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"
95 #include "libavutil/mem_internal.h"
100 #include "put_bits.h"
101 #include "wmaprodata.h"
104 #include "wma_common.h"
106 /** current decoder limitations */
107 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
108 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
109 #define MAX_BANDS 29 ///< max number of scale factor bands
110 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
111 #define XMA_MAX_STREAMS 8
112 #define XMA_MAX_CHANNELS_STREAM 2
113 #define XMA_MAX_CHANNELS (XMA_MAX_STREAMS * XMA_MAX_CHANNELS_STREAM)
115 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
116 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
117 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
118 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
119 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
123 #define SCALEVLCBITS 8
124 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
125 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
126 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
127 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
128 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
130 static VLC sf_vlc; ///< scale factor DPCM vlc
131 static VLC sf_rl_vlc; ///< scale factor run length vlc
132 static VLC vec4_vlc; ///< 4 coefficients per symbol
133 static VLC vec2_vlc; ///< 2 coefficients per symbol
134 static VLC vec1_vlc; ///< 1 coefficient per symbol
135 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
136 static float sin64[33]; ///< sine table for decorrelation
139 * @brief frame specific decoder context for a single channel
141 typedef struct WMAProChannelCtx {
142 int16_t prev_block_len; ///< length of the previous block
143 uint8_t transmit_coefs;
144 uint8_t num_subframes;
145 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
146 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
147 uint8_t cur_subframe; ///< current subframe number
148 uint16_t decoded_samples; ///< number of already processed samples
149 uint8_t grouped; ///< channel is part of a group
150 int quant_step; ///< quantization step for the current subframe
151 int8_t reuse_sf; ///< share scale factors between subframes
152 int8_t scale_factor_step; ///< scaling step for the current subframe
153 int max_scale_factor; ///< maximum scale factor for the current subframe
154 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
155 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
156 int* scale_factors; ///< pointer to the scale factor values used for decoding
157 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
158 float* coeffs; ///< pointer to the subframe decode buffer
159 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
160 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
164 * @brief channel group for channel transformations
166 typedef struct WMAProChannelGrp {
167 uint8_t num_channels; ///< number of channels in the group
168 int8_t transform; ///< transform on / off
169 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
170 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
171 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
175 * @brief main decoder context
177 typedef struct WMAProDecodeCtx {
178 /* generic decoder variables */
179 AVCodecContext* avctx; ///< codec context for av_log
180 AVFloatDSPContext *fdsp;
181 uint8_t frame_data[MAX_FRAMESIZE +
182 AV_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
183 PutBitContext pb; ///< context for filling the frame_data buffer
184 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
185 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
186 const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
188 /* frame size dependent frame information (set during initialization) */
189 uint32_t decode_flags; ///< used compression features
190 uint8_t len_prefix; ///< frame is prefixed with its length
191 uint8_t dynamic_range_compression; ///< frame contains DRC data
192 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
193 uint16_t samples_per_frame; ///< number of samples to output
194 uint16_t log2_frame_size;
195 int8_t lfe_channel; ///< lfe channel index
196 uint8_t max_num_subframes;
197 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
198 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
199 uint16_t min_samples_per_subframe;
200 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
201 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
202 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
203 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
205 /* packet decode state */
206 GetBitContext pgb; ///< bitstream reader context for the packet
207 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
208 uint8_t packet_offset; ///< frame offset in the packet
209 uint8_t packet_sequence_number; ///< current packet number
210 int num_saved_bits; ///< saved number of bits
211 int frame_offset; ///< frame offset in the bit reservoir
212 int subframe_offset; ///< subframe offset in the bit reservoir
213 uint8_t packet_loss; ///< set in case of bitstream error
214 uint8_t packet_done; ///< set when a packet is fully decoded
215 uint8_t eof_done; ///< set when EOF reached and extra subframe is written (XMA1/2)
217 /* frame decode state */
218 uint32_t frame_num; ///< current frame number (not used for decoding)
219 GetBitContext gb; ///< bitstream reader context
220 int buf_bit_size; ///< buffer size in bits
221 uint8_t drc_gain; ///< gain for the DRC tool
222 int8_t skip_frame; ///< skip output step
223 int8_t parsed_all_subframes; ///< all subframes decoded?
224 uint8_t skip_packets; ///< packets to skip to find next packet in a stream (XMA1/2)
226 /* subframe/block decode state */
227 int16_t subframe_len; ///< current subframe length
228 int8_t nb_channels; ///< number of channels in stream (XMA1/2)
229 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
230 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
231 int8_t num_bands; ///< number of scale factor bands
232 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
233 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
234 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
235 int8_t esc_len; ///< length of escaped coefficients
237 uint8_t num_chgroups; ///< number of channel groups
238 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
240 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
243 typedef struct XMADecodeCtx {
244 WMAProDecodeCtx xma[XMA_MAX_STREAMS];
245 AVFrame *frames[XMA_MAX_STREAMS];
248 float samples[XMA_MAX_CHANNELS][512 * 64];
249 int offset[XMA_MAX_STREAMS];
250 int start_channel[XMA_MAX_STREAMS];
254 *@brief helper function to print the most important members of the context
257 static av_cold void dump_context(WMAProDecodeCtx *s)
259 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
260 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
262 PRINT("ed sample bit depth", s->bits_per_sample);
263 PRINT_HEX("ed decode flags", s->decode_flags);
264 PRINT("samples per frame", s->samples_per_frame);
265 PRINT("log2 frame size", s->log2_frame_size);
266 PRINT("max num subframes", s->max_num_subframes);
267 PRINT("len prefix", s->len_prefix);
268 PRINT("num channels", s->nb_channels);
272 *@brief Uninitialize the decoder and free all resources.
273 *@param avctx codec context
274 *@return 0 on success, < 0 otherwise
276 static av_cold int decode_end(WMAProDecodeCtx *s)
282 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
283 ff_mdct_end(&s->mdct_ctx[i]);
288 static av_cold int wmapro_decode_end(AVCodecContext *avctx)
290 WMAProDecodeCtx *s = avctx->priv_data;
297 static av_cold int get_rate(AVCodecContext *avctx)
299 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA?
300 if (avctx->sample_rate > 44100)
302 else if (avctx->sample_rate > 32000)
304 else if (avctx->sample_rate > 24000)
309 return avctx->sample_rate;
313 *@brief Initialize the decoder.
314 *@param avctx codec context
315 *@return 0 on success, -1 otherwise
317 static av_cold int decode_init(WMAProDecodeCtx *s, AVCodecContext *avctx, int num_stream)
319 uint8_t *edata_ptr = avctx->extradata;
320 unsigned int channel_mask;
322 int log2_max_num_subframes;
323 int num_possible_block_sizes;
325 if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2)
326 avctx->block_align = 2048;
328 if (!avctx->block_align) {
329 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
330 return AVERROR(EINVAL);
335 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
337 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
339 /** dump the extradata */
340 av_log(avctx, AV_LOG_DEBUG, "extradata:\n");
341 for (i = 0; i < avctx->extradata_size; i++)
342 av_log(avctx, AV_LOG_DEBUG, "[%x] ", avctx->extradata[i]);
343 av_log(avctx, AV_LOG_DEBUG, "\n");
345 if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
346 s->decode_flags = 0x10d6;
347 s->bits_per_sample = 16;
348 channel_mask = 0; //AV_RL32(edata_ptr+2); /* not always in expected order */
349 if ((num_stream+1) * XMA_MAX_CHANNELS_STREAM > avctx->channels) /* stream config is 2ch + 2ch + ... + 1/2ch */
353 } else if (avctx->codec_id == AV_CODEC_ID_XMA2) { /* XMA2WAVEFORMAT */
354 s->decode_flags = 0x10d6;
355 s->bits_per_sample = 16;
356 channel_mask = 0; /* would need to aggregate from all streams */
357 s->nb_channels = edata_ptr[32 + ((edata_ptr[0]==3)?0:8) + 4*num_stream + 0]; /* nth stream config */
358 } else if (avctx->codec_id == AV_CODEC_ID_XMA1) { /* XMAWAVEFORMAT */
359 s->decode_flags = 0x10d6;
360 s->bits_per_sample = 16;
361 channel_mask = 0; /* would need to aggregate from all streams */
362 s->nb_channels = edata_ptr[8 + 20*num_stream + 17]; /* nth stream config */
363 } else if (avctx->codec_id == AV_CODEC_ID_WMAPRO && avctx->extradata_size >= 18) {
364 s->decode_flags = AV_RL16(edata_ptr+14);
365 channel_mask = AV_RL32(edata_ptr+2);
366 s->bits_per_sample = AV_RL16(edata_ptr);
367 s->nb_channels = avctx->channels;
369 if (s->bits_per_sample > 32 || s->bits_per_sample < 1) {
370 avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample);
371 return AVERROR_PATCHWELCOME;
374 avpriv_request_sample(avctx, "Unknown extradata size");
375 return AVERROR_PATCHWELCOME;
379 s->log2_frame_size = av_log2(avctx->block_align) + 4;
380 if (s->log2_frame_size > 25) {
381 avpriv_request_sample(avctx, "Large block align");
382 return AVERROR_PATCHWELCOME;
386 if (avctx->codec_id != AV_CODEC_ID_WMAPRO)
389 s->skip_frame = 1; /* skip first frame */
392 s->len_prefix = (s->decode_flags & 0x40);
395 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
396 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
397 if (bits > WMAPRO_BLOCK_MAX_BITS) {
398 avpriv_request_sample(avctx, "14-bit block sizes");
399 return AVERROR_PATCHWELCOME;
401 s->samples_per_frame = 1 << bits;
403 s->samples_per_frame = 512;
407 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
408 s->max_num_subframes = 1 << log2_max_num_subframes;
409 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
410 s->max_subframe_len_bit = 1;
411 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
413 num_possible_block_sizes = log2_max_num_subframes + 1;
414 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
415 s->dynamic_range_compression = (s->decode_flags & 0x80);
417 if (s->max_num_subframes > MAX_SUBFRAMES) {
418 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
419 s->max_num_subframes);
420 return AVERROR_INVALIDDATA;
423 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
424 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
425 s->min_samples_per_subframe);
426 return AVERROR_INVALIDDATA;
429 if (s->avctx->sample_rate <= 0) {
430 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
431 return AVERROR_INVALIDDATA;
434 if (s->nb_channels <= 0) {
435 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
437 return AVERROR_INVALIDDATA;
438 } else if (avctx->codec_id != AV_CODEC_ID_WMAPRO && s->nb_channels > XMA_MAX_CHANNELS_STREAM) {
439 av_log(avctx, AV_LOG_ERROR, "invalid number of channels per XMA stream %d\n",
441 return AVERROR_INVALIDDATA;
442 } else if (s->nb_channels > WMAPRO_MAX_CHANNELS || s->nb_channels > avctx->channels) {
443 avpriv_request_sample(avctx,
444 "More than %d channels", WMAPRO_MAX_CHANNELS);
445 return AVERROR_PATCHWELCOME;
448 /** init previous block len */
449 for (i = 0; i < s->nb_channels; i++)
450 s->channel[i].prev_block_len = s->samples_per_frame;
452 /** extract lfe channel position */
455 if (channel_mask & 8) {
457 for (mask = 1; mask < 16; mask <<= 1) {
458 if (channel_mask & mask)
463 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
464 scale_huffbits, 1, 1,
465 scale_huffcodes, 2, 2, 616);
467 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
468 scale_rl_huffbits, 1, 1,
469 scale_rl_huffcodes, 4, 4, 1406);
471 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
472 coef0_huffbits, 1, 1,
473 coef0_huffcodes, 4, 4, 2108);
475 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
476 coef1_huffbits, 1, 1,
477 coef1_huffcodes, 4, 4, 3912);
479 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
481 vec4_huffcodes, 2, 2, 604);
483 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
485 vec2_huffcodes, 2, 2, 562);
487 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
489 vec1_huffcodes, 2, 2, 562);
491 /** calculate number of scale factor bands and their offsets
492 for every possible block size */
493 for (i = 0; i < num_possible_block_sizes; i++) {
494 int subframe_len = s->samples_per_frame >> i;
497 int rate = get_rate(avctx);
499 s->sfb_offsets[i][0] = 0;
501 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
502 int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2;
504 if (offset > s->sfb_offsets[i][band - 1])
505 s->sfb_offsets[i][band++] = offset;
507 if (offset >= subframe_len)
510 s->sfb_offsets[i][band - 1] = subframe_len;
511 s->num_sfb[i] = band - 1;
512 if (s->num_sfb[i] <= 0) {
513 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
514 return AVERROR_INVALIDDATA;
519 /** Scale factors can be shared between blocks of different size
520 as every block has a different scale factor band layout.
521 The matrix sf_offsets is needed to find the correct scale factor.
524 for (i = 0; i < num_possible_block_sizes; i++) {
526 for (b = 0; b < s->num_sfb[i]; b++) {
528 int offset = ((s->sfb_offsets[i][b]
529 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
530 for (x = 0; x < num_possible_block_sizes; x++) {
532 while (s->sfb_offsets[x][v + 1] << x < offset) {
534 av_assert0(v < MAX_BANDS);
536 s->sf_offsets[i][x][b] = v;
541 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
543 return AVERROR(ENOMEM);
545 /** init MDCT, FIXME: only init needed sizes */
546 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
547 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
548 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
549 / (1ll << (s->bits_per_sample - 1)));
551 /** init MDCT windows: simple sine window */
552 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
553 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
554 ff_init_ff_sine_windows(win_idx);
555 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
558 /** calculate subwoofer cutoff values */
559 for (i = 0; i < num_possible_block_sizes; i++) {
560 int block_size = s->samples_per_frame >> i;
561 int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1)
562 / s->avctx->sample_rate;
563 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
566 /** calculate sine values for the decorrelation matrix */
567 for (i = 0; i < 33; i++)
568 sin64[i] = sin(i*M_PI / 64.0);
570 if (avctx->debug & FF_DEBUG_BITSTREAM)
573 avctx->channel_layout = channel_mask;
579 *@brief Initialize the decoder.
580 *@param avctx codec context
581 *@return 0 on success, -1 otherwise
583 static av_cold int wmapro_decode_init(AVCodecContext *avctx)
585 WMAProDecodeCtx *s = avctx->priv_data;
587 return decode_init(s, avctx, 0);
591 *@brief Decode the subframe length.
593 *@param offset sample offset in the frame
594 *@return decoded subframe length on success, < 0 in case of an error
596 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
598 int frame_len_shift = 0;
601 /** no need to read from the bitstream when only one length is possible */
602 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
603 return s->min_samples_per_subframe;
605 if (get_bits_left(&s->gb) < 1)
606 return AVERROR_INVALIDDATA;
608 /** 1 bit indicates if the subframe is of maximum length */
609 if (s->max_subframe_len_bit) {
610 if (get_bits1(&s->gb))
611 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
613 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
615 subframe_len = s->samples_per_frame >> frame_len_shift;
617 /** sanity check the length */
618 if (subframe_len < s->min_samples_per_subframe ||
619 subframe_len > s->samples_per_frame) {
620 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
622 return AVERROR_INVALIDDATA;
628 *@brief Decode how the data in the frame is split into subframes.
629 * Every WMA frame contains the encoded data for a fixed number of
630 * samples per channel. The data for every channel might be split
631 * into several subframes. This function will reconstruct the list of
632 * subframes for every channel.
634 * If the subframes are not evenly split, the algorithm estimates the
635 * channels with the lowest number of total samples.
636 * Afterwards, for each of these channels a bit is read from the
637 * bitstream that indicates if the channel contains a subframe with the
638 * next subframe size that is going to be read from the bitstream or not.
639 * If a channel contains such a subframe, the subframe size gets added to
640 * the channel's subframe list.
641 * The algorithm repeats these steps until the frame is properly divided
642 * between the individual channels.
645 *@return 0 on success, < 0 in case of an error
647 static int decode_tilehdr(WMAProDecodeCtx *s)
649 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
650 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
651 int channels_for_cur_subframe = s->nb_channels; /**< number of channels that contain the current subframe */
652 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
653 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
656 /* Should never consume more than 3073 bits (256 iterations for the
657 * while loop when always the minimum amount of 128 samples is subtracted
658 * from missing samples in the 8 channel case).
659 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
662 /** reset tiling information */
663 for (c = 0; c < s->nb_channels; c++)
664 s->channel[c].num_subframes = 0;
666 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
667 fixed_channel_layout = 1;
669 /** loop until the frame data is split between the subframes */
673 /** check which channels contain the subframe */
674 for (c = 0; c < s->nb_channels; c++) {
675 if (num_samples[c] == min_channel_len) {
676 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
677 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
678 contains_subframe[c] = 1;
680 contains_subframe[c] = get_bits1(&s->gb);
682 contains_subframe[c] = 0;
685 /** get subframe length, subframe_len == 0 is not allowed */
686 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
687 return AVERROR_INVALIDDATA;
689 /** add subframes to the individual channels and find new min_channel_len */
690 min_channel_len += subframe_len;
691 for (c = 0; c < s->nb_channels; c++) {
692 WMAProChannelCtx* chan = &s->channel[c];
694 if (contains_subframe[c]) {
695 if (chan->num_subframes >= MAX_SUBFRAMES) {
696 av_log(s->avctx, AV_LOG_ERROR,
697 "broken frame: num subframes > 31\n");
698 return AVERROR_INVALIDDATA;
700 chan->subframe_len[chan->num_subframes] = subframe_len;
701 num_samples[c] += subframe_len;
702 ++chan->num_subframes;
703 if (num_samples[c] > s->samples_per_frame) {
704 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
705 "channel len > samples_per_frame\n");
706 return AVERROR_INVALIDDATA;
708 } else if (num_samples[c] <= min_channel_len) {
709 if (num_samples[c] < min_channel_len) {
710 channels_for_cur_subframe = 0;
711 min_channel_len = num_samples[c];
713 ++channels_for_cur_subframe;
716 } while (min_channel_len < s->samples_per_frame);
718 for (c = 0; c < s->nb_channels; c++) {
721 for (i = 0; i < s->channel[c].num_subframes; i++) {
722 ff_dlog(s->avctx, "frame[%"PRIu32"] channel[%i] subframe[%i]"
723 " len %i\n", s->frame_num, c, i,
724 s->channel[c].subframe_len[i]);
725 s->channel[c].subframe_offset[i] = offset;
726 offset += s->channel[c].subframe_len[i];
734 *@brief Calculate a decorrelation matrix from the bitstream parameters.
735 *@param s codec context
736 *@param chgroup channel group for which the matrix needs to be calculated
738 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
739 WMAProChannelGrp *chgroup)
743 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
744 memset(chgroup->decorrelation_matrix, 0, s->nb_channels *
745 s->nb_channels * sizeof(*chgroup->decorrelation_matrix));
747 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
748 rotation_offset[i] = get_bits(&s->gb, 6);
750 for (i = 0; i < chgroup->num_channels; i++)
751 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
752 get_bits1(&s->gb) ? 1.0 : -1.0;
754 for (i = 1; i < chgroup->num_channels; i++) {
756 for (x = 0; x < i; x++) {
758 for (y = 0; y < i + 1; y++) {
759 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
760 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
761 int n = rotation_offset[offset + x];
767 cosv = sin64[32 - n];
769 sinv = sin64[64 - n];
770 cosv = -sin64[n - 32];
773 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
774 (v1 * sinv) - (v2 * cosv);
775 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
776 (v1 * cosv) + (v2 * sinv);
784 *@brief Decode channel transformation parameters
785 *@param s codec context
786 *@return >= 0 in case of success, < 0 in case of bitstream errors
788 static int decode_channel_transform(WMAProDecodeCtx* s)
791 /* should never consume more than 1921 bits for the 8 channel case
792 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
793 * + MAX_CHANNELS + MAX_BANDS + 1)
796 /** in the one channel case channel transforms are pointless */
798 if (s->nb_channels > 1) {
799 int remaining_channels = s->channels_for_cur_subframe;
801 if (get_bits1(&s->gb)) {
802 avpriv_request_sample(s->avctx,
803 "Channel transform bit");
804 return AVERROR_PATCHWELCOME;
807 for (s->num_chgroups = 0; remaining_channels &&
808 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
809 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
810 float** channel_data = chgroup->channel_data;
811 chgroup->num_channels = 0;
812 chgroup->transform = 0;
814 /** decode channel mask */
815 if (remaining_channels > 2) {
816 for (i = 0; i < s->channels_for_cur_subframe; i++) {
817 int channel_idx = s->channel_indexes_for_cur_subframe[i];
818 if (!s->channel[channel_idx].grouped
819 && get_bits1(&s->gb)) {
820 ++chgroup->num_channels;
821 s->channel[channel_idx].grouped = 1;
822 *channel_data++ = s->channel[channel_idx].coeffs;
826 chgroup->num_channels = remaining_channels;
827 for (i = 0; i < s->channels_for_cur_subframe; i++) {
828 int channel_idx = s->channel_indexes_for_cur_subframe[i];
829 if (!s->channel[channel_idx].grouped)
830 *channel_data++ = s->channel[channel_idx].coeffs;
831 s->channel[channel_idx].grouped = 1;
835 /** decode transform type */
836 if (chgroup->num_channels == 2) {
837 if (get_bits1(&s->gb)) {
838 if (get_bits1(&s->gb)) {
839 avpriv_request_sample(s->avctx,
840 "Unknown channel transform type");
841 return AVERROR_PATCHWELCOME;
844 chgroup->transform = 1;
845 if (s->nb_channels == 2) {
846 chgroup->decorrelation_matrix[0] = 1.0;
847 chgroup->decorrelation_matrix[1] = -1.0;
848 chgroup->decorrelation_matrix[2] = 1.0;
849 chgroup->decorrelation_matrix[3] = 1.0;
852 chgroup->decorrelation_matrix[0] = 0.70703125;
853 chgroup->decorrelation_matrix[1] = -0.70703125;
854 chgroup->decorrelation_matrix[2] = 0.70703125;
855 chgroup->decorrelation_matrix[3] = 0.70703125;
858 } else if (chgroup->num_channels > 2) {
859 if (get_bits1(&s->gb)) {
860 chgroup->transform = 1;
861 if (get_bits1(&s->gb)) {
862 decode_decorrelation_matrix(s, chgroup);
864 /** FIXME: more than 6 coupled channels not supported */
865 if (chgroup->num_channels > 6) {
866 avpriv_request_sample(s->avctx,
867 "Coupled channels > 6");
869 memcpy(chgroup->decorrelation_matrix,
870 default_decorrelation[chgroup->num_channels],
871 chgroup->num_channels * chgroup->num_channels *
872 sizeof(*chgroup->decorrelation_matrix));
878 /** decode transform on / off */
879 if (chgroup->transform) {
880 if (!get_bits1(&s->gb)) {
882 /** transform can be enabled for individual bands */
883 for (i = 0; i < s->num_bands; i++) {
884 chgroup->transform_band[i] = get_bits1(&s->gb);
887 memset(chgroup->transform_band, 1, s->num_bands);
890 remaining_channels -= chgroup->num_channels;
897 *@brief Extract the coefficients from the bitstream.
898 *@param s codec context
899 *@param c current channel number
900 *@return 0 on success, < 0 in case of bitstream errors
902 static int decode_coeffs(WMAProDecodeCtx *s, int c)
904 /* Integers 0..15 as single-precision floats. The table saves a
905 costly int to float conversion, and storing the values as
906 integers allows fast sign-flipping. */
907 static const uint32_t fval_tab[16] = {
908 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
909 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
910 0x41000000, 0x41100000, 0x41200000, 0x41300000,
911 0x41400000, 0x41500000, 0x41600000, 0x41700000,
915 WMAProChannelCtx* ci = &s->channel[c];
922 ff_dlog(s->avctx, "decode coefficients for channel %i\n", c);
924 vlctable = get_bits1(&s->gb);
925 vlc = &coef_vlc[vlctable];
935 /** decode vector coefficients (consumes up to 167 bits per iteration for
936 4 vector coded large values) */
937 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
938 (cur_coeff + 3 < ci->num_vec_coeffs)) {
943 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
945 if (idx == HUFF_VEC4_SIZE - 1) {
946 for (i = 0; i < 4; i += 2) {
947 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
948 if (idx == HUFF_VEC2_SIZE - 1) {
950 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
951 if (v0 == HUFF_VEC1_SIZE - 1)
952 v0 += ff_wma_get_large_val(&s->gb);
953 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
954 if (v1 == HUFF_VEC1_SIZE - 1)
955 v1 += ff_wma_get_large_val(&s->gb);
956 vals[i ] = av_float2int(v0);
957 vals[i+1] = av_float2int(v1);
959 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
960 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
964 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
965 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
966 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
967 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
971 for (i = 0; i < 4; i++) {
973 uint32_t sign = get_bits1(&s->gb) - 1;
974 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
977 ci->coeffs[cur_coeff] = 0;
978 /** switch to run level mode when subframe_len / 128 zeros
979 were found in a row */
980 rl_mode |= (++num_zeros > s->subframe_len >> 8);
986 /** decode run level coded coefficients */
987 if (cur_coeff < s->subframe_len) {
988 memset(&ci->coeffs[cur_coeff], 0,
989 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
990 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
991 level, run, 1, ci->coeffs,
992 cur_coeff, s->subframe_len,
993 s->subframe_len, s->esc_len, 0))
994 return AVERROR_INVALIDDATA;
1001 *@brief Extract scale factors from the bitstream.
1002 *@param s codec context
1003 *@return 0 on success, < 0 in case of bitstream errors
1005 static int decode_scale_factors(WMAProDecodeCtx* s)
1009 /** should never consume more than 5344 bits
1010 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
1013 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1014 int c = s->channel_indexes_for_cur_subframe[i];
1017 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
1018 sf_end = s->channel[c].scale_factors + s->num_bands;
1020 /** resample scale factors for the new block size
1021 * as the scale factors might need to be resampled several times
1022 * before some new values are transmitted, a backup of the last
1023 * transmitted scale factors is kept in saved_scale_factors
1025 if (s->channel[c].reuse_sf) {
1026 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
1028 for (b = 0; b < s->num_bands; b++)
1029 s->channel[c].scale_factors[b] =
1030 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
1033 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
1035 if (!s->channel[c].reuse_sf) {
1037 /** decode DPCM coded scale factors */
1038 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
1039 val = 45 / s->channel[c].scale_factor_step;
1040 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
1041 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
1046 /** run level decode differences to the resampled factors */
1047 for (i = 0; i < s->num_bands; i++) {
1053 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
1056 uint32_t code = get_bits(&s->gb, 14);
1058 sign = (code & 1) - 1;
1059 skip = (code & 0x3f) >> 1;
1060 } else if (idx == 1) {
1063 skip = scale_rl_run[idx];
1064 val = scale_rl_level[idx];
1065 sign = get_bits1(&s->gb)-1;
1069 if (i >= s->num_bands) {
1070 av_log(s->avctx, AV_LOG_ERROR,
1071 "invalid scale factor coding\n");
1072 return AVERROR_INVALIDDATA;
1074 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
1078 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
1079 s->channel[c].table_idx = s->table_idx;
1080 s->channel[c].reuse_sf = 1;
1083 /** calculate new scale factor maximum */
1084 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
1085 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
1086 s->channel[c].max_scale_factor =
1087 FFMAX(s->channel[c].max_scale_factor, *sf);
1095 *@brief Reconstruct the individual channel data.
1096 *@param s codec context
1098 static void inverse_channel_transform(WMAProDecodeCtx *s)
1102 for (i = 0; i < s->num_chgroups; i++) {
1103 if (s->chgroup[i].transform) {
1104 float data[WMAPRO_MAX_CHANNELS];
1105 const int num_channels = s->chgroup[i].num_channels;
1106 float** ch_data = s->chgroup[i].channel_data;
1107 float** ch_end = ch_data + num_channels;
1108 const int8_t* tb = s->chgroup[i].transform_band;
1111 /** multichannel decorrelation */
1112 for (sfb = s->cur_sfb_offsets;
1113 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1116 /** multiply values with the decorrelation_matrix */
1117 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1118 const float* mat = s->chgroup[i].decorrelation_matrix;
1119 const float* data_end = data + num_channels;
1120 float* data_ptr = data;
1123 for (ch = ch_data; ch < ch_end; ch++)
1124 *data_ptr++ = (*ch)[y];
1126 for (ch = ch_data; ch < ch_end; ch++) {
1129 while (data_ptr < data_end)
1130 sum += *data_ptr++ * *mat++;
1135 } else if (s->nb_channels == 2) {
1136 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1137 s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0],
1138 ch_data[0] + sfb[0],
1140 s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0],
1141 ch_data[1] + sfb[0],
1150 *@brief Apply sine window and reconstruct the output buffer.
1151 *@param s codec context
1153 static void wmapro_window(WMAProDecodeCtx *s)
1156 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1157 int c = s->channel_indexes_for_cur_subframe[i];
1158 const float* window;
1159 int winlen = s->channel[c].prev_block_len;
1160 float* start = s->channel[c].coeffs - (winlen >> 1);
1162 if (s->subframe_len < winlen) {
1163 start += (winlen - s->subframe_len) >> 1;
1164 winlen = s->subframe_len;
1167 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1171 s->fdsp->vector_fmul_window(start, start, start + winlen,
1174 s->channel[c].prev_block_len = s->subframe_len;
1179 *@brief Decode a single subframe (block).
1180 *@param s codec context
1181 *@return 0 on success, < 0 when decoding failed
1183 static int decode_subframe(WMAProDecodeCtx *s)
1185 int offset = s->samples_per_frame;
1186 int subframe_len = s->samples_per_frame;
1188 int total_samples = s->samples_per_frame * s->nb_channels;
1189 int transmit_coeffs = 0;
1190 int cur_subwoofer_cutoff;
1192 s->subframe_offset = get_bits_count(&s->gb);
1194 /** reset channel context and find the next block offset and size
1195 == the next block of the channel with the smallest number of
1198 for (i = 0; i < s->nb_channels; i++) {
1199 s->channel[i].grouped = 0;
1200 if (offset > s->channel[i].decoded_samples) {
1201 offset = s->channel[i].decoded_samples;
1203 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1208 "processing subframe with offset %i len %i\n", offset, subframe_len);
1210 /** get a list of all channels that contain the estimated block */
1211 s->channels_for_cur_subframe = 0;
1212 for (i = 0; i < s->nb_channels; i++) {
1213 const int cur_subframe = s->channel[i].cur_subframe;
1214 /** subtract already processed samples */
1215 total_samples -= s->channel[i].decoded_samples;
1217 /** and count if there are multiple subframes that match our profile */
1218 if (offset == s->channel[i].decoded_samples &&
1219 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1220 total_samples -= s->channel[i].subframe_len[cur_subframe];
1221 s->channel[i].decoded_samples +=
1222 s->channel[i].subframe_len[cur_subframe];
1223 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1224 ++s->channels_for_cur_subframe;
1228 /** check if the frame will be complete after processing the
1231 s->parsed_all_subframes = 1;
1234 ff_dlog(s->avctx, "subframe is part of %i channels\n",
1235 s->channels_for_cur_subframe);
1237 /** calculate number of scale factor bands and their offsets */
1238 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1239 s->num_bands = s->num_sfb[s->table_idx];
1240 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1241 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1243 /** configure the decoder for the current subframe */
1244 offset += s->samples_per_frame >> 1;
1246 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1247 int c = s->channel_indexes_for_cur_subframe[i];
1249 s->channel[c].coeffs = &s->channel[c].out[offset];
1252 s->subframe_len = subframe_len;
1253 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1255 /** skip extended header if any */
1256 if (get_bits1(&s->gb)) {
1258 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1259 int len = get_bits(&s->gb, 4);
1260 num_fill_bits = get_bitsz(&s->gb, len) + 1;
1263 if (num_fill_bits >= 0) {
1264 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1265 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1266 return AVERROR_INVALIDDATA;
1269 skip_bits_long(&s->gb, num_fill_bits);
1273 /** no idea for what the following bit is used */
1274 if (get_bits1(&s->gb)) {
1275 avpriv_request_sample(s->avctx, "Reserved bit");
1276 return AVERROR_PATCHWELCOME;
1280 if (decode_channel_transform(s) < 0)
1281 return AVERROR_INVALIDDATA;
1284 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1285 int c = s->channel_indexes_for_cur_subframe[i];
1286 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1287 transmit_coeffs = 1;
1290 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1291 if (transmit_coeffs) {
1293 int quant_step = 90 * s->bits_per_sample >> 4;
1295 /** decode number of vector coded coefficients */
1296 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1297 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1298 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1299 int c = s->channel_indexes_for_cur_subframe[i];
1300 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1301 if (num_vec_coeffs > s->subframe_len) {
1302 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1303 return AVERROR_INVALIDDATA;
1305 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1306 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1309 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1310 int c = s->channel_indexes_for_cur_subframe[i];
1311 s->channel[c].num_vec_coeffs = s->subframe_len;
1314 /** decode quantization step */
1315 step = get_sbits(&s->gb, 6);
1317 if (step == -32 || step == 31) {
1318 const int sign = (step == 31) - 1;
1320 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1321 (step = get_bits(&s->gb, 5)) == 31) {
1324 quant_step += ((quant + step) ^ sign) - sign;
1326 if (quant_step < 0) {
1327 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1330 /** decode quantization step modifiers for every channel */
1332 if (s->channels_for_cur_subframe == 1) {
1333 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1335 int modifier_len = get_bits(&s->gb, 3);
1336 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1337 int c = s->channel_indexes_for_cur_subframe[i];
1338 s->channel[c].quant_step = quant_step;
1339 if (get_bits1(&s->gb)) {
1341 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1343 ++s->channel[c].quant_step;
1348 /** decode scale factors */
1349 if (decode_scale_factors(s) < 0)
1350 return AVERROR_INVALIDDATA;
1353 ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1354 get_bits_count(&s->gb) - s->subframe_offset);
1356 /** parse coefficients */
1357 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1358 int c = s->channel_indexes_for_cur_subframe[i];
1359 if (s->channel[c].transmit_coefs &&
1360 get_bits_count(&s->gb) < s->num_saved_bits) {
1361 decode_coeffs(s, c);
1363 memset(s->channel[c].coeffs, 0,
1364 sizeof(*s->channel[c].coeffs) * subframe_len);
1367 ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1368 get_bits_count(&s->gb) - s->subframe_offset);
1370 if (transmit_coeffs) {
1371 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1372 /** reconstruct the per channel data */
1373 inverse_channel_transform(s);
1374 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1375 int c = s->channel_indexes_for_cur_subframe[i];
1376 const int* sf = s->channel[c].scale_factors;
1379 if (c == s->lfe_channel)
1380 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1381 (subframe_len - cur_subwoofer_cutoff));
1383 /** inverse quantization and rescaling */
1384 for (b = 0; b < s->num_bands; b++) {
1385 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1386 const int exp = s->channel[c].quant_step -
1387 (s->channel[c].max_scale_factor - *sf++) *
1388 s->channel[c].scale_factor_step;
1389 const float quant = ff_exp10(exp / 20.0);
1390 int start = s->cur_sfb_offsets[b];
1391 s->fdsp->vector_fmul_scalar(s->tmp + start,
1392 s->channel[c].coeffs + start,
1393 quant, end - start);
1396 /** apply imdct (imdct_half == DCTIV with reverse) */
1397 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1401 /** window and overlapp-add */
1404 /** handled one subframe */
1405 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1406 int c = s->channel_indexes_for_cur_subframe[i];
1407 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1408 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1409 return AVERROR_INVALIDDATA;
1411 ++s->channel[c].cur_subframe;
1418 *@brief Decode one WMA frame.
1419 *@param s codec context
1420 *@return 0 if the trailer bit indicates that this is the last frame,
1421 * 1 if there are additional frames
1423 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1425 GetBitContext* gb = &s->gb;
1426 int more_frames = 0;
1430 /** get frame length */
1432 len = get_bits(gb, s->log2_frame_size);
1434 ff_dlog(s->avctx, "decoding frame with length %x\n", len);
1436 /** decode tile information */
1437 if (decode_tilehdr(s)) {
1442 /** read postproc transform */
1443 if (s->nb_channels > 1 && get_bits1(gb)) {
1444 if (get_bits1(gb)) {
1445 for (i = 0; i < s->nb_channels * s->nb_channels; i++)
1450 /** read drc info */
1451 if (s->dynamic_range_compression) {
1452 s->drc_gain = get_bits(gb, 8);
1453 ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1456 /** no idea what these are for, might be the number of samples
1457 that need to be skipped at the beginning or end of a stream */
1458 if (get_bits1(gb)) {
1461 /** usually true for the first frame */
1462 if (get_bits1(gb)) {
1463 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1464 ff_dlog(s->avctx, "start skip: %i\n", skip);
1467 /** sometimes true for the last frame */
1468 if (get_bits1(gb)) {
1469 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1470 ff_dlog(s->avctx, "end skip: %i\n", skip);
1475 ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1476 get_bits_count(gb) - s->frame_offset);
1478 /** reset subframe states */
1479 s->parsed_all_subframes = 0;
1480 for (i = 0; i < s->nb_channels; i++) {
1481 s->channel[i].decoded_samples = 0;
1482 s->channel[i].cur_subframe = 0;
1483 s->channel[i].reuse_sf = 0;
1486 /** decode all subframes */
1487 while (!s->parsed_all_subframes) {
1488 if (decode_subframe(s) < 0) {
1494 /** copy samples to the output buffer */
1495 for (i = 0; i < s->nb_channels; i++)
1496 memcpy(frame->extended_data[i], s->channel[i].out,
1497 s->samples_per_frame * sizeof(*s->channel[i].out));
1499 for (i = 0; i < s->nb_channels; i++) {
1500 /** reuse second half of the IMDCT output for the next frame */
1501 memcpy(&s->channel[i].out[0],
1502 &s->channel[i].out[s->samples_per_frame],
1503 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1506 if (s->skip_frame) {
1509 av_frame_unref(frame);
1514 if (s->len_prefix) {
1515 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1516 /** FIXME: not sure if this is always an error */
1517 av_log(s->avctx, AV_LOG_ERROR,
1518 "frame[%"PRIu32"] would have to skip %i bits\n",
1520 len - (get_bits_count(gb) - s->frame_offset) - 1);
1525 /** skip the rest of the frame data */
1526 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1528 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1532 /** decode trailer bit */
1533 more_frames = get_bits1(gb);
1540 *@brief Calculate remaining input buffer length.
1541 *@param s codec context
1542 *@param gb bitstream reader context
1543 *@return remaining size in bits
1545 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1547 return s->buf_bit_size - get_bits_count(gb);
1551 *@brief Fill the bit reservoir with a (partial) frame.
1552 *@param s codec context
1553 *@param gb bitstream reader context
1554 *@param len length of the partial frame
1555 *@param append decides whether to reset the buffer or not
1557 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1562 /** when the frame data does not need to be concatenated, the input buffer
1563 is reset and additional bits from the previous frame are copied
1564 and skipped later so that a fast byte copy is possible */
1567 s->frame_offset = get_bits_count(gb) & 7;
1568 s->num_saved_bits = s->frame_offset;
1569 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1570 buflen = (s->num_saved_bits + len + 7) >> 3;
1572 buflen = (put_bits_count(&s->pb) + len + 7) >> 3;
1574 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1575 avpriv_request_sample(s->avctx, "Too small input buffer");
1580 av_assert0(len <= put_bits_left(&s->pb));
1582 s->num_saved_bits += len;
1584 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1587 int align = 8 - (get_bits_count(gb) & 7);
1588 align = FFMIN(align, len);
1589 put_bits(&s->pb, align, get_bits(gb, align));
1591 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1593 skip_bits_long(gb, len);
1596 PutBitContext tmp = s->pb;
1597 flush_put_bits(&tmp);
1600 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1601 skip_bits(&s->gb, s->frame_offset);
1604 static int decode_packet(AVCodecContext *avctx, WMAProDecodeCtx *s,
1605 void *data, int *got_frame_ptr, AVPacket *avpkt)
1607 GetBitContext* gb = &s->pgb;
1608 const uint8_t* buf = avpkt->data;
1609 int buf_size = avpkt->size;
1610 int num_bits_prev_frame;
1611 int packet_sequence_number;
1616 AVFrame *frame = data;
1619 /** Must output remaining samples after stream end. WMAPRO 5.1 created
1620 * by XWMA encoder don't though (maybe only 1/2ch streams need it). */
1625 /** clean output buffer and copy last IMDCT samples */
1626 for (i = 0; i < s->nb_channels; i++) {
1627 memset(frame->extended_data[i], 0,
1628 s->samples_per_frame * sizeof(*s->channel[i].out));
1630 memcpy(frame->extended_data[i], s->channel[i].out,
1631 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1634 /* TODO: XMA should output 128 samples only (instead of 512) and WMAPRO
1635 * maybe 768 (with 2048), XMA needs changes in multi-stream handling though. */
1642 else if (s->packet_done || s->packet_loss) {
1645 /** sanity check for the buffer length */
1646 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) {
1647 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1648 buf_size, avctx->block_align);
1650 return AVERROR_INVALIDDATA;
1653 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1654 s->next_packet_start = buf_size - avctx->block_align;
1655 buf_size = avctx->block_align;
1657 s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align);
1658 buf_size = FFMIN(buf_size, avctx->block_align);
1660 s->buf_bit_size = buf_size << 3;
1662 /** parse packet header */
1663 init_get_bits(gb, buf, s->buf_bit_size);
1664 if (avctx->codec_id != AV_CODEC_ID_XMA2) {
1665 packet_sequence_number = get_bits(gb, 4);
1668 int num_frames = get_bits(gb, 6);
1669 ff_dlog(avctx, "packet[%d]: number of frames %d\n", avctx->frame_number, num_frames);
1670 packet_sequence_number = 0;
1673 /** get number of bits that need to be added to the previous frame */
1674 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1675 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) {
1677 s->skip_packets = get_bits(gb, 8);
1678 ff_dlog(avctx, "packet[%d]: skip packets %d\n", avctx->frame_number, s->skip_packets);
1681 ff_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1682 num_bits_prev_frame);
1684 /** check for packet loss */
1685 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && !s->packet_loss &&
1686 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1688 av_log(avctx, AV_LOG_ERROR,
1689 "Packet loss detected! seq %"PRIx8" vs %x\n",
1690 s->packet_sequence_number, packet_sequence_number);
1692 s->packet_sequence_number = packet_sequence_number;
1694 if (num_bits_prev_frame > 0) {
1695 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1696 if (num_bits_prev_frame >= remaining_packet_bits) {
1697 num_bits_prev_frame = remaining_packet_bits;
1701 /** append the previous frame data to the remaining data from the
1702 previous packet to create a full frame */
1703 save_bits(s, gb, num_bits_prev_frame, 1);
1704 ff_dlog(avctx, "accumulated %x bits of frame data\n",
1705 s->num_saved_bits - s->frame_offset);
1707 /** decode the cross packet frame if it is valid */
1708 if (!s->packet_loss)
1709 decode_frame(s, data, got_frame_ptr);
1710 } else if (s->num_saved_bits - s->frame_offset) {
1711 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1712 s->num_saved_bits - s->frame_offset);
1715 if (s->packet_loss) {
1716 /** reset number of saved bits so that the decoder
1717 does not start to decode incomplete frames in the
1718 s->len_prefix == 0 case */
1719 s->num_saved_bits = 0;
1725 if (avpkt->size < s->next_packet_start) {
1727 return AVERROR_INVALIDDATA;
1730 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1731 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1732 skip_bits(gb, s->packet_offset);
1733 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1734 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1735 frame_size <= remaining_bits(s, gb)) {
1736 save_bits(s, gb, frame_size, 0);
1737 if (!s->packet_loss)
1738 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1739 } else if (!s->len_prefix
1740 && s->num_saved_bits > get_bits_count(&s->gb)) {
1741 /** when the frames do not have a length prefix, we don't know
1742 the compressed length of the individual frames
1743 however, we know what part of a new packet belongs to the
1745 therefore we save the incoming packet first, then we append
1746 the "previous frame" data from the next packet so that
1747 we get a buffer that only contains full frames */
1748 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1754 if (remaining_bits(s, gb) < 0) {
1755 av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1759 if (s->packet_done && !s->packet_loss &&
1760 remaining_bits(s, gb) > 0) {
1761 /** save the rest of the data so that it can be decoded
1762 with the next packet */
1763 save_bits(s, gb, remaining_bits(s, gb), 0);
1766 s->packet_offset = get_bits_count(gb) & 7;
1768 return AVERROR_INVALIDDATA;
1770 return get_bits_count(gb) >> 3;
1774 *@brief Decode a single WMA packet.
1775 *@param avctx codec context
1776 *@param data the output buffer
1777 *@param avpkt input packet
1778 *@return number of bytes that were read from the input buffer
1780 static int wmapro_decode_packet(AVCodecContext *avctx, void *data,
1781 int *got_frame_ptr, AVPacket *avpkt)
1783 WMAProDecodeCtx *s = avctx->priv_data;
1784 AVFrame *frame = data;
1787 /* get output buffer */
1788 frame->nb_samples = s->samples_per_frame;
1789 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1794 return decode_packet(avctx, s, data, got_frame_ptr, avpkt);
1797 static int xma_decode_packet(AVCodecContext *avctx, void *data,
1798 int *got_frame_ptr, AVPacket *avpkt)
1800 XMADecodeCtx *s = avctx->priv_data;
1801 int got_stream_frame_ptr = 0;
1802 AVFrame *frame = data;
1803 int i, ret, offset = INT_MAX;
1805 if (!s->frames[s->current_stream]->data[0]) {
1806 s->frames[s->current_stream]->nb_samples = 512;
1807 if ((ret = ff_get_buffer(avctx, s->frames[s->current_stream], 0)) < 0) {
1811 /* decode current stream packet */
1812 ret = decode_packet(avctx, &s->xma[s->current_stream], s->frames[s->current_stream],
1813 &got_stream_frame_ptr, avpkt);
1815 if (got_stream_frame_ptr && s->offset[s->current_stream] >= 64) {
1816 got_stream_frame_ptr = 0;
1817 ret = AVERROR_INVALIDDATA;
1820 /* copy stream samples (1/2ch) to sample buffer (Nch) */
1821 if (got_stream_frame_ptr) {
1822 int start_ch = s->start_channel[s->current_stream];
1823 memcpy(&s->samples[start_ch + 0][s->offset[s->current_stream] * 512],
1824 s->frames[s->current_stream]->extended_data[0], 512 * 4);
1825 if (s->xma[s->current_stream].nb_channels > 1)
1826 memcpy(&s->samples[start_ch + 1][s->offset[s->current_stream] * 512],
1827 s->frames[s->current_stream]->extended_data[1], 512 * 4);
1828 s->offset[s->current_stream]++;
1829 } else if (ret < 0) {
1830 memset(s->offset, 0, sizeof(s->offset));
1831 s->current_stream = 0;
1835 /* find next XMA packet's owner stream, and update.
1836 * XMA streams find their packets following packet_skips
1837 * (at start there is one packet per stream, then interleave non-linearly). */
1838 if (s->xma[s->current_stream].packet_done ||
1839 s->xma[s->current_stream].packet_loss) {
1841 /* select stream with 0 skip_packets (= uses next packet) */
1842 if (s->xma[s->current_stream].skip_packets != 0) {
1845 min[0] = s->xma[0].skip_packets;
1848 for (i = 1; i < s->num_streams; i++) {
1849 if (s->xma[i].skip_packets < min[0]) {
1850 min[0] = s->xma[i].skip_packets;
1855 s->current_stream = min[1];
1858 /* all other streams skip next packet */
1859 for (i = 0; i < s->num_streams; i++) {
1860 s->xma[i].skip_packets = FFMAX(0, s->xma[i].skip_packets - 1);
1863 /* copy samples from buffer to output if possible */
1864 for (i = 0; i < s->num_streams; i++) {
1865 offset = FFMIN(offset, s->offset[i]);
1870 frame->nb_samples = 512 * offset;
1871 if ((bret = ff_get_buffer(avctx, frame, 0)) < 0)
1874 /* copy samples buffer (Nch) to frame samples (Nch), move unconsumed samples */
1875 for (i = 0; i < s->num_streams; i++) {
1876 int start_ch = s->start_channel[i];
1877 memcpy(frame->extended_data[start_ch + 0], s->samples[start_ch + 0], frame->nb_samples * 4);
1878 if (s->xma[i].nb_channels > 1)
1879 memcpy(frame->extended_data[start_ch + 1], s->samples[start_ch + 1], frame->nb_samples * 4);
1881 s->offset[i] -= offset;
1883 memmove(s->samples[start_ch + 0], s->samples[start_ch + 0] + frame->nb_samples, s->offset[i] * 4 * 512);
1884 if (s->xma[i].nb_channels > 1)
1885 memmove(s->samples[start_ch + 1], s->samples[start_ch + 1] + frame->nb_samples, s->offset[i] * 4 * 512);
1896 static av_cold int xma_decode_init(AVCodecContext *avctx)
1898 XMADecodeCtx *s = avctx->priv_data;
1899 int i, ret, start_channels = 0;
1901 if (avctx->channels <= 0 || avctx->extradata_size == 0)
1902 return AVERROR_INVALIDDATA;
1904 /* get stream config */
1905 if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
1906 s->num_streams = (avctx->channels + 1) / 2;
1907 } else if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size >= 2) { /* XMA2WAVEFORMAT */
1908 s->num_streams = avctx->extradata[1];
1909 if (avctx->extradata_size != (32 + ((avctx->extradata[0]==3)?0:8) + 4*s->num_streams)) {
1910 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA2 extradata size\n");
1912 return AVERROR(EINVAL);
1914 } else if (avctx->codec_id == AV_CODEC_ID_XMA1 && avctx->extradata_size >= 4) { /* XMAWAVEFORMAT */
1915 s->num_streams = avctx->extradata[4];
1916 if (avctx->extradata_size != (8 + 20*s->num_streams)) {
1917 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA1 extradata size\n");
1919 return AVERROR(EINVAL);
1922 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA config\n");
1923 return AVERROR(EINVAL);
1926 /* encoder supports up to 64 streams / 64*2 channels (would have to alloc arrays) */
1927 if (avctx->channels > XMA_MAX_CHANNELS || s->num_streams > XMA_MAX_STREAMS ||
1930 avpriv_request_sample(avctx, "More than %d channels in %d streams", XMA_MAX_CHANNELS, s->num_streams);
1932 return AVERROR_PATCHWELCOME;
1935 /* init all streams (several streams of 1/2ch make Nch files) */
1936 for (i = 0; i < s->num_streams; i++) {
1937 ret = decode_init(&s->xma[i], avctx, i);
1940 s->frames[i] = av_frame_alloc();
1942 return AVERROR(ENOMEM);
1944 s->start_channel[i] = start_channels;
1945 start_channels += s->xma[i].nb_channels;
1947 if (start_channels != avctx->channels)
1948 return AVERROR_INVALIDDATA;
1953 static av_cold int xma_decode_end(AVCodecContext *avctx)
1955 XMADecodeCtx *s = avctx->priv_data;
1958 for (i = 0; i < s->num_streams; i++) {
1959 decode_end(&s->xma[i]);
1960 av_frame_free(&s->frames[i]);
1967 static void flush(WMAProDecodeCtx *s)
1970 /** reset output buffer as a part of it is used during the windowing of a
1972 for (i = 0; i < s->nb_channels; i++)
1973 memset(s->channel[i].out, 0, s->samples_per_frame *
1974 sizeof(*s->channel[i].out));
1976 s->skip_packets = 0;
1982 *@brief Clear decoder buffers (for seeking).
1983 *@param avctx codec context
1985 static void wmapro_flush(AVCodecContext *avctx)
1987 WMAProDecodeCtx *s = avctx->priv_data;
1992 static void xma_flush(AVCodecContext *avctx)
1994 XMADecodeCtx *s = avctx->priv_data;
1997 for (i = 0; i < s->num_streams; i++)
2000 memset(s->offset, 0, sizeof(s->offset));
2001 s->current_stream = 0;
2006 *@brief wmapro decoder
2008 const AVCodec ff_wmapro_decoder = {
2010 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
2011 .type = AVMEDIA_TYPE_AUDIO,
2012 .id = AV_CODEC_ID_WMAPRO,
2013 .priv_data_size = sizeof(WMAProDecodeCtx),
2014 .init = wmapro_decode_init,
2015 .close = wmapro_decode_end,
2016 .decode = wmapro_decode_packet,
2017 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
2018 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2019 .flush = wmapro_flush,
2020 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2021 AV_SAMPLE_FMT_NONE },
2024 const AVCodec ff_xma1_decoder = {
2026 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 1"),
2027 .type = AVMEDIA_TYPE_AUDIO,
2028 .id = AV_CODEC_ID_XMA1,
2029 .priv_data_size = sizeof(XMADecodeCtx),
2030 .init = xma_decode_init,
2031 .close = xma_decode_end,
2032 .decode = xma_decode_packet,
2033 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
2034 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2035 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2036 AV_SAMPLE_FMT_NONE },
2039 const AVCodec ff_xma2_decoder = {
2041 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 2"),
2042 .type = AVMEDIA_TYPE_AUDIO,
2043 .id = AV_CODEC_ID_XMA2,
2044 .priv_data_size = sizeof(XMADecodeCtx),
2045 .init = xma_decode_init,
2046 .close = xma_decode_end,
2047 .decode = xma_decode_packet,
2049 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
2050 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2051 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2052 AV_SAMPLE_FMT_NONE },