context |
| 1469 | *@param len length of the partial frame |
| 1470 | *@param append decides whether to reset the buffer or not |
| 1471 | */ |
| 1472 | static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len, |
| 1473 | int append) |
| 1474 | { |
| 1475 | int buflen; |
| 1476 | |
| 1477 | /** when the frame data does not need to be concatenated, the input buffer |
| 1478 | is reset and additional bits from the previous frame are copied |
| 1479 | and skipped later so that a fast byte copy is possible */ |
| 1480 | |
| 1481 | if (!append) { |
| 1482 | s->frame_offset = get_bits_count(gb) & 7; |
| 1483 | s->num_saved_bits = s->frame_offset; |
| 1484 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
| 1485 | } |
| 1486 | |
| 1487 | buflen = (put_bits_count(&s->pb) + len + 8) >> 3; |
| 1488 | |
| 1489 | if (len <= 0 || buflen > MAX_FRAMESIZE) { |
| 1490 | avpriv_request_sample(s->avctx, "Too small input buffer"); |
| 1491 | s->packet_loss = 1; |
| 1492 | return; |
| 1493 | } |
| 1494 | |
| 1495 | av_assert0(len <= put_bits_left(&s->pb)); |
| 1496 | |
| 1497 | s->num_saved_bits += len; |
| 1498 | if (!append) { |
| 1499 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), |
| 1500 | s->num_saved_bits); |
| 1501 | } else { |
| 1502 | int align = 8 - (get_bits_count(gb) & 7); |
| 1503 | align = FFMIN(align, len); |
| 1504 | put_bits(&s->pb, align, get_bits(gb, align)); |
| 1505 | len -= align; |
| 1506 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len); |
| 1507 | } |
| 1508 | skip_bits_long(gb, len); |
| 1509 | |
| 1510 | { |
| 1511 | PutBitContext tmp = s->pb; |
| 1512 | flush_put_bits(&tmp); |
| 1513 | } |
| 1514 | |
| 1515 | init_get_bits(&s->gb, s->frame_data, s->num_saved_bits); |
| 1516 | skip_bits(&s->gb, s->frame_offset); |
| 1517 | } |
| 1518 | |
| 1519 | /** |
| 1520 | *@brief Decode a single WMA packet. |
| 1521 | *@param avctx codec context |
| 1522 | *@param data the output buffer |
| 1523 | *@param avpkt input packet |
| 1524 | *@return number of bytes that were read from the input buffer |
| 1525 | */ |
| 1526 | static int decode_packet(AVCodecContext *avctx, void *data, |
| 1527 | int *got_frame_ptr, AVPacket* avpkt) |
| 1528 | { |
| 1529 | WMAProDecodeCtx *s = avctx->priv_data; |
| 1530 | GetBitContext* gb = &s->pgb; |
| 1531 | const uint8_t* buf = avpkt->data; |
| 1532 | int buf_size = avpkt->size; |
| 1533 | int num_bits_prev_frame; |
| 1534 | int packet_sequence_number; |
| 1535 | |
| 1536 | *got_frame_ptr = 0; |
| 1537 | |
| 1538 | if (s->packet_done || s->packet_loss) { |
| 1539 | s->packet_done = 0; |
| 1540 | |
| 1541 | /** sanity check for the buffer length */ |
| 1542 | if (buf_size < avctx->block_align) { |
| 1543 | av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n", |
| 1544 | buf_size, avctx->block_align); |
| 1545 | return AVERROR_INVALIDDATA; |
| 1546 | } |
| 1547 | |
| 1548 | s->next_packet_start = buf_size - avctx->block_align; |
| 1549 | buf_size = avctx->block_align; |
| 1550 | s->buf_bit_size = buf_size << 3; |
| 1551 | |
| 1552 | /** parse packet header */ |
| 1553 | init_get_bits(gb, buf, s->buf_bit_size); |
| 1554 | packet_sequence_number = get_bits(gb, 4); |
| 1555 | skip_bits(gb, 2); |
| 1556 | |
| 1557 | /** get number of bits that need to be added to the previous frame */ |
| 1558 | num_bits_prev_frame = get_bits(gb, s->log2_frame_size); |
| 1559 | av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number, |
| 1560 | num_bits_prev_frame); |
| 1561 | |
| 1562 | /** check for packet loss */ |
| 1563 | if (!s->packet_loss && |
| 1564 | ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) { |
| 1565 | s->packet_loss = 1; |
| 1566 | av_log(avctx, AV_LOG_ERROR, |
| 1567 | "Packet loss detected! seq %"PRIx8" vs %x\n", |
| 1568 | s->packet_sequence_number, packet_sequence_number); |
| 1569 | } |
| 1570 | s->packet_sequence_number = packet_sequence_number; |
| 1571 | |
| 1572 | if (num_bits_prev_frame > 0) { |
| 1573 | int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb); |
| 1574 | if (num_bits_prev_frame >= remaining_packet_bits) { |
| 1575 | num_bits_prev_frame = remaining_packet_bits; |
| 1576 | s->packet_done = 1; |
| 1577 | } |
| 1578 | |
| 1579 | /** append the previous frame data to the remaining data from the |
| 1580 | previous packet to create a full frame */ |
| 1581 | save_bits(s, gb, num_bits_prev_frame, 1); |
| 1582 | av_dlog(avctx, "accumulated %x bits of frame data\n", |
| 1583 | s->num_saved_bits - s->frame_offset); |
| 1584 | |
| 1585 | /** decode the cross packet frame if it is valid */ |
| 1586 | if (!s->packet_loss) |
| 1587 | decode_frame(s, data, got_frame_ptr); |
| 1588 | } else if (s->num_saved_bits - s->frame_offset) { |
| 1589 | av_dlog(avctx, "ignoring %x previously saved bits\n", |
| 1590 | s->num_saved_bits - s->frame_offset); |
| 1591 | } |
| 1592 | |
| 1593 | if (s->packet_loss) { |
| 1594 | /** reset number of saved bits so that the decoder |
| 1595 | does not start to decode incomplete frames in the |
| 1596 | s->len_prefix == 0 case */ |
| 1597 | s->num_saved_bits = 0; |
| 1598 | s->packet_loss = 0; |
| 1599 | } |
| 1600 | |
| 1601 | } else { |
| 1602 | int frame_size; |
| 1603 | s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3; |
| 1604 | init_get_bits(gb, avpkt->data, s->buf_bit_size); |
| 1605 | skip_bits(gb, s->packet_offset); |
| 1606 | if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size && |
| 1607 | (frame_size = show_bits(gb, s->log2_frame_size)) && |
| 1608 | frame_size <= remaining_bits(s, gb)) { |
| 1609 | save_bits(s, gb, frame_size, 0); |
| 1610 | if (!s->packet_loss) |
| 1611 | s->packet_done = !decode_frame(s, data, got_frame_ptr); |
| 1612 | } else if (!s->len_prefix |
| 1613 | && s->num_saved_bits > get_bits_count(&s->gb)) { |
| 1614 | /** when the frames do not have a length prefix, we don't know |
| 1615 | the compressed length of the individual frames |
| 1616 | however, we know what part of a new packet belongs to the |
| 1617 | previous frame |
| 1618 | therefore we save the incoming packet first, then we append |
| 1619 | the "previous frame" data from the next packet so that |
| 1620 | we get a buffer that only contains full frames */ |
| 1621 | s->packet_done = !decode_frame(s, data, got_frame_ptr); |
| 1622 | } else |
| 1623 | s->packet_done = 1; |
| 1624 | } |
| 1625 | |
| 1626 | if (s->packet_done && !s->packet_loss && |
| 1627 | remaining_bits(s, gb) > 0) { |
| 1628 | /** save the rest of the data so that it can be decoded |
| 1629 | with the next packet */ |
| 1630 | save_bits(s, gb, remaining_bits(s, gb), 0); |
| 1631 | } |
| 1632 | |
| 1633 | s->packet_offset = get_bits_count(gb) & 7; |
| 1634 | if (s->packet_loss) |
| 1635 | return AVERROR_INVALIDDATA; |
| 1636 | |
| 1637 | return get_bits_count(gb) >> 3; |
| 1638 | } |
| 1639 | |
| 1640 | /** |
| 1641 | *@brief Clear decoder buffers (for seeking). |
| 1642 | *@param avctx codec context |
| 1643 | */ |
| 1644 | static void flush(AVCodecContext *avctx) |
| 1645 | { |
| 1646 | WMAProDecodeCtx *s = avctx->priv_data; |
| 1647 | int i; |
| 1648 | /** reset output buffer as a part of it is used during the windowing of a |
| 1649 | new frame */ |
| 1650 | for (i = 0; i < avctx->channels; i++) |
| 1651 | memset(s->channel[i].out, 0, s->samples_per_frame * |
| 1652 | sizeof(*s->channel[i].out)); |
| 1653 | s->packet_loss = 1; |
| 1654 | } |
| 1655 | |
| 1656 | |
| 1657 | /** |
| 1658 | *@brief wmapro decoder |
| 1659 | */ |
| 1660 | AVCodec ff_wmapro_decoder = { |
| 1661 | .name = "wmapro", |
| 1662 | .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"), |
| 1663 | .type = AVMEDIA_TYPE_AUDIO, |
| 1664 | .id = AV_CODEC_ID_WMAPRO, |
| 1665 | .priv_data_size = sizeof(WMAProDecodeCtx), |
| 1666 | .init = decode_init, |
| 1667 | .close = decode_end, |
| 1668 | .decode = decode_packet, |
| 1669 | .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1, |
| 1670 | .flush = flush, |
| 1671 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
| 1672 | AV_SAMPLE_FMT_NONE }, |
| 1673 | }; |