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/audio_fifo.h"
92 #include "libavutil/tx.h"
93 #include "libavutil/ffmath.h"
94 #include "libavutil/float_dsp.h"
95 #include "libavutil/intfloat.h"
96 #include "libavutil/intreadwrite.h"
97 #include "libavutil/mem_internal.h"
98 #include "libavutil/thread.h"
101 #include "codec_internal.h"
103 #include "get_bits.h"
104 #include "internal.h"
105 #include "put_bits.h"
106 #include "wmaprodata.h"
109 #include "wma_common.h"
111 /** current decoder limitations */
112 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
113 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
114 #define MAX_BANDS 29 ///< max number of scale factor bands
115 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
116 #define XMA_MAX_STREAMS 8
117 #define XMA_MAX_CHANNELS_STREAM 2
118 #define XMA_MAX_CHANNELS (XMA_MAX_STREAMS * XMA_MAX_CHANNELS_STREAM)
120 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
121 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
122 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
123 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
124 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
128 #define SCALEVLCBITS 8
129 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
130 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
131 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
132 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
133 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
135 static VLC sf_vlc; ///< scale factor DPCM vlc
136 static VLC sf_rl_vlc; ///< scale factor run length vlc
137 static VLC vec4_vlc; ///< 4 coefficients per symbol
138 static VLC vec2_vlc; ///< 2 coefficients per symbol
139 static VLC vec1_vlc; ///< 1 coefficient per symbol
140 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
141 static float sin64[33]; ///< sine table for decorrelation
144 * @brief frame specific decoder context for a single channel
146 typedef struct WMAProChannelCtx {
147 int16_t prev_block_len; ///< length of the previous block
148 uint8_t transmit_coefs;
149 uint8_t num_subframes;
150 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
151 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
152 uint8_t cur_subframe; ///< current subframe number
153 uint16_t decoded_samples; ///< number of already processed samples
154 uint8_t grouped; ///< channel is part of a group
155 int quant_step; ///< quantization step for the current subframe
156 int8_t reuse_sf; ///< share scale factors between subframes
157 int8_t scale_factor_step; ///< scaling step for the current subframe
158 int max_scale_factor; ///< maximum scale factor for the current subframe
159 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
160 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
161 int* scale_factors; ///< pointer to the scale factor values used for decoding
162 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
163 float* coeffs; ///< pointer to the subframe decode buffer
164 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
165 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
169 * @brief channel group for channel transformations
171 typedef struct WMAProChannelGrp {
172 uint8_t num_channels; ///< number of channels in the group
173 int8_t transform; ///< transform on / off
174 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
175 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
176 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
180 * @brief main decoder context
182 typedef struct WMAProDecodeCtx {
183 /* generic decoder variables */
184 AVCodecContext* avctx; ///< codec context for av_log
185 AVFloatDSPContext *fdsp;
186 uint8_t frame_data[MAX_FRAMESIZE +
187 AV_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
188 PutBitContext pb; ///< context for filling the frame_data buffer
189 AVTXContext *tx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
190 av_tx_fn tx_fn[WMAPRO_BLOCK_SIZES];
191 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
192 const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
194 /* frame size dependent frame information (set during initialization) */
195 uint32_t decode_flags; ///< used compression features
196 uint8_t len_prefix; ///< frame is prefixed with its length
197 uint8_t dynamic_range_compression; ///< frame contains DRC data
198 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
199 uint16_t samples_per_frame; ///< number of samples to output
200 uint16_t trim_start; ///< number of samples to skip at start
201 uint16_t trim_end; ///< number of samples to skip at end
202 uint16_t log2_frame_size;
203 int8_t lfe_channel; ///< lfe channel index
204 uint8_t max_num_subframes;
205 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
206 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
207 uint16_t min_samples_per_subframe;
208 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
209 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
210 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
211 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
213 /* packet decode state */
214 GetBitContext pgb; ///< bitstream reader context for the packet
215 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
216 uint8_t packet_offset; ///< frame offset in the packet
217 uint8_t packet_sequence_number; ///< current packet number
218 int num_saved_bits; ///< saved number of bits
219 int frame_offset; ///< frame offset in the bit reservoir
220 int subframe_offset; ///< subframe offset in the bit reservoir
221 uint8_t packet_loss; ///< set in case of bitstream error
222 uint8_t packet_done; ///< set when a packet is fully decoded
223 uint8_t eof_done; ///< set when EOF reached and extra subframe is written (XMA1/2)
225 /* frame decode state */
226 uint32_t frame_num; ///< current frame number (not used for decoding)
227 GetBitContext gb; ///< bitstream reader context
228 int buf_bit_size; ///< buffer size in bits
229 uint8_t drc_gain; ///< gain for the DRC tool
230 int8_t skip_frame; ///< skip output step
231 int8_t parsed_all_subframes; ///< all subframes decoded?
232 uint8_t skip_packets; ///< packets to skip to find next packet in a stream (XMA1/2)
234 /* subframe/block decode state */
235 int16_t subframe_len; ///< current subframe length
236 int8_t nb_channels; ///< number of channels in stream (XMA1/2)
237 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
238 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
239 int8_t num_bands; ///< number of scale factor bands
240 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
241 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
242 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
243 int8_t esc_len; ///< length of escaped coefficients
245 uint8_t num_chgroups; ///< number of channel groups
246 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
248 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
251 typedef struct XMADecodeCtx {
252 WMAProDecodeCtx xma[XMA_MAX_STREAMS];
253 AVFrame *frames[XMA_MAX_STREAMS];
256 AVAudioFifo *samples[2][XMA_MAX_STREAMS];
257 int start_channel[XMA_MAX_STREAMS];
258 int trim_start, trim_end;
263 *@brief helper function to print the most important members of the context
266 static av_cold void dump_context(WMAProDecodeCtx *s)
268 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
269 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
271 PRINT("ed sample bit depth", s->bits_per_sample);
272 PRINT_HEX("ed decode flags", s->decode_flags);
273 PRINT("samples per frame", s->samples_per_frame);
274 PRINT("log2 frame size", s->log2_frame_size);
275 PRINT("max num subframes", s->max_num_subframes);
276 PRINT("len prefix", s->len_prefix);
277 PRINT("num channels", s->nb_channels);
281 *@brief Uninitialize the decoder and free all resources.
282 *@param avctx codec context
283 *@return 0 on success, < 0 otherwise
285 static av_cold int decode_end(WMAProDecodeCtx *s)
291 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
292 av_tx_uninit(&s->tx[i]);
297 static av_cold int wmapro_decode_end(AVCodecContext *avctx)
299 WMAProDecodeCtx *s = avctx->priv_data;
306 static av_cold int get_rate(AVCodecContext *avctx)
308 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA?
309 if (avctx->sample_rate > 44100)
311 else if (avctx->sample_rate > 32000)
313 else if (avctx->sample_rate > 24000)
318 return avctx->sample_rate;
321 static av_cold void decode_init_static(void)
323 VLC_INIT_STATIC_FROM_LENGTHS(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
324 &scale_table[0][1], 2,
325 &scale_table[0][0], 2, 1, -60, 0, 616);
326 VLC_INIT_STATIC_FROM_LENGTHS(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
327 &scale_rl_table[0][1], 2,
328 &scale_rl_table[0][0], 2, 1, 0, 0, 1406);
329 VLC_INIT_STATIC_FROM_LENGTHS(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
331 coef0_syms, 2, 2, 0, 0, 2108);
332 VLC_INIT_STATIC_FROM_LENGTHS(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
333 &coef1_table[0][1], 2,
334 &coef1_table[0][0], 2, 1, 0, 0, 3912);
335 VLC_INIT_STATIC_FROM_LENGTHS(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
337 vec4_syms, 2, 2, -1, 0, 604);
338 VLC_INIT_STATIC_FROM_LENGTHS(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
339 &vec2_table[0][1], 2,
340 &vec2_table[0][0], 2, 1, -1, 0, 562);
341 VLC_INIT_STATIC_FROM_LENGTHS(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
342 &vec1_table[0][1], 2,
343 &vec1_table[0][0], 2, 1, 0, 0, 562);
345 /** calculate sine values for the decorrelation matrix */
346 for (int i = 0; i < 33; i++)
347 sin64[i] = sin(i * M_PI / 64.0);
349 for (int i = WMAPRO_BLOCK_MIN_BITS; i <= WMAPRO_BLOCK_MAX_BITS; i++)
350 ff_init_ff_sine_windows(i);
354 *@brief Initialize the decoder.
355 *@param avctx codec context
356 *@return 0 on success, -1 otherwise
358 static av_cold int decode_init(WMAProDecodeCtx *s, AVCodecContext *avctx, int num_stream)
360 static AVOnce init_static_once = AV_ONCE_INIT;
361 uint8_t *edata_ptr = avctx->extradata;
362 unsigned int channel_mask;
364 int log2_max_num_subframes;
365 int num_possible_block_sizes;
367 if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2)
368 avctx->block_align = 2048;
370 if (!avctx->block_align) {
371 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
372 return AVERROR(EINVAL);
377 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
379 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
381 /** dump the extradata */
382 av_log(avctx, AV_LOG_DEBUG, "extradata:\n");
383 for (i = 0; i < avctx->extradata_size; i++)
384 av_log(avctx, AV_LOG_DEBUG, "[%x] ", avctx->extradata[i]);
385 av_log(avctx, AV_LOG_DEBUG, "\n");
387 if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
388 s->decode_flags = 0x10d6;
389 s->bits_per_sample = 16;
390 channel_mask = 0; //AV_RL32(edata_ptr+2); /* not always in expected order */
391 if ((num_stream+1) * XMA_MAX_CHANNELS_STREAM > avctx->ch_layout.nb_channels) /* stream config is 2ch + 2ch + ... + 1/2ch */
395 } else if (avctx->codec_id == AV_CODEC_ID_XMA2) { /* XMA2WAVEFORMAT */
396 s->decode_flags = 0x10d6;
397 s->bits_per_sample = 16;
398 channel_mask = 0; /* would need to aggregate from all streams */
399 s->nb_channels = edata_ptr[32 + ((edata_ptr[0]==3)?0:8) + 4*num_stream + 0]; /* nth stream config */
400 } else if (avctx->codec_id == AV_CODEC_ID_XMA1) { /* XMAWAVEFORMAT */
401 s->decode_flags = 0x10d6;
402 s->bits_per_sample = 16;
403 channel_mask = 0; /* would need to aggregate from all streams */
404 s->nb_channels = edata_ptr[8 + 20*num_stream + 17]; /* nth stream config */
405 } else if (avctx->codec_id == AV_CODEC_ID_WMAPRO && avctx->extradata_size >= 18) {
406 s->decode_flags = AV_RL16(edata_ptr+14);
407 channel_mask = AV_RL32(edata_ptr+2);
408 s->bits_per_sample = AV_RL16(edata_ptr);
409 s->nb_channels = channel_mask ? av_popcount(channel_mask) : avctx->ch_layout.nb_channels;
411 if (s->bits_per_sample > 32 || s->bits_per_sample < 1) {
412 avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample);
413 return AVERROR_PATCHWELCOME;
416 avpriv_request_sample(avctx, "Unknown extradata size");
417 return AVERROR_PATCHWELCOME;
421 s->log2_frame_size = av_log2(avctx->block_align) + 4;
422 if (s->log2_frame_size > 25) {
423 avpriv_request_sample(avctx, "Large block align");
424 return AVERROR_PATCHWELCOME;
428 s->skip_frame = 1; /* skip first frame */
431 s->len_prefix = (s->decode_flags & 0x40);
434 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
435 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
436 if (bits > WMAPRO_BLOCK_MAX_BITS) {
437 avpriv_request_sample(avctx, "14-bit block sizes");
438 return AVERROR_PATCHWELCOME;
440 s->samples_per_frame = 1 << bits;
442 s->samples_per_frame = 512;
446 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
447 s->max_num_subframes = 1 << log2_max_num_subframes;
448 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
449 s->max_subframe_len_bit = 1;
450 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
452 num_possible_block_sizes = log2_max_num_subframes + 1;
453 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
454 s->dynamic_range_compression = (s->decode_flags & 0x80);
456 if (s->max_num_subframes > MAX_SUBFRAMES) {
457 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
458 s->max_num_subframes);
459 return AVERROR_INVALIDDATA;
462 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
463 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
464 s->min_samples_per_subframe);
465 return AVERROR_INVALIDDATA;
468 if (s->avctx->sample_rate <= 0) {
469 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
470 return AVERROR_INVALIDDATA;
473 if (s->nb_channels <= 0) {
474 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
476 return AVERROR_INVALIDDATA;
477 } else if (avctx->codec_id != AV_CODEC_ID_WMAPRO && s->nb_channels > XMA_MAX_CHANNELS_STREAM) {
478 av_log(avctx, AV_LOG_ERROR, "invalid number of channels per XMA stream %d\n",
480 return AVERROR_INVALIDDATA;
481 } else if (s->nb_channels > WMAPRO_MAX_CHANNELS || s->nb_channels > avctx->ch_layout.nb_channels) {
482 avpriv_request_sample(avctx,
483 "More than %d channels", WMAPRO_MAX_CHANNELS);
484 return AVERROR_PATCHWELCOME;
487 /** init previous block len */
488 for (i = 0; i < s->nb_channels; i++)
489 s->channel[i].prev_block_len = s->samples_per_frame;
491 /** extract lfe channel position */
494 if (channel_mask & 8) {
496 for (mask = 1; mask < 16; mask <<= 1) {
497 if (channel_mask & mask)
502 /** calculate number of scale factor bands and their offsets
503 for every possible block size */
504 for (i = 0; i < num_possible_block_sizes; i++) {
505 int subframe_len = s->samples_per_frame >> i;
508 int rate = get_rate(avctx);
510 s->sfb_offsets[i][0] = 0;
512 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
513 int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2;
515 if (offset > s->sfb_offsets[i][band - 1])
516 s->sfb_offsets[i][band++] = offset;
518 if (offset >= subframe_len)
521 s->sfb_offsets[i][band - 1] = subframe_len;
522 s->num_sfb[i] = band - 1;
523 if (s->num_sfb[i] <= 0) {
524 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
525 return AVERROR_INVALIDDATA;
530 /** Scale factors can be shared between blocks of different size
531 as every block has a different scale factor band layout.
532 The matrix sf_offsets is needed to find the correct scale factor.
535 for (i = 0; i < num_possible_block_sizes; i++) {
537 for (b = 0; b < s->num_sfb[i]; b++) {
539 int offset = ((s->sfb_offsets[i][b]
540 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
541 for (x = 0; x < num_possible_block_sizes; x++) {
543 while (s->sfb_offsets[x][v + 1] << x < offset) {
545 av_assert0(v < MAX_BANDS);
547 s->sf_offsets[i][x][b] = v;
552 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
554 return AVERROR(ENOMEM);
556 /** init MDCT, FIXME: only init needed sizes */
557 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
558 const float scale = 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
559 / (1ll << (s->bits_per_sample - 1));
560 int err = av_tx_init(&s->tx[i], &s->tx_fn[i], AV_TX_FLOAT_MDCT, 1,
561 1 << (WMAPRO_BLOCK_MIN_BITS + i), &scale, 0);
566 /** init MDCT windows: simple sine window */
567 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
568 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
569 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
572 /** calculate subwoofer cutoff values */
573 for (i = 0; i < num_possible_block_sizes; i++) {
574 int block_size = s->samples_per_frame >> i;
575 int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1)
576 / s->avctx->sample_rate;
577 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
580 if (avctx->debug & FF_DEBUG_BITSTREAM)
583 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
585 av_channel_layout_uninit(&avctx->ch_layout);
586 av_channel_layout_from_mask(&avctx->ch_layout, channel_mask);
588 avctx->ch_layout.order = AV_CHANNEL_ORDER_UNSPEC;
591 ff_thread_once(&init_static_once, decode_init_static);
597 *@brief Initialize the decoder.
598 *@param avctx codec context
599 *@return 0 on success, -1 otherwise
601 static av_cold int wmapro_decode_init(AVCodecContext *avctx)
603 WMAProDecodeCtx *s = avctx->priv_data;
605 return decode_init(s, avctx, 0);
609 *@brief Decode the subframe length.
611 *@param offset sample offset in the frame
612 *@return decoded subframe length on success, < 0 in case of an error
614 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
616 int frame_len_shift = 0;
619 /** no need to read from the bitstream when only one length is possible */
620 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
621 return s->min_samples_per_subframe;
623 if (get_bits_left(&s->gb) < 1)
624 return AVERROR_INVALIDDATA;
626 /** 1 bit indicates if the subframe is of maximum length */
627 if (s->max_subframe_len_bit) {
628 if (get_bits1(&s->gb))
629 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
631 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
633 subframe_len = s->samples_per_frame >> frame_len_shift;
635 /** sanity check the length */
636 if (subframe_len < s->min_samples_per_subframe ||
637 subframe_len > s->samples_per_frame) {
638 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
640 return AVERROR_INVALIDDATA;
646 *@brief Decode how the data in the frame is split into subframes.
647 * Every WMA frame contains the encoded data for a fixed number of
648 * samples per channel. The data for every channel might be split
649 * into several subframes. This function will reconstruct the list of
650 * subframes for every channel.
652 * If the subframes are not evenly split, the algorithm estimates the
653 * channels with the lowest number of total samples.
654 * Afterwards, for each of these channels a bit is read from the
655 * bitstream that indicates if the channel contains a subframe with the
656 * next subframe size that is going to be read from the bitstream or not.
657 * If a channel contains such a subframe, the subframe size gets added to
658 * the channel's subframe list.
659 * The algorithm repeats these steps until the frame is properly divided
660 * between the individual channels.
663 *@return 0 on success, < 0 in case of an error
665 static int decode_tilehdr(WMAProDecodeCtx *s)
667 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
668 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
669 int channels_for_cur_subframe = s->nb_channels; /**< number of channels that contain the current subframe */
670 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
671 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
674 /* Should never consume more than 3073 bits (256 iterations for the
675 * while loop when always the minimum amount of 128 samples is subtracted
676 * from missing samples in the 8 channel case).
677 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
680 /** reset tiling information */
681 for (c = 0; c < s->nb_channels; c++)
682 s->channel[c].num_subframes = 0;
684 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
685 fixed_channel_layout = 1;
687 /** loop until the frame data is split between the subframes */
691 /** check which channels contain the subframe */
692 for (c = 0; c < s->nb_channels; c++) {
693 if (num_samples[c] == min_channel_len) {
694 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
695 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
696 contains_subframe[c] = 1;
698 contains_subframe[c] = get_bits1(&s->gb);
700 contains_subframe[c] = 0;
703 /** get subframe length, subframe_len == 0 is not allowed */
704 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
705 return AVERROR_INVALIDDATA;
707 /** add subframes to the individual channels and find new min_channel_len */
708 min_channel_len += subframe_len;
709 for (c = 0; c < s->nb_channels; c++) {
710 WMAProChannelCtx* chan = &s->channel[c];
712 if (contains_subframe[c]) {
713 if (chan->num_subframes >= MAX_SUBFRAMES) {
714 av_log(s->avctx, AV_LOG_ERROR,
715 "broken frame: num subframes > 31\n");
716 return AVERROR_INVALIDDATA;
718 chan->subframe_len[chan->num_subframes] = subframe_len;
719 num_samples[c] += subframe_len;
720 ++chan->num_subframes;
721 if (num_samples[c] > s->samples_per_frame) {
722 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
723 "channel len > samples_per_frame\n");
724 return AVERROR_INVALIDDATA;
726 } else if (num_samples[c] <= min_channel_len) {
727 if (num_samples[c] < min_channel_len) {
728 channels_for_cur_subframe = 0;
729 min_channel_len = num_samples[c];
731 ++channels_for_cur_subframe;
734 } while (min_channel_len < s->samples_per_frame);
736 for (c = 0; c < s->nb_channels; c++) {
739 for (i = 0; i < s->channel[c].num_subframes; i++) {
740 ff_dlog(s->avctx, "frame[%"PRIu32"] channel[%i] subframe[%i]"
741 " len %i\n", s->frame_num, c, i,
742 s->channel[c].subframe_len[i]);
743 s->channel[c].subframe_offset[i] = offset;
744 offset += s->channel[c].subframe_len[i];
752 *@brief Calculate a decorrelation matrix from the bitstream parameters.
753 *@param s codec context
754 *@param chgroup channel group for which the matrix needs to be calculated
756 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
757 WMAProChannelGrp *chgroup)
761 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
762 memset(chgroup->decorrelation_matrix, 0, s->nb_channels *
763 s->nb_channels * sizeof(*chgroup->decorrelation_matrix));
765 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
766 rotation_offset[i] = get_bits(&s->gb, 6);
768 for (i = 0; i < chgroup->num_channels; i++)
769 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
770 get_bits1(&s->gb) ? 1.0 : -1.0;
772 for (i = 1; i < chgroup->num_channels; i++) {
774 for (x = 0; x < i; x++) {
776 for (y = 0; y < i + 1; y++) {
777 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
778 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
779 int n = rotation_offset[offset + x];
785 cosv = sin64[32 - n];
787 sinv = sin64[64 - n];
788 cosv = -sin64[n - 32];
791 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
792 (v1 * sinv) - (v2 * cosv);
793 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
794 (v1 * cosv) + (v2 * sinv);
802 *@brief Decode channel transformation parameters
803 *@param s codec context
804 *@return >= 0 in case of success, < 0 in case of bitstream errors
806 static int decode_channel_transform(WMAProDecodeCtx* s)
809 /* should never consume more than 1921 bits for the 8 channel case
810 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
811 * + MAX_CHANNELS + MAX_BANDS + 1)
814 /** in the one channel case channel transforms are pointless */
816 if (s->nb_channels > 1) {
817 int remaining_channels = s->channels_for_cur_subframe;
819 if (get_bits1(&s->gb)) {
820 avpriv_request_sample(s->avctx,
821 "Channel transform bit");
822 return AVERROR_PATCHWELCOME;
825 for (s->num_chgroups = 0; remaining_channels &&
826 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
827 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
828 float** channel_data = chgroup->channel_data;
829 chgroup->num_channels = 0;
830 chgroup->transform = 0;
832 /** decode channel mask */
833 if (remaining_channels > 2) {
834 for (i = 0; i < s->channels_for_cur_subframe; i++) {
835 int channel_idx = s->channel_indexes_for_cur_subframe[i];
836 if (!s->channel[channel_idx].grouped
837 && get_bits1(&s->gb)) {
838 ++chgroup->num_channels;
839 s->channel[channel_idx].grouped = 1;
840 *channel_data++ = s->channel[channel_idx].coeffs;
844 chgroup->num_channels = remaining_channels;
845 for (i = 0; i < s->channels_for_cur_subframe; i++) {
846 int channel_idx = s->channel_indexes_for_cur_subframe[i];
847 if (!s->channel[channel_idx].grouped)
848 *channel_data++ = s->channel[channel_idx].coeffs;
849 s->channel[channel_idx].grouped = 1;
853 /** decode transform type */
854 if (chgroup->num_channels == 2) {
855 if (get_bits1(&s->gb)) {
856 if (get_bits1(&s->gb)) {
857 avpriv_request_sample(s->avctx,
858 "Unknown channel transform type");
859 return AVERROR_PATCHWELCOME;
862 chgroup->transform = 1;
863 if (s->nb_channels == 2) {
864 chgroup->decorrelation_matrix[0] = 1.0;
865 chgroup->decorrelation_matrix[1] = -1.0;
866 chgroup->decorrelation_matrix[2] = 1.0;
867 chgroup->decorrelation_matrix[3] = 1.0;
870 chgroup->decorrelation_matrix[0] = 0.70703125;
871 chgroup->decorrelation_matrix[1] = -0.70703125;
872 chgroup->decorrelation_matrix[2] = 0.70703125;
873 chgroup->decorrelation_matrix[3] = 0.70703125;
876 } else if (chgroup->num_channels > 2) {
877 if (get_bits1(&s->gb)) {
878 chgroup->transform = 1;
879 if (get_bits1(&s->gb)) {
880 decode_decorrelation_matrix(s, chgroup);
882 /** FIXME: more than 6 coupled channels not supported */
883 if (chgroup->num_channels > 6) {
884 avpriv_request_sample(s->avctx,
885 "Coupled channels > 6");
887 memcpy(chgroup->decorrelation_matrix,
888 default_decorrelation[chgroup->num_channels],
889 chgroup->num_channels * chgroup->num_channels *
890 sizeof(*chgroup->decorrelation_matrix));
896 /** decode transform on / off */
897 if (chgroup->transform) {
898 if (!get_bits1(&s->gb)) {
900 /** transform can be enabled for individual bands */
901 for (i = 0; i < s->num_bands; i++) {
902 chgroup->transform_band[i] = get_bits1(&s->gb);
905 memset(chgroup->transform_band, 1, s->num_bands);
908 remaining_channels -= chgroup->num_channels;
915 *@brief Extract the coefficients from the bitstream.
916 *@param s codec context
917 *@param c current channel number
918 *@return 0 on success, < 0 in case of bitstream errors
920 static int decode_coeffs(WMAProDecodeCtx *s, int c)
922 /* Integers 0..15 as single-precision floats. The table saves a
923 costly int to float conversion, and storing the values as
924 integers allows fast sign-flipping. */
925 static const uint32_t fval_tab[16] = {
926 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
927 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
928 0x41000000, 0x41100000, 0x41200000, 0x41300000,
929 0x41400000, 0x41500000, 0x41600000, 0x41700000,
933 WMAProChannelCtx* ci = &s->channel[c];
940 ff_dlog(s->avctx, "decode coefficients for channel %i\n", c);
942 vlctable = get_bits1(&s->gb);
943 vlc = &coef_vlc[vlctable];
953 /** decode vector coefficients (consumes up to 167 bits per iteration for
954 4 vector coded large values) */
955 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
956 (cur_coeff + 3 < ci->num_vec_coeffs)) {
961 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
964 for (i = 0; i < 4; i += 2) {
965 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
968 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
969 if (v0 == HUFF_VEC1_SIZE - 1)
970 v0 += ff_wma_get_large_val(&s->gb);
971 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
972 if (v1 == HUFF_VEC1_SIZE - 1)
973 v1 += ff_wma_get_large_val(&s->gb);
974 vals[i ] = av_float2int(v0);
975 vals[i+1] = av_float2int(v1);
977 vals[i] = fval_tab[idx >> 4 ];
978 vals[i+1] = fval_tab[idx & 0xF];
982 vals[0] = fval_tab[ idx >> 12 ];
983 vals[1] = fval_tab[(idx >> 8) & 0xF];
984 vals[2] = fval_tab[(idx >> 4) & 0xF];
985 vals[3] = fval_tab[ idx & 0xF];
989 for (i = 0; i < 4; i++) {
991 uint32_t sign = get_bits1(&s->gb) - 1;
992 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
995 ci->coeffs[cur_coeff] = 0;
996 /** switch to run level mode when subframe_len / 128 zeros
997 were found in a row */
998 rl_mode |= (++num_zeros > s->subframe_len >> 8);
1004 /** decode run level coded coefficients */
1005 if (cur_coeff < s->subframe_len) {
1008 memset(&ci->coeffs[cur_coeff], 0,
1009 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
1010 ret = ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
1011 level, run, 1, ci->coeffs,
1012 cur_coeff, s->subframe_len,
1013 s->subframe_len, s->esc_len, 0);
1022 *@brief Extract scale factors from the bitstream.
1023 *@param s codec context
1024 *@return 0 on success, < 0 in case of bitstream errors
1026 static int decode_scale_factors(WMAProDecodeCtx* s)
1030 /** should never consume more than 5344 bits
1031 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
1034 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1035 int c = s->channel_indexes_for_cur_subframe[i];
1038 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
1039 sf_end = s->channel[c].scale_factors + s->num_bands;
1041 /** resample scale factors for the new block size
1042 * as the scale factors might need to be resampled several times
1043 * before some new values are transmitted, a backup of the last
1044 * transmitted scale factors is kept in saved_scale_factors
1046 if (s->channel[c].reuse_sf) {
1047 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
1049 for (b = 0; b < s->num_bands; b++)
1050 s->channel[c].scale_factors[b] =
1051 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
1054 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
1056 if (!s->channel[c].reuse_sf) {
1058 /** decode DPCM coded scale factors */
1059 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
1060 val = 45 / s->channel[c].scale_factor_step;
1061 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
1062 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH);
1067 /** run level decode differences to the resampled factors */
1068 for (i = 0; i < s->num_bands; i++) {
1074 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
1077 uint32_t code = get_bits(&s->gb, 14);
1079 sign = (code & 1) - 1;
1080 skip = (code & 0x3f) >> 1;
1081 } else if (idx == 1) {
1084 skip = scale_rl_run[idx];
1085 val = scale_rl_level[idx];
1086 sign = get_bits1(&s->gb)-1;
1090 if (i >= s->num_bands) {
1091 av_log(s->avctx, AV_LOG_ERROR,
1092 "invalid scale factor coding\n");
1093 return AVERROR_INVALIDDATA;
1095 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
1099 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
1100 s->channel[c].table_idx = s->table_idx;
1101 s->channel[c].reuse_sf = 1;
1104 /** calculate new scale factor maximum */
1105 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
1106 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
1107 s->channel[c].max_scale_factor =
1108 FFMAX(s->channel[c].max_scale_factor, *sf);
1116 *@brief Reconstruct the individual channel data.
1117 *@param s codec context
1119 static void inverse_channel_transform(WMAProDecodeCtx *s)
1123 for (i = 0; i < s->num_chgroups; i++) {
1124 if (s->chgroup[i].transform) {
1125 float data[WMAPRO_MAX_CHANNELS];
1126 const int num_channels = s->chgroup[i].num_channels;
1127 float** ch_data = s->chgroup[i].channel_data;
1128 float** ch_end = ch_data + num_channels;
1129 const int8_t* tb = s->chgroup[i].transform_band;
1132 /** multichannel decorrelation */
1133 for (sfb = s->cur_sfb_offsets;
1134 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1137 /** multiply values with the decorrelation_matrix */
1138 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1139 const float* mat = s->chgroup[i].decorrelation_matrix;
1140 const float* data_end = data + num_channels;
1141 float* data_ptr = data;
1144 for (ch = ch_data; ch < ch_end; ch++)
1145 *data_ptr++ = (*ch)[y];
1147 for (ch = ch_data; ch < ch_end; ch++) {
1150 while (data_ptr < data_end)
1151 sum += *data_ptr++ * *mat++;
1156 } else if (s->nb_channels == 2) {
1157 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1158 s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0],
1159 ch_data[0] + sfb[0],
1161 s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0],
1162 ch_data[1] + sfb[0],
1171 *@brief Apply sine window and reconstruct the output buffer.
1172 *@param s codec context
1174 static void wmapro_window(WMAProDecodeCtx *s)
1177 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1178 int c = s->channel_indexes_for_cur_subframe[i];
1179 const float* window;
1180 int winlen = s->channel[c].prev_block_len;
1181 float* start = s->channel[c].coeffs - (winlen >> 1);
1183 if (s->subframe_len < winlen) {
1184 start += (winlen - s->subframe_len) >> 1;
1185 winlen = s->subframe_len;
1188 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1192 s->fdsp->vector_fmul_window(start, start, start + winlen,
1195 s->channel[c].prev_block_len = s->subframe_len;
1200 *@brief Decode a single subframe (block).
1201 *@param s codec context
1202 *@return 0 on success, < 0 when decoding failed
1204 static int decode_subframe(WMAProDecodeCtx *s)
1206 int offset = s->samples_per_frame;
1207 int subframe_len = s->samples_per_frame;
1209 int total_samples = s->samples_per_frame * s->nb_channels;
1210 int transmit_coeffs = 0;
1211 int cur_subwoofer_cutoff;
1213 s->subframe_offset = get_bits_count(&s->gb);
1215 /** reset channel context and find the next block offset and size
1216 == the next block of the channel with the smallest number of
1219 for (i = 0; i < s->nb_channels; i++) {
1220 s->channel[i].grouped = 0;
1221 if (offset > s->channel[i].decoded_samples) {
1222 offset = s->channel[i].decoded_samples;
1224 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1229 "processing subframe with offset %i len %i\n", offset, subframe_len);
1231 /** get a list of all channels that contain the estimated block */
1232 s->channels_for_cur_subframe = 0;
1233 for (i = 0; i < s->nb_channels; i++) {
1234 const int cur_subframe = s->channel[i].cur_subframe;
1235 /** subtract already processed samples */
1236 total_samples -= s->channel[i].decoded_samples;
1238 /** and count if there are multiple subframes that match our profile */
1239 if (offset == s->channel[i].decoded_samples &&
1240 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1241 total_samples -= s->channel[i].subframe_len[cur_subframe];
1242 s->channel[i].decoded_samples +=
1243 s->channel[i].subframe_len[cur_subframe];
1244 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1245 ++s->channels_for_cur_subframe;
1249 /** check if the frame will be complete after processing the
1252 s->parsed_all_subframes = 1;
1255 ff_dlog(s->avctx, "subframe is part of %i channels\n",
1256 s->channels_for_cur_subframe);
1258 /** calculate number of scale factor bands and their offsets */
1259 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1260 s->num_bands = s->num_sfb[s->table_idx];
1261 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1262 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1264 /** configure the decoder for the current subframe */
1265 offset += s->samples_per_frame >> 1;
1267 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1268 int c = s->channel_indexes_for_cur_subframe[i];
1270 s->channel[c].coeffs = &s->channel[c].out[offset];
1273 s->subframe_len = subframe_len;
1274 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1276 /** skip extended header if any */
1277 if (get_bits1(&s->gb)) {
1279 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1280 int len = get_bits(&s->gb, 4);
1281 num_fill_bits = get_bitsz(&s->gb, len) + 1;
1284 if (num_fill_bits >= 0) {
1285 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1286 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1287 return AVERROR_INVALIDDATA;
1290 skip_bits_long(&s->gb, num_fill_bits);
1294 /** no idea for what the following bit is used */
1295 if (get_bits1(&s->gb)) {
1296 avpriv_request_sample(s->avctx, "Reserved bit");
1297 return AVERROR_PATCHWELCOME;
1301 if (decode_channel_transform(s) < 0)
1302 return AVERROR_INVALIDDATA;
1305 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1306 int c = s->channel_indexes_for_cur_subframe[i];
1307 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1308 transmit_coeffs = 1;
1311 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1312 if (transmit_coeffs) {
1314 int quant_step = 90 * s->bits_per_sample >> 4;
1316 /** decode number of vector coded coefficients */
1317 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1318 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1319 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1320 int c = s->channel_indexes_for_cur_subframe[i];
1321 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1322 if (num_vec_coeffs > s->subframe_len) {
1323 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1324 return AVERROR_INVALIDDATA;
1326 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1327 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1330 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1331 int c = s->channel_indexes_for_cur_subframe[i];
1332 s->channel[c].num_vec_coeffs = s->subframe_len;
1335 /** decode quantization step */
1336 step = get_sbits(&s->gb, 6);
1338 if (step == -32 || step == 31) {
1339 const int sign = (step == 31) - 1;
1341 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1342 (step = get_bits(&s->gb, 5)) == 31) {
1345 quant_step += ((quant + step) ^ sign) - sign;
1347 if (quant_step < 0) {
1348 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1351 /** decode quantization step modifiers for every channel */
1353 if (s->channels_for_cur_subframe == 1) {
1354 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1356 int modifier_len = get_bits(&s->gb, 3);
1357 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1358 int c = s->channel_indexes_for_cur_subframe[i];
1359 s->channel[c].quant_step = quant_step;
1360 if (get_bits1(&s->gb)) {
1362 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1364 ++s->channel[c].quant_step;
1369 /** decode scale factors */
1370 if (decode_scale_factors(s) < 0)
1371 return AVERROR_INVALIDDATA;
1374 ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1375 get_bits_count(&s->gb) - s->subframe_offset);
1377 /** parse coefficients */
1378 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1379 int c = s->channel_indexes_for_cur_subframe[i];
1380 if (s->channel[c].transmit_coefs &&
1381 get_bits_count(&s->gb) < s->num_saved_bits) {
1382 decode_coeffs(s, c);
1384 memset(s->channel[c].coeffs, 0,
1385 sizeof(*s->channel[c].coeffs) * subframe_len);
1388 ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1389 get_bits_count(&s->gb) - s->subframe_offset);
1391 if (transmit_coeffs) {
1392 AVTXContext *tx = s->tx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1393 av_tx_fn tx_fn = s->tx_fn[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1394 /** reconstruct the per channel data */
1395 inverse_channel_transform(s);
1396 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1397 int c = s->channel_indexes_for_cur_subframe[i];
1398 const int* sf = s->channel[c].scale_factors;
1401 if (c == s->lfe_channel)
1402 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1403 (subframe_len - cur_subwoofer_cutoff));
1405 /** inverse quantization and rescaling */
1406 for (b = 0; b < s->num_bands; b++) {
1407 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1408 const int exp = s->channel[c].quant_step -
1409 (s->channel[c].max_scale_factor - *sf++) *
1410 s->channel[c].scale_factor_step;
1411 const float quant = ff_exp10(exp / 20.0);
1412 int start = s->cur_sfb_offsets[b];
1413 s->fdsp->vector_fmul_scalar(s->tmp + start,
1414 s->channel[c].coeffs + start,
1415 quant, end - start);
1418 /** apply imdct (imdct_half == DCTIV with reverse) */
1419 tx_fn(tx, s->channel[c].coeffs, s->tmp, sizeof(float));
1423 /** window and overlapp-add */
1426 /** handled one subframe */
1427 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1428 int c = s->channel_indexes_for_cur_subframe[i];
1429 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1430 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1431 return AVERROR_INVALIDDATA;
1433 ++s->channel[c].cur_subframe;
1440 *@brief Decode one WMA frame.
1441 *@param s codec context
1442 *@return 0 if the trailer bit indicates that this is the last frame,
1443 * 1 if there are additional frames
1445 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1447 GetBitContext* gb = &s->gb;
1448 int more_frames = 0;
1452 /** get frame length */
1454 len = get_bits(gb, s->log2_frame_size);
1456 ff_dlog(s->avctx, "decoding frame with length %x\n", len);
1458 /** decode tile information */
1459 if (decode_tilehdr(s)) {
1464 /** read postproc transform */
1465 if (s->nb_channels > 1 && get_bits1(gb)) {
1466 if (get_bits1(gb)) {
1467 for (i = 0; i < s->nb_channels * s->nb_channels; i++)
1472 /** read drc info */
1473 if (s->dynamic_range_compression) {
1474 s->drc_gain = get_bits(gb, 8);
1475 ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1478 if (get_bits1(gb)) {
1480 s->trim_start = get_bits(gb, av_log2(s->samples_per_frame * 2));
1483 s->trim_end = get_bits(gb, av_log2(s->samples_per_frame * 2));
1485 s->trim_start = s->trim_end = 0;
1488 ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1489 get_bits_count(gb) - s->frame_offset);
1491 /** reset subframe states */
1492 s->parsed_all_subframes = 0;
1493 for (i = 0; i < s->nb_channels; i++) {
1494 s->channel[i].decoded_samples = 0;
1495 s->channel[i].cur_subframe = 0;
1496 s->channel[i].reuse_sf = 0;
1499 /** decode all subframes */
1500 while (!s->parsed_all_subframes) {
1501 if (decode_subframe(s) < 0) {
1507 /** copy samples to the output buffer */
1508 for (i = 0; i < s->nb_channels; i++)
1509 memcpy(frame->extended_data[i], s->channel[i].out,
1510 s->samples_per_frame * sizeof(*s->channel[i].out));
1512 for (i = 0; i < s->nb_channels; i++) {
1513 /** reuse second half of the IMDCT output for the next frame */
1514 memcpy(&s->channel[i].out[0],
1515 &s->channel[i].out[s->samples_per_frame],
1516 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1519 if (s->skip_frame) {
1522 av_frame_unref(frame);
1527 if (s->len_prefix) {
1528 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1529 /** FIXME: not sure if this is always an error */
1530 av_log(s->avctx, AV_LOG_ERROR,
1531 "frame[%"PRIu32"] would have to skip %i bits\n",
1533 len - (get_bits_count(gb) - s->frame_offset) - 1);
1538 /** skip the rest of the frame data */
1539 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1541 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1545 /** decode trailer bit */
1546 more_frames = get_bits1(gb);
1553 *@brief Calculate remaining input buffer length.
1554 *@param s codec context
1555 *@param gb bitstream reader context
1556 *@return remaining size in bits
1558 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1560 return s->buf_bit_size - get_bits_count(gb);
1564 *@brief Fill the bit reservoir with a (partial) frame.
1565 *@param s codec context
1566 *@param gb bitstream reader context
1567 *@param len length of the partial frame
1568 *@param append decides whether to reset the buffer or not
1570 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1575 /** when the frame data does not need to be concatenated, the input buffer
1576 is reset and additional bits from the previous frame are copied
1577 and skipped later so that a fast byte copy is possible */
1580 s->frame_offset = get_bits_count(gb) & 7;
1581 s->num_saved_bits = s->frame_offset;
1582 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1583 buflen = (s->num_saved_bits + len + 7) >> 3;
1585 buflen = (put_bits_count(&s->pb) + len + 7) >> 3;
1587 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1588 avpriv_request_sample(s->avctx, "Too small input buffer");
1593 av_assert0(len <= put_bits_left(&s->pb));
1595 s->num_saved_bits += len;
1597 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1600 int align = 8 - (get_bits_count(gb) & 7);
1601 align = FFMIN(align, len);
1602 put_bits(&s->pb, align, get_bits(gb, align));
1604 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1606 skip_bits_long(gb, len);
1609 PutBitContext tmp = s->pb;
1610 flush_put_bits(&tmp);
1613 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1614 skip_bits(&s->gb, s->frame_offset);
1617 static int decode_packet(AVCodecContext *avctx, WMAProDecodeCtx *s,
1618 AVFrame *frame, int *got_frame_ptr, AVPacket *avpkt)
1620 GetBitContext* gb = &s->pgb;
1621 const uint8_t* buf = avpkt->data;
1622 int buf_size = avpkt->size;
1623 int num_bits_prev_frame;
1624 int packet_sequence_number;
1632 /** Must output remaining samples after stream end. WMAPRO 5.1 created
1633 * by XWMA encoder don't though (maybe only 1/2ch streams need it). */
1638 /** clean output buffer and copy last IMDCT samples */
1639 for (i = 0; i < s->nb_channels; i++) {
1640 memset(frame->extended_data[i], 0,
1641 s->samples_per_frame * sizeof(*s->channel[i].out));
1643 memcpy(frame->extended_data[i], s->channel[i].out,
1644 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1652 else if (s->packet_done || s->packet_loss) {
1655 /** sanity check for the buffer length */
1656 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) {
1657 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1658 buf_size, avctx->block_align);
1660 return AVERROR_INVALIDDATA;
1663 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1664 s->next_packet_start = buf_size - avctx->block_align;
1665 buf_size = avctx->block_align;
1667 s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align);
1668 buf_size = FFMIN(buf_size, avctx->block_align);
1670 s->buf_bit_size = buf_size << 3;
1672 /** parse packet header */
1673 ret = init_get_bits8(gb, buf, buf_size);
1676 if (avctx->codec_id != AV_CODEC_ID_XMA2) {
1677 packet_sequence_number = get_bits(gb, 4);
1680 int num_frames = get_bits(gb, 6);
1681 ff_dlog(avctx, "packet[%"PRId64"]: number of frames %d\n", avctx->frame_num, num_frames);
1682 packet_sequence_number = 0;
1685 /** get number of bits that need to be added to the previous frame */
1686 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1687 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) {
1689 s->skip_packets = get_bits(gb, 8);
1690 ff_dlog(avctx, "packet[%"PRId64"]: skip packets %d\n", avctx->frame_num, s->skip_packets);
1693 ff_dlog(avctx, "packet[%"PRId64"]: nbpf %x\n", avctx->frame_num,
1694 num_bits_prev_frame);
1696 /** check for packet loss */
1697 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && !s->packet_loss &&
1698 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1700 av_log(avctx, AV_LOG_ERROR,
1701 "Packet loss detected! seq %"PRIx8" vs %x\n",
1702 s->packet_sequence_number, packet_sequence_number);
1704 s->packet_sequence_number = packet_sequence_number;
1706 if (num_bits_prev_frame > 0) {
1707 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1708 if (num_bits_prev_frame >= remaining_packet_bits) {
1709 num_bits_prev_frame = remaining_packet_bits;
1713 /** append the previous frame data to the remaining data from the
1714 previous packet to create a full frame */
1715 save_bits(s, gb, num_bits_prev_frame, 1);
1716 ff_dlog(avctx, "accumulated %x bits of frame data\n",
1717 s->num_saved_bits - s->frame_offset);
1719 /** decode the cross packet frame if it is valid */
1720 if (!s->packet_loss)
1721 decode_frame(s, frame, got_frame_ptr);
1722 } else if (s->num_saved_bits - s->frame_offset) {
1723 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1724 s->num_saved_bits - s->frame_offset);
1727 if (s->packet_loss) {
1728 /** reset number of saved bits so that the decoder
1729 does not start to decode incomplete frames in the
1730 s->len_prefix == 0 case */
1731 s->num_saved_bits = 0;
1737 if (avpkt->size < s->next_packet_start) {
1739 return AVERROR_INVALIDDATA;
1742 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1743 ret = init_get_bits8(gb, avpkt->data, avpkt->size - s->next_packet_start);
1746 skip_bits(gb, s->packet_offset);
1747 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1748 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1749 frame_size <= remaining_bits(s, gb)) {
1750 save_bits(s, gb, frame_size, 0);
1751 if (!s->packet_loss)
1752 s->packet_done = !decode_frame(s, frame, got_frame_ptr);
1753 } else if (!s->len_prefix
1754 && s->num_saved_bits > get_bits_count(&s->gb)) {
1755 /** when the frames do not have a length prefix, we don't know
1756 the compressed length of the individual frames
1757 however, we know what part of a new packet belongs to the
1759 therefore we save the incoming packet first, then we append
1760 the "previous frame" data from the next packet so that
1761 we get a buffer that only contains full frames */
1762 s->packet_done = !decode_frame(s, frame, got_frame_ptr);
1768 if (remaining_bits(s, gb) < 0) {
1769 av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1773 if (s->packet_done && !s->packet_loss &&
1774 remaining_bits(s, gb) > 0) {
1775 /** save the rest of the data so that it can be decoded
1776 with the next packet */
1777 save_bits(s, gb, remaining_bits(s, gb), 0);
1780 s->packet_offset = get_bits_count(gb) & 7;
1782 return AVERROR_INVALIDDATA;
1784 if (s->trim_start && avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1785 if (s->trim_start < frame->nb_samples) {
1786 for (int ch = 0; ch < frame->ch_layout.nb_channels; ch++)
1787 frame->extended_data[ch] += s->trim_start * 4;
1789 frame->nb_samples -= s->trim_start;
1797 if (s->trim_end && avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1798 if (s->trim_end < frame->nb_samples) {
1799 frame->nb_samples -= s->trim_end;
1807 return get_bits_count(gb) >> 3;
1811 *@brief Decode a single WMA packet.
1812 *@param avctx codec context
1813 *@param data the output buffer
1814 *@param avpkt input packet
1815 *@return number of bytes that were read from the input buffer
1817 static int wmapro_decode_packet(AVCodecContext *avctx, AVFrame *frame,
1818 int *got_frame_ptr, AVPacket *avpkt)
1820 WMAProDecodeCtx *s = avctx->priv_data;
1823 /* get output buffer */
1824 frame->nb_samples = s->samples_per_frame;
1825 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1830 return decode_packet(avctx, s, frame, got_frame_ptr, avpkt);
1833 static int xma_decode_packet(AVCodecContext *avctx, AVFrame *frame,
1834 int *got_frame_ptr, AVPacket *avpkt)
1836 XMADecodeCtx *s = avctx->priv_data;
1837 int got_stream_frame_ptr = 0;
1838 int i, ret = 0, eof = 0;
1840 if (!s->frames[s->current_stream]->data[0]) {
1841 avctx->internal->skip_samples = 64;
1842 s->frames[s->current_stream]->nb_samples = 512;
1843 if ((ret = ff_get_buffer(avctx, s->frames[s->current_stream], 0)) < 0)
1845 } else if (s->frames[s->current_stream]->nb_samples != 512) {
1846 avctx->internal->skip_samples = 64;
1847 av_frame_unref(s->frames[s->current_stream]);
1848 s->frames[s->current_stream]->nb_samples = 512;
1849 if ((ret = ff_get_buffer(avctx, s->frames[s->current_stream], 0)) < 0)
1852 /* decode current stream packet */
1853 if (!s->xma[s->current_stream].eof_done) {
1854 ret = decode_packet(avctx, &s->xma[s->current_stream], s->frames[s->current_stream],
1855 &got_stream_frame_ptr, avpkt);
1861 for (i = 0; i < s->num_streams; i++) {
1862 if (!s->xma[i].eof_done && s->frames[i]->data[0]) {
1863 ret = decode_packet(avctx, &s->xma[i], s->frames[i],
1864 &got_stream_frame_ptr, avpkt);
1867 eof &= s->xma[i].eof_done;
1871 if (s->xma[0].trim_start)
1872 s->trim_start = s->xma[0].trim_start;
1873 if (s->xma[0].trim_end)
1874 s->trim_end = s->xma[0].trim_end;
1876 /* copy stream samples (1/2ch) to sample buffer (Nch) */
1877 if (got_stream_frame_ptr) {
1878 const int nb_samples = s->frames[s->current_stream]->nb_samples;
1879 void *left[1] = { s->frames[s->current_stream]->extended_data[0] };
1880 void *right[1] = { s->frames[s->current_stream]->extended_data[1] };
1882 av_audio_fifo_write(s->samples[0][s->current_stream], left, nb_samples);
1883 if (s->xma[s->current_stream].nb_channels > 1)
1884 av_audio_fifo_write(s->samples[1][s->current_stream], right, nb_samples);
1885 } else if (ret < 0) {
1886 s->current_stream = 0;
1890 /* find next XMA packet's owner stream, and update.
1891 * XMA streams find their packets following packet_skips
1892 * (at start there is one packet per stream, then interleave non-linearly). */
1893 if (s->xma[s->current_stream].packet_done ||
1894 s->xma[s->current_stream].packet_loss) {
1895 int nb_samples = INT_MAX;
1897 /* select stream with 0 skip_packets (= uses next packet) */
1898 if (s->xma[s->current_stream].skip_packets != 0) {
1901 min[0] = s->xma[0].skip_packets;
1904 for (i = 1; i < s->num_streams; i++) {
1905 if (s->xma[i].skip_packets < min[0]) {
1906 min[0] = s->xma[i].skip_packets;
1911 s->current_stream = min[1];
1914 /* all other streams skip next packet */
1915 for (i = 0; i < s->num_streams; i++) {
1916 s->xma[i].skip_packets = FFMAX(0, s->xma[i].skip_packets - 1);
1917 nb_samples = FFMIN(nb_samples, av_audio_fifo_size(s->samples[0][i]));
1920 if (!eof && avpkt->size)
1921 nb_samples -= FFMIN(nb_samples, 4096);
1923 /* copy samples from buffer to output if possible */
1924 if ((nb_samples > 0 || eof || !avpkt->size) && !s->flushed) {
1928 nb_samples -= av_clip(s->trim_end + s->trim_start - 128 - 64, 0, nb_samples);
1932 frame->nb_samples = nb_samples;
1933 if ((bret = ff_get_buffer(avctx, frame, 0)) < 0)
1936 for (i = 0; i < s->num_streams; i++) {
1937 const int start_ch = s->start_channel[i];
1938 void *left[1] = { frame->extended_data[start_ch + 0] };
1940 av_audio_fifo_read(s->samples[0][i], left, nb_samples);
1941 if (s->xma[i].nb_channels > 1) {
1942 void *right[1] = { frame->extended_data[start_ch + 1] };
1943 av_audio_fifo_read(s->samples[1][i], right, nb_samples);
1947 *got_frame_ptr = nb_samples > 0;
1954 static av_cold int xma_decode_init(AVCodecContext *avctx)
1956 XMADecodeCtx *s = avctx->priv_data;
1957 int i, ret, start_channels = 0;
1959 if (avctx->ch_layout.nb_channels <= 0 || avctx->extradata_size == 0)
1960 return AVERROR_INVALIDDATA;
1962 /* get stream config */
1963 if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
1964 unsigned int channel_mask = AV_RL32(avctx->extradata + 2);
1966 av_channel_layout_uninit(&avctx->ch_layout);
1967 av_channel_layout_from_mask(&avctx->ch_layout, channel_mask);
1969 avctx->ch_layout.order = AV_CHANNEL_ORDER_UNSPEC;
1970 s->num_streams = AV_RL16(avctx->extradata);
1971 } else if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size >= 2) { /* XMA2WAVEFORMAT */
1972 s->num_streams = avctx->extradata[1];
1973 if (avctx->extradata_size != (32 + ((avctx->extradata[0]==3)?0:8) + 4*s->num_streams)) {
1974 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA2 extradata size\n");
1976 return AVERROR(EINVAL);
1978 } else if (avctx->codec_id == AV_CODEC_ID_XMA1 && avctx->extradata_size >= 4) { /* XMAWAVEFORMAT */
1979 s->num_streams = avctx->extradata[4];
1980 if (avctx->extradata_size != (8 + 20*s->num_streams)) {
1981 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA1 extradata size\n");
1983 return AVERROR(EINVAL);
1986 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA config\n");
1987 return AVERROR(EINVAL);
1990 /* encoder supports up to 64 streams / 64*2 channels (would have to alloc arrays) */
1991 if (avctx->ch_layout.nb_channels > XMA_MAX_CHANNELS || s->num_streams > XMA_MAX_STREAMS ||
1994 avpriv_request_sample(avctx, "More than %d channels in %d streams", XMA_MAX_CHANNELS, s->num_streams);
1996 return AVERROR_PATCHWELCOME;
1999 /* init all streams (several streams of 1/2ch make Nch files) */
2000 for (i = 0; i < s->num_streams; i++) {
2001 ret = decode_init(&s->xma[i], avctx, i);
2004 s->frames[i] = av_frame_alloc();
2006 return AVERROR(ENOMEM);
2008 s->start_channel[i] = start_channels;
2009 start_channels += s->xma[i].nb_channels;
2011 if (start_channels != avctx->ch_layout.nb_channels)
2012 return AVERROR_INVALIDDATA;
2014 for (int i = 0; i < XMA_MAX_STREAMS; i++) {
2015 s->samples[0][i] = av_audio_fifo_alloc(avctx->sample_fmt, 1, 64 * 512);
2016 s->samples[1][i] = av_audio_fifo_alloc(avctx->sample_fmt, 1, 64 * 512);
2017 if (!s->samples[0][i] || !s->samples[1][i])
2018 return AVERROR(ENOMEM);
2024 static av_cold int xma_decode_end(AVCodecContext *avctx)
2026 XMADecodeCtx *s = avctx->priv_data;
2029 for (i = 0; i < s->num_streams; i++) {
2030 decode_end(&s->xma[i]);
2031 av_frame_free(&s->frames[i]);
2035 for (i = 0; i < XMA_MAX_STREAMS; i++) {
2036 av_audio_fifo_free(s->samples[0][i]);
2037 av_audio_fifo_free(s->samples[1][i]);
2043 static void flush(WMAProDecodeCtx *s)
2046 /** reset output buffer as a part of it is used during the windowing of a
2048 for (i = 0; i < s->nb_channels; i++)
2049 memset(s->channel[i].out, 0, s->samples_per_frame *
2050 sizeof(*s->channel[i].out));
2052 s->skip_packets = 0;
2058 *@brief Clear decoder buffers (for seeking).
2059 *@param avctx codec context
2061 static void wmapro_flush(AVCodecContext *avctx)
2063 WMAProDecodeCtx *s = avctx->priv_data;
2068 static void xma_flush(AVCodecContext *avctx)
2070 XMADecodeCtx *s = avctx->priv_data;
2073 for (i = 0; i < XMA_MAX_STREAMS; i++) {
2074 av_audio_fifo_reset(s->samples[0][i]);
2075 av_audio_fifo_reset(s->samples[1][i]);
2078 for (i = 0; i < s->num_streams; i++)
2081 s->current_stream = 0;
2086 *@brief wmapro decoder
2088 const FFCodec ff_wmapro_decoder = {
2090 CODEC_LONG_NAME("Windows Media Audio 9 Professional"),
2091 .p.type = AVMEDIA_TYPE_AUDIO,
2092 .p.id = AV_CODEC_ID_WMAPRO,
2093 .priv_data_size = sizeof(WMAProDecodeCtx),
2094 .init = wmapro_decode_init,
2095 .close = wmapro_decode_end,
2096 FF_CODEC_DECODE_CB(wmapro_decode_packet),
2098 #if FF_API_SUBFRAMES
2099 AV_CODEC_CAP_SUBFRAMES |
2102 .flush = wmapro_flush,
2103 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2104 AV_SAMPLE_FMT_NONE },
2105 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2108 const FFCodec ff_xma1_decoder = {
2110 CODEC_LONG_NAME("Xbox Media Audio 1"),
2111 .p.type = AVMEDIA_TYPE_AUDIO,
2112 .p.id = AV_CODEC_ID_XMA1,
2113 .priv_data_size = sizeof(XMADecodeCtx),
2114 .init = xma_decode_init,
2115 .close = xma_decode_end,
2116 FF_CODEC_DECODE_CB(xma_decode_packet),
2119 #if FF_API_SUBFRAMES
2120 AV_CODEC_CAP_SUBFRAMES |
2122 AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
2123 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2124 AV_SAMPLE_FMT_NONE },
2125 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2128 const FFCodec ff_xma2_decoder = {
2130 CODEC_LONG_NAME("Xbox Media Audio 2"),
2131 .p.type = AVMEDIA_TYPE_AUDIO,
2132 .p.id = AV_CODEC_ID_XMA2,
2133 .priv_data_size = sizeof(XMADecodeCtx),
2134 .init = xma_decode_init,
2135 .close = xma_decode_end,
2136 FF_CODEC_DECODE_CB(xma_decode_packet),
2139 #if FF_API_SUBFRAMES
2140 AV_CODEC_CAP_SUBFRAMES |
2142 AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
2143 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2144 AV_SAMPLE_FMT_NONE },
2145 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,