2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
6 * This file is part of FFmpeg.
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.
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.
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
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
32 * - windowing and overlapp-add
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
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
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.
47 * Example wmapro bitstream (in samples):
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 * ---------------------------------------------------
58 * The frame layouts for the individual channels of a wma frame does not need
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.
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
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
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.
91 #include "libavutil/ffmath.h"
92 #include "libavutil/float_dsp.h"
93 #include "libavutil/intfloat.h"
94 #include "libavutil/intreadwrite.h"
99 #include "wmaprodata.h"
102 #include "wma_common.h"
104 /** current decoder limitations */
105 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
106 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
107 #define MAX_BANDS 29 ///< max number of scale factor bands
108 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
110 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
111 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
112 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
113 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
114 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
118 #define SCALEVLCBITS 8
119 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
120 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
121 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
122 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
123 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
125 static VLC sf_vlc; ///< scale factor DPCM vlc
126 static VLC sf_rl_vlc; ///< scale factor run length vlc
127 static VLC vec4_vlc; ///< 4 coefficients per symbol
128 static VLC vec2_vlc; ///< 2 coefficients per symbol
129 static VLC vec1_vlc; ///< 1 coefficient per symbol
130 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
131 static float sin64[33]; ///< sine table for decorrelation
134 * @brief frame specific decoder context for a single channel
136 typedef struct WMAProChannelCtx {
137 int16_t prev_block_len; ///< length of the previous block
138 uint8_t transmit_coefs;
139 uint8_t num_subframes;
140 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
141 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
142 uint8_t cur_subframe; ///< current subframe number
143 uint16_t decoded_samples; ///< number of already processed samples
144 uint8_t grouped; ///< channel is part of a group
145 int quant_step; ///< quantization step for the current subframe
146 int8_t reuse_sf; ///< share scale factors between subframes
147 int8_t scale_factor_step; ///< scaling step for the current subframe
148 int max_scale_factor; ///< maximum scale factor for the current subframe
149 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
150 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
151 int* scale_factors; ///< pointer to the scale factor values used for decoding
152 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
153 float* coeffs; ///< pointer to the subframe decode buffer
154 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
155 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
159 * @brief channel group for channel transformations
161 typedef struct WMAProChannelGrp {
162 uint8_t num_channels; ///< number of channels in the group
163 int8_t transform; ///< transform on / off
164 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
165 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
166 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
170 * @brief main decoder context
172 typedef struct WMAProDecodeCtx {
173 /* generic decoder variables */
174 AVCodecContext* avctx; ///< codec context for av_log
175 AVFloatDSPContext *fdsp;
176 uint8_t frame_data[MAX_FRAMESIZE +
177 AV_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
178 PutBitContext pb; ///< context for filling the frame_data buffer
179 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
180 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
181 const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
183 /* frame size dependent frame information (set during initialization) */
184 uint32_t decode_flags; ///< used compression features
185 uint8_t len_prefix; ///< frame is prefixed with its length
186 uint8_t dynamic_range_compression; ///< frame contains DRC data
187 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
188 uint16_t samples_per_frame; ///< number of samples to output
189 uint16_t log2_frame_size;
190 int8_t lfe_channel; ///< lfe channel index
191 uint8_t max_num_subframes;
192 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
193 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
194 uint16_t min_samples_per_subframe;
195 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
196 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
197 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
198 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
200 /* packet decode state */
201 GetBitContext pgb; ///< bitstream reader context for the packet
202 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
203 uint8_t packet_offset; ///< frame offset in the packet
204 uint8_t packet_sequence_number; ///< current packet number
205 int num_saved_bits; ///< saved number of bits
206 int frame_offset; ///< frame offset in the bit reservoir
207 int subframe_offset; ///< subframe offset in the bit reservoir
208 uint8_t packet_loss; ///< set in case of bitstream error
209 uint8_t packet_done; ///< set when a packet is fully decoded
210 uint8_t skip_packets;
212 /* frame decode state */
213 uint32_t frame_num; ///< current frame number (not used for decoding)
215 GetBitContext gb; ///< bitstream reader context
216 int buf_bit_size; ///< buffer size in bits
217 uint8_t drc_gain; ///< gain for the DRC tool
218 int8_t skip_frame; ///< skip output step
219 int8_t parsed_all_subframes; ///< all subframes decoded?
221 /* subframe/block decode state */
222 int16_t subframe_len; ///< current subframe length
223 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
224 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
225 int8_t num_bands; ///< number of scale factor bands
226 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
227 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
228 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
229 int8_t esc_len; ///< length of escaped coefficients
231 uint8_t num_chgroups; ///< number of channel groups
232 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
234 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
239 *@brief helper function to print the most important members of the context
242 static av_cold void dump_context(WMAProDecodeCtx *s)
244 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
245 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
247 PRINT("ed sample bit depth", s->bits_per_sample);
248 PRINT_HEX("ed decode flags", s->decode_flags);
249 PRINT("samples per frame", s->samples_per_frame);
250 PRINT("log2 frame size", s->log2_frame_size);
251 PRINT("max num subframes", s->max_num_subframes);
252 PRINT("len prefix", s->len_prefix);
253 PRINT("num channels", s->avctx->channels);
257 *@brief Uninitialize the decoder and free all resources.
258 *@param avctx codec context
259 *@return 0 on success, < 0 otherwise
261 static av_cold int decode_end(AVCodecContext *avctx)
263 WMAProDecodeCtx *s = avctx->priv_data;
268 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
269 ff_mdct_end(&s->mdct_ctx[i]);
274 static av_cold int get_rate(AVCodecContext *avctx)
276 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA?
277 if (avctx->sample_rate > 44100)
279 else if (avctx->sample_rate > 32000)
281 else if (avctx->sample_rate > 24000)
286 return avctx->sample_rate;
290 *@brief Initialize the decoder.
291 *@param avctx codec context
292 *@return 0 on success, -1 otherwise
294 static av_cold int decode_init(AVCodecContext *avctx)
296 WMAProDecodeCtx *s = avctx->priv_data;
297 uint8_t *edata_ptr = avctx->extradata;
298 unsigned int channel_mask;
300 int log2_max_num_subframes;
301 int num_possible_block_sizes;
303 if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2)
304 avctx->block_align = 2048;
306 if (!avctx->block_align) {
307 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
308 return AVERROR(EINVAL);
312 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
314 return AVERROR(ENOMEM);
316 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
318 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
320 if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size >= 34) {
321 s->decode_flags = 0x10d6;
322 channel_mask = AV_RL32(edata_ptr+2);
323 s->bits_per_sample = 16;
324 /** dump the extradata */
325 for (i = 0; i < avctx->extradata_size; i++)
326 ff_dlog(avctx, "[%x] ", avctx->extradata[i]);
327 ff_dlog(avctx, "\n");
329 } else if (avctx->codec_id == AV_CODEC_ID_XMA1 && avctx->extradata_size >= 28) {
330 s->decode_flags = 0x10d6;
331 s->bits_per_sample = 16;
333 /** dump the extradata */
334 for (i = 0; i < avctx->extradata_size; i++)
335 ff_dlog(avctx, "[%x] ", avctx->extradata[i]);
336 ff_dlog(avctx, "\n");
338 } else if (avctx->extradata_size >= 18) {
339 s->decode_flags = AV_RL16(edata_ptr+14);
340 channel_mask = AV_RL32(edata_ptr+2);
341 s->bits_per_sample = AV_RL16(edata_ptr);
343 if (s->bits_per_sample > 32 || s->bits_per_sample < 1) {
344 avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample);
345 return AVERROR_PATCHWELCOME;
348 /** dump the extradata */
349 for (i = 0; i < avctx->extradata_size; i++)
350 ff_dlog(avctx, "[%x] ", avctx->extradata[i]);
351 ff_dlog(avctx, "\n");
354 avpriv_request_sample(avctx, "Unknown extradata size");
355 return AVERROR_PATCHWELCOME;
358 if (avctx->codec_id != AV_CODEC_ID_WMAPRO && avctx->channels > 2) {
359 avpriv_report_missing_feature(avctx, ">2 channels support");
360 return AVERROR_PATCHWELCOME;
364 s->log2_frame_size = av_log2(avctx->block_align) + 4;
365 if (s->log2_frame_size > 25) {
366 avpriv_request_sample(avctx, "Large block align");
367 return AVERROR_PATCHWELCOME;
371 if (avctx->codec_id != AV_CODEC_ID_WMAPRO)
374 s->skip_frame = 1; /* skip first frame */
377 s->len_prefix = (s->decode_flags & 0x40);
380 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
381 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
382 if (bits > WMAPRO_BLOCK_MAX_BITS) {
383 avpriv_request_sample(avctx, "14-bit block sizes");
384 return AVERROR_PATCHWELCOME;
386 s->samples_per_frame = 1 << bits;
388 s->samples_per_frame = 512;
392 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
393 s->max_num_subframes = 1 << log2_max_num_subframes;
394 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
395 s->max_subframe_len_bit = 1;
396 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
398 num_possible_block_sizes = log2_max_num_subframes + 1;
399 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
400 s->dynamic_range_compression = (s->decode_flags & 0x80);
402 if (s->max_num_subframes > MAX_SUBFRAMES) {
403 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
404 s->max_num_subframes);
405 return AVERROR_INVALIDDATA;
408 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
409 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
410 s->min_samples_per_subframe);
411 return AVERROR_INVALIDDATA;
414 if (s->avctx->sample_rate <= 0) {
415 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
416 return AVERROR_INVALIDDATA;
419 if (avctx->channels < 0) {
420 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
422 return AVERROR_INVALIDDATA;
423 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
424 avpriv_request_sample(avctx,
425 "More than %d channels", WMAPRO_MAX_CHANNELS);
426 return AVERROR_PATCHWELCOME;
429 /** init previous block len */
430 for (i = 0; i < avctx->channels; i++)
431 s->channel[i].prev_block_len = s->samples_per_frame;
433 /** extract lfe channel position */
436 if (channel_mask & 8) {
438 for (mask = 1; mask < 16; mask <<= 1) {
439 if (channel_mask & mask)
444 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
445 scale_huffbits, 1, 1,
446 scale_huffcodes, 2, 2, 616);
448 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
449 scale_rl_huffbits, 1, 1,
450 scale_rl_huffcodes, 4, 4, 1406);
452 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
453 coef0_huffbits, 1, 1,
454 coef0_huffcodes, 4, 4, 2108);
456 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
457 coef1_huffbits, 1, 1,
458 coef1_huffcodes, 4, 4, 3912);
460 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
462 vec4_huffcodes, 2, 2, 604);
464 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
466 vec2_huffcodes, 2, 2, 562);
468 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
470 vec1_huffcodes, 2, 2, 562);
472 /** calculate number of scale factor bands and their offsets
473 for every possible block size */
474 for (i = 0; i < num_possible_block_sizes; i++) {
475 int subframe_len = s->samples_per_frame >> i;
478 int rate = get_rate(avctx);
480 s->sfb_offsets[i][0] = 0;
482 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
483 int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2;
485 if (offset > s->sfb_offsets[i][band - 1])
486 s->sfb_offsets[i][band++] = offset;
488 if (offset >= subframe_len)
491 s->sfb_offsets[i][band - 1] = subframe_len;
492 s->num_sfb[i] = band - 1;
493 if (s->num_sfb[i] <= 0) {
494 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
495 return AVERROR_INVALIDDATA;
500 /** Scale factors can be shared between blocks of different size
501 as every block has a different scale factor band layout.
502 The matrix sf_offsets is needed to find the correct scale factor.
505 for (i = 0; i < num_possible_block_sizes; i++) {
507 for (b = 0; b < s->num_sfb[i]; b++) {
509 int offset = ((s->sfb_offsets[i][b]
510 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
511 for (x = 0; x < num_possible_block_sizes; x++) {
513 while (s->sfb_offsets[x][v + 1] << x < offset) {
515 av_assert0(v < MAX_BANDS);
517 s->sf_offsets[i][x][b] = v;
522 /** init MDCT, FIXME: only init needed sizes */
523 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
524 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
525 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
526 / (1 << (s->bits_per_sample - 1)));
528 /** init MDCT windows: simple sine window */
529 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
530 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
531 ff_init_ff_sine_windows(win_idx);
532 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
535 /** calculate subwoofer cutoff values */
536 for (i = 0; i < num_possible_block_sizes; i++) {
537 int block_size = s->samples_per_frame >> i;
538 int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1)
539 / s->avctx->sample_rate;
540 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
543 /** calculate sine values for the decorrelation matrix */
544 for (i = 0; i < 33; i++)
545 sin64[i] = sin(i*M_PI / 64.0);
547 if (avctx->debug & FF_DEBUG_BITSTREAM)
550 avctx->channel_layout = channel_mask;
556 *@brief Decode the subframe length.
558 *@param offset sample offset in the frame
559 *@return decoded subframe length on success, < 0 in case of an error
561 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
563 int frame_len_shift = 0;
566 /** no need to read from the bitstream when only one length is possible */
567 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
568 return s->min_samples_per_subframe;
570 if (get_bits_left(&s->gb) < 1)
571 return AVERROR_INVALIDDATA;
573 /** 1 bit indicates if the subframe is of maximum length */
574 if (s->max_subframe_len_bit) {
575 if (get_bits1(&s->gb))
576 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
578 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
580 subframe_len = s->samples_per_frame >> frame_len_shift;
582 /** sanity check the length */
583 if (subframe_len < s->min_samples_per_subframe ||
584 subframe_len > s->samples_per_frame) {
585 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
587 return AVERROR_INVALIDDATA;
593 *@brief Decode how the data in the frame is split into subframes.
594 * Every WMA frame contains the encoded data for a fixed number of
595 * samples per channel. The data for every channel might be split
596 * into several subframes. This function will reconstruct the list of
597 * subframes for every channel.
599 * If the subframes are not evenly split, the algorithm estimates the
600 * channels with the lowest number of total samples.
601 * Afterwards, for each of these channels a bit is read from the
602 * bitstream that indicates if the channel contains a subframe with the
603 * next subframe size that is going to be read from the bitstream or not.
604 * If a channel contains such a subframe, the subframe size gets added to
605 * the channel's subframe list.
606 * The algorithm repeats these steps until the frame is properly divided
607 * between the individual channels.
610 *@return 0 on success, < 0 in case of an error
612 static int decode_tilehdr(WMAProDecodeCtx *s)
614 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
615 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
616 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
617 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
618 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
621 /* Should never consume more than 3073 bits (256 iterations for the
622 * while loop when always the minimum amount of 128 samples is subtracted
623 * from missing samples in the 8 channel case).
624 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
627 /** reset tiling information */
628 for (c = 0; c < s->avctx->channels; c++)
629 s->channel[c].num_subframes = 0;
631 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
632 fixed_channel_layout = 1;
634 /** loop until the frame data is split between the subframes */
638 /** check which channels contain the subframe */
639 for (c = 0; c < s->avctx->channels; c++) {
640 if (num_samples[c] == min_channel_len) {
641 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
642 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
643 contains_subframe[c] = 1;
645 contains_subframe[c] = get_bits1(&s->gb);
647 contains_subframe[c] = 0;
650 /** get subframe length, subframe_len == 0 is not allowed */
651 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
652 return AVERROR_INVALIDDATA;
654 /** add subframes to the individual channels and find new min_channel_len */
655 min_channel_len += subframe_len;
656 for (c = 0; c < s->avctx->channels; c++) {
657 WMAProChannelCtx* chan = &s->channel[c];
659 if (contains_subframe[c]) {
660 if (chan->num_subframes >= MAX_SUBFRAMES) {
661 av_log(s->avctx, AV_LOG_ERROR,
662 "broken frame: num subframes > 31\n");
663 return AVERROR_INVALIDDATA;
665 chan->subframe_len[chan->num_subframes] = subframe_len;
666 num_samples[c] += subframe_len;
667 ++chan->num_subframes;
668 if (num_samples[c] > s->samples_per_frame) {
669 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
670 "channel len > samples_per_frame\n");
671 return AVERROR_INVALIDDATA;
673 } else if (num_samples[c] <= min_channel_len) {
674 if (num_samples[c] < min_channel_len) {
675 channels_for_cur_subframe = 0;
676 min_channel_len = num_samples[c];
678 ++channels_for_cur_subframe;
681 } while (min_channel_len < s->samples_per_frame);
683 for (c = 0; c < s->avctx->channels; c++) {
686 for (i = 0; i < s->channel[c].num_subframes; i++) {
687 ff_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
688 " len %i\n", s->frame_num, c, i,
689 s->channel[c].subframe_len[i]);
690 s->channel[c].subframe_offset[i] = offset;
691 offset += s->channel[c].subframe_len[i];
699 *@brief Calculate a decorrelation matrix from the bitstream parameters.
700 *@param s codec context
701 *@param chgroup channel group for which the matrix needs to be calculated
703 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
704 WMAProChannelGrp *chgroup)
708 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
709 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
710 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
712 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
713 rotation_offset[i] = get_bits(&s->gb, 6);
715 for (i = 0; i < chgroup->num_channels; i++)
716 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
717 get_bits1(&s->gb) ? 1.0 : -1.0;
719 for (i = 1; i < chgroup->num_channels; i++) {
721 for (x = 0; x < i; x++) {
723 for (y = 0; y < i + 1; y++) {
724 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
725 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
726 int n = rotation_offset[offset + x];
732 cosv = sin64[32 - n];
734 sinv = sin64[64 - n];
735 cosv = -sin64[n - 32];
738 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
739 (v1 * sinv) - (v2 * cosv);
740 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
741 (v1 * cosv) + (v2 * sinv);
749 *@brief Decode channel transformation parameters
750 *@param s codec context
751 *@return >= 0 in case of success, < 0 in case of bitstream errors
753 static int decode_channel_transform(WMAProDecodeCtx* s)
756 /* should never consume more than 1921 bits for the 8 channel case
757 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
758 * + MAX_CHANNELS + MAX_BANDS + 1)
761 /** in the one channel case channel transforms are pointless */
763 if (s->avctx->channels > 1) {
764 int remaining_channels = s->channels_for_cur_subframe;
766 if (get_bits1(&s->gb)) {
767 avpriv_request_sample(s->avctx,
768 "Channel transform bit");
769 return AVERROR_PATCHWELCOME;
772 for (s->num_chgroups = 0; remaining_channels &&
773 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
774 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
775 float** channel_data = chgroup->channel_data;
776 chgroup->num_channels = 0;
777 chgroup->transform = 0;
779 /** decode channel mask */
780 if (remaining_channels > 2) {
781 for (i = 0; i < s->channels_for_cur_subframe; i++) {
782 int channel_idx = s->channel_indexes_for_cur_subframe[i];
783 if (!s->channel[channel_idx].grouped
784 && get_bits1(&s->gb)) {
785 ++chgroup->num_channels;
786 s->channel[channel_idx].grouped = 1;
787 *channel_data++ = s->channel[channel_idx].coeffs;
791 chgroup->num_channels = remaining_channels;
792 for (i = 0; i < s->channels_for_cur_subframe; i++) {
793 int channel_idx = s->channel_indexes_for_cur_subframe[i];
794 if (!s->channel[channel_idx].grouped)
795 *channel_data++ = s->channel[channel_idx].coeffs;
796 s->channel[channel_idx].grouped = 1;
800 /** decode transform type */
801 if (chgroup->num_channels == 2) {
802 if (get_bits1(&s->gb)) {
803 if (get_bits1(&s->gb)) {
804 avpriv_request_sample(s->avctx,
805 "Unknown channel transform type");
806 return AVERROR_PATCHWELCOME;
809 chgroup->transform = 1;
810 if (s->avctx->channels == 2) {
811 chgroup->decorrelation_matrix[0] = 1.0;
812 chgroup->decorrelation_matrix[1] = -1.0;
813 chgroup->decorrelation_matrix[2] = 1.0;
814 chgroup->decorrelation_matrix[3] = 1.0;
817 chgroup->decorrelation_matrix[0] = 0.70703125;
818 chgroup->decorrelation_matrix[1] = -0.70703125;
819 chgroup->decorrelation_matrix[2] = 0.70703125;
820 chgroup->decorrelation_matrix[3] = 0.70703125;
823 } else if (chgroup->num_channels > 2) {
824 if (get_bits1(&s->gb)) {
825 chgroup->transform = 1;
826 if (get_bits1(&s->gb)) {
827 decode_decorrelation_matrix(s, chgroup);
829 /** FIXME: more than 6 coupled channels not supported */
830 if (chgroup->num_channels > 6) {
831 avpriv_request_sample(s->avctx,
832 "Coupled channels > 6");
834 memcpy(chgroup->decorrelation_matrix,
835 default_decorrelation[chgroup->num_channels],
836 chgroup->num_channels * chgroup->num_channels *
837 sizeof(*chgroup->decorrelation_matrix));
843 /** decode transform on / off */
844 if (chgroup->transform) {
845 if (!get_bits1(&s->gb)) {
847 /** transform can be enabled for individual bands */
848 for (i = 0; i < s->num_bands; i++) {
849 chgroup->transform_band[i] = get_bits1(&s->gb);
852 memset(chgroup->transform_band, 1, s->num_bands);
855 remaining_channels -= chgroup->num_channels;
862 *@brief Extract the coefficients from the bitstream.
863 *@param s codec context
864 *@param c current channel number
865 *@return 0 on success, < 0 in case of bitstream errors
867 static int decode_coeffs(WMAProDecodeCtx *s, int c)
869 /* Integers 0..15 as single-precision floats. The table saves a
870 costly int to float conversion, and storing the values as
871 integers allows fast sign-flipping. */
872 static const uint32_t fval_tab[16] = {
873 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
874 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
875 0x41000000, 0x41100000, 0x41200000, 0x41300000,
876 0x41400000, 0x41500000, 0x41600000, 0x41700000,
880 WMAProChannelCtx* ci = &s->channel[c];
887 ff_dlog(s->avctx, "decode coefficients for channel %i\n", c);
889 vlctable = get_bits1(&s->gb);
890 vlc = &coef_vlc[vlctable];
900 /** decode vector coefficients (consumes up to 167 bits per iteration for
901 4 vector coded large values) */
902 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
903 (cur_coeff + 3 < ci->num_vec_coeffs)) {
908 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
910 if (idx == HUFF_VEC4_SIZE - 1) {
911 for (i = 0; i < 4; i += 2) {
912 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
913 if (idx == HUFF_VEC2_SIZE - 1) {
915 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
916 if (v0 == HUFF_VEC1_SIZE - 1)
917 v0 += ff_wma_get_large_val(&s->gb);
918 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
919 if (v1 == HUFF_VEC1_SIZE - 1)
920 v1 += ff_wma_get_large_val(&s->gb);
921 vals[i ] = av_float2int(v0);
922 vals[i+1] = av_float2int(v1);
924 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
925 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
929 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
930 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
931 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
932 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
936 for (i = 0; i < 4; i++) {
938 uint32_t sign = get_bits1(&s->gb) - 1;
939 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
942 ci->coeffs[cur_coeff] = 0;
943 /** switch to run level mode when subframe_len / 128 zeros
944 were found in a row */
945 rl_mode |= (++num_zeros > s->subframe_len >> 8);
951 /** decode run level coded coefficients */
952 if (cur_coeff < s->subframe_len) {
953 memset(&ci->coeffs[cur_coeff], 0,
954 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
955 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
956 level, run, 1, ci->coeffs,
957 cur_coeff, s->subframe_len,
958 s->subframe_len, s->esc_len, 0))
959 return AVERROR_INVALIDDATA;
966 *@brief Extract scale factors from the bitstream.
967 *@param s codec context
968 *@return 0 on success, < 0 in case of bitstream errors
970 static int decode_scale_factors(WMAProDecodeCtx* s)
974 /** should never consume more than 5344 bits
975 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
978 for (i = 0; i < s->channels_for_cur_subframe; i++) {
979 int c = s->channel_indexes_for_cur_subframe[i];
982 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
983 sf_end = s->channel[c].scale_factors + s->num_bands;
985 /** resample scale factors for the new block size
986 * as the scale factors might need to be resampled several times
987 * before some new values are transmitted, a backup of the last
988 * transmitted scale factors is kept in saved_scale_factors
990 if (s->channel[c].reuse_sf) {
991 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
993 for (b = 0; b < s->num_bands; b++)
994 s->channel[c].scale_factors[b] =
995 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
998 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
1000 if (!s->channel[c].reuse_sf) {
1002 /** decode DPCM coded scale factors */
1003 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
1004 val = 45 / s->channel[c].scale_factor_step;
1005 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
1006 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
1011 /** run level decode differences to the resampled factors */
1012 for (i = 0; i < s->num_bands; i++) {
1018 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
1021 uint32_t code = get_bits(&s->gb, 14);
1023 sign = (code & 1) - 1;
1024 skip = (code & 0x3f) >> 1;
1025 } else if (idx == 1) {
1028 skip = scale_rl_run[idx];
1029 val = scale_rl_level[idx];
1030 sign = get_bits1(&s->gb)-1;
1034 if (i >= s->num_bands) {
1035 av_log(s->avctx, AV_LOG_ERROR,
1036 "invalid scale factor coding\n");
1037 return AVERROR_INVALIDDATA;
1039 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
1043 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
1044 s->channel[c].table_idx = s->table_idx;
1045 s->channel[c].reuse_sf = 1;
1048 /** calculate new scale factor maximum */
1049 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
1050 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
1051 s->channel[c].max_scale_factor =
1052 FFMAX(s->channel[c].max_scale_factor, *sf);
1060 *@brief Reconstruct the individual channel data.
1061 *@param s codec context
1063 static void inverse_channel_transform(WMAProDecodeCtx *s)
1067 for (i = 0; i < s->num_chgroups; i++) {
1068 if (s->chgroup[i].transform) {
1069 float data[WMAPRO_MAX_CHANNELS];
1070 const int num_channels = s->chgroup[i].num_channels;
1071 float** ch_data = s->chgroup[i].channel_data;
1072 float** ch_end = ch_data + num_channels;
1073 const int8_t* tb = s->chgroup[i].transform_band;
1076 /** multichannel decorrelation */
1077 for (sfb = s->cur_sfb_offsets;
1078 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1081 /** multiply values with the decorrelation_matrix */
1082 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1083 const float* mat = s->chgroup[i].decorrelation_matrix;
1084 const float* data_end = data + num_channels;
1085 float* data_ptr = data;
1088 for (ch = ch_data; ch < ch_end; ch++)
1089 *data_ptr++ = (*ch)[y];
1091 for (ch = ch_data; ch < ch_end; ch++) {
1094 while (data_ptr < data_end)
1095 sum += *data_ptr++ * *mat++;
1100 } else if (s->avctx->channels == 2) {
1101 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1102 s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0],
1103 ch_data[0] + sfb[0],
1105 s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0],
1106 ch_data[1] + sfb[0],
1115 *@brief Apply sine window and reconstruct the output buffer.
1116 *@param s codec context
1118 static void wmapro_window(WMAProDecodeCtx *s)
1121 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1122 int c = s->channel_indexes_for_cur_subframe[i];
1123 const float* window;
1124 int winlen = s->channel[c].prev_block_len;
1125 float* start = s->channel[c].coeffs - (winlen >> 1);
1127 if (s->subframe_len < winlen) {
1128 start += (winlen - s->subframe_len) >> 1;
1129 winlen = s->subframe_len;
1132 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1136 s->fdsp->vector_fmul_window(start, start, start + winlen,
1139 s->channel[c].prev_block_len = s->subframe_len;
1144 *@brief Decode a single subframe (block).
1145 *@param s codec context
1146 *@return 0 on success, < 0 when decoding failed
1148 static int decode_subframe(WMAProDecodeCtx *s)
1150 int offset = s->samples_per_frame;
1151 int subframe_len = s->samples_per_frame;
1153 int total_samples = s->samples_per_frame * s->avctx->channels;
1154 int transmit_coeffs = 0;
1155 int cur_subwoofer_cutoff;
1157 s->subframe_offset = get_bits_count(&s->gb);
1159 /** reset channel context and find the next block offset and size
1160 == the next block of the channel with the smallest number of
1163 for (i = 0; i < s->avctx->channels; i++) {
1164 s->channel[i].grouped = 0;
1165 if (offset > s->channel[i].decoded_samples) {
1166 offset = s->channel[i].decoded_samples;
1168 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1173 "processing subframe with offset %i len %i\n", offset, subframe_len);
1175 /** get a list of all channels that contain the estimated block */
1176 s->channels_for_cur_subframe = 0;
1177 for (i = 0; i < s->avctx->channels; i++) {
1178 const int cur_subframe = s->channel[i].cur_subframe;
1179 /** subtract already processed samples */
1180 total_samples -= s->channel[i].decoded_samples;
1182 /** and count if there are multiple subframes that match our profile */
1183 if (offset == s->channel[i].decoded_samples &&
1184 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1185 total_samples -= s->channel[i].subframe_len[cur_subframe];
1186 s->channel[i].decoded_samples +=
1187 s->channel[i].subframe_len[cur_subframe];
1188 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1189 ++s->channels_for_cur_subframe;
1193 /** check if the frame will be complete after processing the
1196 s->parsed_all_subframes = 1;
1199 ff_dlog(s->avctx, "subframe is part of %i channels\n",
1200 s->channels_for_cur_subframe);
1202 /** calculate number of scale factor bands and their offsets */
1203 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1204 s->num_bands = s->num_sfb[s->table_idx];
1205 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1206 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1208 /** configure the decoder for the current subframe */
1209 offset += s->samples_per_frame >> 1;
1211 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1212 int c = s->channel_indexes_for_cur_subframe[i];
1214 s->channel[c].coeffs = &s->channel[c].out[offset];
1217 s->subframe_len = subframe_len;
1218 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1220 /** skip extended header if any */
1221 if (get_bits1(&s->gb)) {
1223 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1224 int len = get_bits(&s->gb, 4);
1225 num_fill_bits = get_bitsz(&s->gb, len) + 1;
1228 if (num_fill_bits >= 0) {
1229 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1230 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1231 return AVERROR_INVALIDDATA;
1234 skip_bits_long(&s->gb, num_fill_bits);
1238 /** no idea for what the following bit is used */
1239 if (get_bits1(&s->gb)) {
1240 avpriv_request_sample(s->avctx, "Reserved bit");
1241 return AVERROR_PATCHWELCOME;
1245 if (decode_channel_transform(s) < 0)
1246 return AVERROR_INVALIDDATA;
1249 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1250 int c = s->channel_indexes_for_cur_subframe[i];
1251 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1252 transmit_coeffs = 1;
1255 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1256 if (transmit_coeffs) {
1258 int quant_step = 90 * s->bits_per_sample >> 4;
1260 /** decode number of vector coded coefficients */
1261 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1262 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1263 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1264 int c = s->channel_indexes_for_cur_subframe[i];
1265 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1266 if (num_vec_coeffs > s->subframe_len) {
1267 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1268 return AVERROR_INVALIDDATA;
1270 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1271 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1274 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1275 int c = s->channel_indexes_for_cur_subframe[i];
1276 s->channel[c].num_vec_coeffs = s->subframe_len;
1279 /** decode quantization step */
1280 step = get_sbits(&s->gb, 6);
1282 if (step == -32 || step == 31) {
1283 const int sign = (step == 31) - 1;
1285 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1286 (step = get_bits(&s->gb, 5)) == 31) {
1289 quant_step += ((quant + step) ^ sign) - sign;
1291 if (quant_step < 0) {
1292 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1295 /** decode quantization step modifiers for every channel */
1297 if (s->channels_for_cur_subframe == 1) {
1298 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1300 int modifier_len = get_bits(&s->gb, 3);
1301 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1302 int c = s->channel_indexes_for_cur_subframe[i];
1303 s->channel[c].quant_step = quant_step;
1304 if (get_bits1(&s->gb)) {
1306 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1308 ++s->channel[c].quant_step;
1313 /** decode scale factors */
1314 if (decode_scale_factors(s) < 0)
1315 return AVERROR_INVALIDDATA;
1318 ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1319 get_bits_count(&s->gb) - s->subframe_offset);
1321 /** parse coefficients */
1322 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1323 int c = s->channel_indexes_for_cur_subframe[i];
1324 if (s->channel[c].transmit_coefs &&
1325 get_bits_count(&s->gb) < s->num_saved_bits) {
1326 decode_coeffs(s, c);
1328 memset(s->channel[c].coeffs, 0,
1329 sizeof(*s->channel[c].coeffs) * subframe_len);
1332 ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1333 get_bits_count(&s->gb) - s->subframe_offset);
1335 if (transmit_coeffs) {
1336 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1337 /** reconstruct the per channel data */
1338 inverse_channel_transform(s);
1339 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1340 int c = s->channel_indexes_for_cur_subframe[i];
1341 const int* sf = s->channel[c].scale_factors;
1344 if (c == s->lfe_channel)
1345 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1346 (subframe_len - cur_subwoofer_cutoff));
1348 /** inverse quantization and rescaling */
1349 for (b = 0; b < s->num_bands; b++) {
1350 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1351 const int exp = s->channel[c].quant_step -
1352 (s->channel[c].max_scale_factor - *sf++) *
1353 s->channel[c].scale_factor_step;
1354 const float quant = ff_exp10(exp / 20.0);
1355 int start = s->cur_sfb_offsets[b];
1356 s->fdsp->vector_fmul_scalar(s->tmp + start,
1357 s->channel[c].coeffs + start,
1358 quant, end - start);
1361 /** apply imdct (imdct_half == DCTIV with reverse) */
1362 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1366 /** window and overlapp-add */
1369 /** handled one subframe */
1370 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1371 int c = s->channel_indexes_for_cur_subframe[i];
1372 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1373 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1374 return AVERROR_INVALIDDATA;
1376 ++s->channel[c].cur_subframe;
1383 *@brief Decode one WMA frame.
1384 *@param s codec context
1385 *@return 0 if the trailer bit indicates that this is the last frame,
1386 * 1 if there are additional frames
1388 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1390 AVCodecContext *avctx = s->avctx;
1391 GetBitContext* gb = &s->gb;
1392 int more_frames = 0;
1396 /** get frame length */
1398 len = get_bits(gb, s->log2_frame_size);
1400 ff_dlog(s->avctx, "decoding frame with length %x\n", len);
1402 /** decode tile information */
1403 if (decode_tilehdr(s)) {
1408 /** read postproc transform */
1409 if (s->avctx->channels > 1 && get_bits1(gb)) {
1410 if (get_bits1(gb)) {
1411 for (i = 0; i < avctx->channels * avctx->channels; i++)
1416 /** read drc info */
1417 if (s->dynamic_range_compression) {
1418 s->drc_gain = get_bits(gb, 8);
1419 ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1422 /** no idea what these are for, might be the number of samples
1423 that need to be skipped at the beginning or end of a stream */
1424 if (get_bits1(gb)) {
1427 /** usually true for the first frame */
1428 if (get_bits1(gb)) {
1429 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1430 ff_dlog(s->avctx, "start skip: %i\n", skip);
1433 /** sometimes true for the last frame */
1434 if (get_bits1(gb)) {
1435 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1436 ff_dlog(s->avctx, "end skip: %i\n", skip);
1441 ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1442 get_bits_count(gb) - s->frame_offset);
1444 /** reset subframe states */
1445 s->parsed_all_subframes = 0;
1446 for (i = 0; i < avctx->channels; i++) {
1447 s->channel[i].decoded_samples = 0;
1448 s->channel[i].cur_subframe = 0;
1449 s->channel[i].reuse_sf = 0;
1452 /** decode all subframes */
1453 while (!s->parsed_all_subframes) {
1454 if (decode_subframe(s) < 0) {
1460 /* get output buffer */
1461 frame->nb_samples = s->samples_per_frame;
1462 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1467 /** copy samples to the output buffer */
1468 for (i = 0; i < avctx->channels; i++)
1469 memcpy(frame->extended_data[i], s->channel[i].out,
1470 s->samples_per_frame * sizeof(*s->channel[i].out));
1472 for (i = 0; i < avctx->channels; i++) {
1473 /** reuse second half of the IMDCT output for the next frame */
1474 memcpy(&s->channel[i].out[0],
1475 &s->channel[i].out[s->samples_per_frame],
1476 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1479 if (s->skip_frame) {
1482 av_frame_unref(frame);
1487 if (s->len_prefix) {
1488 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1489 /** FIXME: not sure if this is always an error */
1490 av_log(s->avctx, AV_LOG_ERROR,
1491 "frame[%"PRIu32"] would have to skip %i bits\n",
1493 len - (get_bits_count(gb) - s->frame_offset) - 1);
1498 /** skip the rest of the frame data */
1499 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1501 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1505 /** decode trailer bit */
1506 more_frames = get_bits1(gb);
1513 *@brief Calculate remaining input buffer length.
1514 *@param s codec context
1515 *@param gb bitstream reader context
1516 *@return remaining size in bits
1518 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1520 return s->buf_bit_size - get_bits_count(gb);
1524 *@brief Fill the bit reservoir with a (partial) frame.
1525 *@param s codec context
1526 *@param gb bitstream reader context
1527 *@param len length of the partial frame
1528 *@param append decides whether to reset the buffer or not
1530 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1535 /** when the frame data does not need to be concatenated, the input buffer
1536 is reset and additional bits from the previous frame are copied
1537 and skipped later so that a fast byte copy is possible */
1540 s->frame_offset = get_bits_count(gb) & 7;
1541 s->num_saved_bits = s->frame_offset;
1542 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1545 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1547 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1548 avpriv_request_sample(s->avctx, "Too small input buffer");
1553 av_assert0(len <= put_bits_left(&s->pb));
1555 s->num_saved_bits += len;
1557 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1560 int align = 8 - (get_bits_count(gb) & 7);
1561 align = FFMIN(align, len);
1562 put_bits(&s->pb, align, get_bits(gb, align));
1564 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1566 skip_bits_long(gb, len);
1569 PutBitContext tmp = s->pb;
1570 flush_put_bits(&tmp);
1573 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1574 skip_bits(&s->gb, s->frame_offset);
1578 *@brief Decode a single WMA packet.
1579 *@param avctx codec context
1580 *@param data the output buffer
1581 *@param avpkt input packet
1582 *@return number of bytes that were read from the input buffer
1584 static int decode_packet(AVCodecContext *avctx, void *data,
1585 int *got_frame_ptr, AVPacket* avpkt)
1587 WMAProDecodeCtx *s = avctx->priv_data;
1588 GetBitContext* gb = &s->pgb;
1589 const uint8_t* buf = avpkt->data;
1590 int buf_size = avpkt->size;
1591 int num_bits_prev_frame;
1592 int packet_sequence_number;
1596 if (s->skip_packets > 0) {
1598 return FFMIN(avpkt->size, avctx->block_align);
1601 if (s->packet_done || s->packet_loss) {
1604 /** sanity check for the buffer length */
1605 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) {
1606 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1607 buf_size, avctx->block_align);
1608 return AVERROR_INVALIDDATA;
1611 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1612 s->next_packet_start = buf_size - avctx->block_align;
1613 buf_size = avctx->block_align;
1615 s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align);
1616 buf_size = FFMIN(buf_size, avctx->block_align);
1618 s->buf_bit_size = buf_size << 3;
1620 /** parse packet header */
1621 init_get_bits(gb, buf, s->buf_bit_size);
1622 if (avctx->codec_id != AV_CODEC_ID_XMA2) {
1623 packet_sequence_number = get_bits(gb, 4);
1626 s->num_frames = get_bits(gb, 6);
1627 packet_sequence_number = 0;
1630 /** get number of bits that need to be added to the previous frame */
1631 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1632 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) {
1634 s->skip_packets = get_bits(gb, 8);
1637 ff_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1638 num_bits_prev_frame);
1640 /** check for packet loss */
1641 if (avctx->codec_id != AV_CODEC_ID_XMA2 && !s->packet_loss &&
1642 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1644 av_log(avctx, AV_LOG_ERROR,
1645 "Packet loss detected! seq %"PRIx8" vs %x\n",
1646 s->packet_sequence_number, packet_sequence_number);
1648 s->packet_sequence_number = packet_sequence_number;
1650 if (num_bits_prev_frame > 0) {
1651 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1652 if (num_bits_prev_frame >= remaining_packet_bits) {
1653 num_bits_prev_frame = remaining_packet_bits;
1657 /** append the previous frame data to the remaining data from the
1658 previous packet to create a full frame */
1659 save_bits(s, gb, num_bits_prev_frame, 1);
1660 ff_dlog(avctx, "accumulated %x bits of frame data\n",
1661 s->num_saved_bits - s->frame_offset);
1663 /** decode the cross packet frame if it is valid */
1664 if (!s->packet_loss)
1665 decode_frame(s, data, got_frame_ptr);
1666 } else if (s->num_saved_bits - s->frame_offset) {
1667 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1668 s->num_saved_bits - s->frame_offset);
1671 if (s->packet_loss) {
1672 /** reset number of saved bits so that the decoder
1673 does not start to decode incomplete frames in the
1674 s->len_prefix == 0 case */
1675 s->num_saved_bits = 0;
1681 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1682 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1683 skip_bits(gb, s->packet_offset);
1684 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1685 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1686 frame_size <= remaining_bits(s, gb)) {
1687 save_bits(s, gb, frame_size, 0);
1688 if (!s->packet_loss)
1689 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1690 } else if (!s->len_prefix
1691 && s->num_saved_bits > get_bits_count(&s->gb)) {
1692 /** when the frames do not have a length prefix, we don't know
1693 the compressed length of the individual frames
1694 however, we know what part of a new packet belongs to the
1696 therefore we save the incoming packet first, then we append
1697 the "previous frame" data from the next packet so that
1698 we get a buffer that only contains full frames */
1699 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1704 if (remaining_bits(s, gb) < 0) {
1705 av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1709 if (s->packet_done && !s->packet_loss &&
1710 remaining_bits(s, gb) > 0) {
1711 /** save the rest of the data so that it can be decoded
1712 with the next packet */
1713 save_bits(s, gb, remaining_bits(s, gb), 0);
1716 s->packet_offset = get_bits_count(gb) & 7;
1718 return AVERROR_INVALIDDATA;
1720 return get_bits_count(gb) >> 3;
1724 *@brief Clear decoder buffers (for seeking).
1725 *@param avctx codec context
1727 static void flush(AVCodecContext *avctx)
1729 WMAProDecodeCtx *s = avctx->priv_data;
1731 /** reset output buffer as a part of it is used during the windowing of a
1733 for (i = 0; i < avctx->channels; i++)
1734 memset(s->channel[i].out, 0, s->samples_per_frame *
1735 sizeof(*s->channel[i].out));
1741 *@brief wmapro decoder
1743 AVCodec ff_wmapro_decoder = {
1745 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1746 .type = AVMEDIA_TYPE_AUDIO,
1747 .id = AV_CODEC_ID_WMAPRO,
1748 .priv_data_size = sizeof(WMAProDecodeCtx),
1749 .init = decode_init,
1750 .close = decode_end,
1751 .decode = decode_packet,
1752 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1754 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1755 AV_SAMPLE_FMT_NONE },
1758 AVCodec ff_xma1_decoder = {
1760 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 1"),
1761 .type = AVMEDIA_TYPE_AUDIO,
1762 .id = AV_CODEC_ID_XMA1,
1763 .priv_data_size = sizeof(WMAProDecodeCtx),
1764 .init = decode_init,
1765 .close = decode_end,
1766 .decode = decode_packet,
1767 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1769 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1770 AV_SAMPLE_FMT_NONE },
1773 AVCodec ff_xma2_decoder = {
1775 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 2"),
1776 .type = AVMEDIA_TYPE_AUDIO,
1777 .id = AV_CODEC_ID_XMA2,
1778 .priv_data_size = sizeof(WMAProDecodeCtx),
1779 .init = decode_init,
1780 .close = decode_end,
1781 .decode = decode_packet,
1782 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1784 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1785 AV_SAMPLE_FMT_NONE },