Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * Windows Media Audio Lossless decoder
  3.  * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
  4.  * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
  5.  * Copyright (c) 2011 Andreas Öman
  6.  * Copyright (c) 2011 - 2012 Mashiat Sarker Shakkhar
  7.  *
  8.  * This file is part of FFmpeg.
  9.  *
  10.  * FFmpeg is free software; you can redistribute it and/or
  11.  * modify it under the terms of the GNU Lesser General Public
  12.  * License as published by the Free Software Foundation; either
  13.  * version 2.1 of the License, or (at your option) any later version.
  14.  *
  15.  * FFmpeg is distributed in the hope that it will be useful,
  16.  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18.  * Lesser General Public License for more details.
  19.  *
  20.  * You should have received a copy of the GNU Lesser General Public
  21.  * License along with FFmpeg; if not, write to the Free Software
  22.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23.  */
  24.  
  25. #include <inttypes.h>
  26.  
  27. #include "libavutil/attributes.h"
  28. #include "libavutil/avassert.h"
  29.  
  30. #include "avcodec.h"
  31. #include "internal.h"
  32. #include "get_bits.h"
  33. #include "put_bits.h"
  34. #include "lossless_audiodsp.h"
  35. #include "wma.h"
  36. #include "wma_common.h"
  37.  
  38. /** current decoder limitations */
  39. #define WMALL_MAX_CHANNELS      8                       ///< max number of handled channels
  40. #define MAX_SUBFRAMES          32                       ///< max number of subframes per channel
  41. #define MAX_BANDS              29                       ///< max number of scale factor bands
  42. #define MAX_FRAMESIZE       32768                       ///< maximum compressed frame size
  43. #define MAX_ORDER             256
  44.  
  45. #define WMALL_BLOCK_MIN_BITS    6                       ///< log2 of min block size
  46. #define WMALL_BLOCK_MAX_BITS   14                       ///< log2 of max block size
  47. #define WMALL_BLOCK_MAX_SIZE (1 << WMALL_BLOCK_MAX_BITS)    ///< maximum block size
  48. #define WMALL_BLOCK_SIZES    (WMALL_BLOCK_MAX_BITS - WMALL_BLOCK_MIN_BITS + 1) ///< possible block sizes
  49.  
  50. #define WMALL_COEFF_PAD_SIZE   16                       ///< pad coef buffers with 0 for use with SIMD
  51.  
  52. /**
  53.  * @brief frame-specific decoder context for a single channel
  54.  */
  55. typedef struct WmallChannelCtx {
  56.     int16_t     prev_block_len;                         ///< length of the previous block
  57.     uint8_t     transmit_coefs;
  58.     uint8_t     num_subframes;
  59.     uint16_t    subframe_len[MAX_SUBFRAMES];            ///< subframe length in samples
  60.     uint16_t    subframe_offsets[MAX_SUBFRAMES];        ///< subframe positions in the current frame
  61.     uint8_t     cur_subframe;                           ///< current subframe number
  62.     uint16_t    decoded_samples;                        ///< number of already processed samples
  63.     int         quant_step;                             ///< quantization step for the current subframe
  64.     int         transient_counter;                      ///< number of transient samples from the beginning of the transient zone
  65. } WmallChannelCtx;
  66.  
  67. /**
  68.  * @brief main decoder context
  69.  */
  70. typedef struct WmallDecodeCtx {
  71.     /* generic decoder variables */
  72.     AVCodecContext  *avctx;
  73.     AVFrame         *frame;
  74.     LLAudDSPContext dsp;                           ///< accelerated DSP functions
  75.     uint8_t         frame_data[MAX_FRAMESIZE + AV_INPUT_BUFFER_PADDING_SIZE];  ///< compressed frame data
  76.     PutBitContext   pb;                             ///< context for filling the frame_data buffer
  77.  
  78.     /* frame size dependent frame information (set during initialization) */
  79.     uint32_t        decode_flags;                   ///< used compression features
  80.     int             len_prefix;                     ///< frame is prefixed with its length
  81.     int             dynamic_range_compression;      ///< frame contains DRC data
  82.     uint8_t         bits_per_sample;                ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
  83.     uint16_t        samples_per_frame;              ///< number of samples to output
  84.     uint16_t        log2_frame_size;
  85.     int8_t          num_channels;                   ///< number of channels in the stream (same as AVCodecContext.num_channels)
  86.     int8_t          lfe_channel;                    ///< lfe channel index
  87.     uint8_t         max_num_subframes;
  88.     uint8_t         subframe_len_bits;              ///< number of bits used for the subframe length
  89.     uint8_t         max_subframe_len_bit;           ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
  90.     uint16_t        min_samples_per_subframe;
  91.  
  92.     /* packet decode state */
  93.     GetBitContext   pgb;                            ///< bitstream reader context for the packet
  94.     int             next_packet_start;              ///< start offset of the next WMA packet in the demuxer packet
  95.     uint8_t         packet_offset;                  ///< offset to the frame in the packet
  96.     uint8_t         packet_sequence_number;         ///< current packet number
  97.     int             num_saved_bits;                 ///< saved number of bits
  98.     int             frame_offset;                   ///< frame offset in the bit reservoir
  99.     int             subframe_offset;                ///< subframe offset in the bit reservoir
  100.     uint8_t         packet_loss;                    ///< set in case of bitstream error
  101.     uint8_t         packet_done;                    ///< set when a packet is fully decoded
  102.  
  103.     /* frame decode state */
  104.     uint32_t        frame_num;                      ///< current frame number (not used for decoding)
  105.     GetBitContext   gb;                             ///< bitstream reader context
  106.     int             buf_bit_size;                   ///< buffer size in bits
  107.     int16_t         *samples_16[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (16-bit)
  108.     int32_t         *samples_32[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (24-bit)
  109.     uint8_t         drc_gain;                       ///< gain for the DRC tool
  110.     int8_t          skip_frame;                     ///< skip output step
  111.     int8_t          parsed_all_subframes;           ///< all subframes decoded?
  112.  
  113.     /* subframe/block decode state */
  114.     int16_t         subframe_len;                   ///< current subframe length
  115.     int8_t          channels_for_cur_subframe;      ///< number of channels that contain the subframe
  116.     int8_t          channel_indexes_for_cur_subframe[WMALL_MAX_CHANNELS];
  117.  
  118.     WmallChannelCtx channel[WMALL_MAX_CHANNELS];    ///< per channel data
  119.  
  120.     // WMA Lossless-specific
  121.  
  122.     uint8_t do_arith_coding;
  123.     uint8_t do_ac_filter;
  124.     uint8_t do_inter_ch_decorr;
  125.     uint8_t do_mclms;
  126.     uint8_t do_lpc;
  127.  
  128.     int8_t  acfilter_order;
  129.     int8_t  acfilter_scaling;
  130.     int16_t acfilter_coeffs[16];
  131.     int     acfilter_prevvalues[WMALL_MAX_CHANNELS][16];
  132.  
  133.     int8_t  mclms_order;
  134.     int8_t  mclms_scaling;
  135.     int16_t mclms_coeffs[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS * 32];
  136.     int16_t mclms_coeffs_cur[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS];
  137.     int16_t mclms_prevvalues[WMALL_MAX_CHANNELS * 2 * 32];
  138.     int16_t mclms_updates[WMALL_MAX_CHANNELS * 2 * 32];
  139.     int     mclms_recent;
  140.  
  141.     int     movave_scaling;
  142.     int     quant_stepsize;
  143.  
  144.     struct {
  145.         int order;
  146.         int scaling;
  147.         int coefsend;
  148.         int bitsend;
  149.         DECLARE_ALIGNED(16, int16_t, coefs)[MAX_ORDER + WMALL_COEFF_PAD_SIZE/sizeof(int16_t)];
  150.         DECLARE_ALIGNED(16, int16_t, lms_prevvalues)[MAX_ORDER * 2 + WMALL_COEFF_PAD_SIZE/sizeof(int16_t)];
  151.         DECLARE_ALIGNED(16, int16_t, lms_updates)[MAX_ORDER * 2 + WMALL_COEFF_PAD_SIZE/sizeof(int16_t)];
  152.         int recent;
  153.     } cdlms[WMALL_MAX_CHANNELS][9];
  154.  
  155.     int cdlms_ttl[WMALL_MAX_CHANNELS];
  156.  
  157.     int bV3RTM;
  158.  
  159.     int is_channel_coded[WMALL_MAX_CHANNELS];
  160.     int update_speed[WMALL_MAX_CHANNELS];
  161.  
  162.     int transient[WMALL_MAX_CHANNELS];
  163.     int transient_pos[WMALL_MAX_CHANNELS];
  164.     int seekable_tile;
  165.  
  166.     int ave_sum[WMALL_MAX_CHANNELS];
  167.  
  168.     int channel_residues[WMALL_MAX_CHANNELS][WMALL_BLOCK_MAX_SIZE];
  169.  
  170.     int lpc_coefs[WMALL_MAX_CHANNELS][40];
  171.     int lpc_order;
  172.     int lpc_scaling;
  173.     int lpc_intbits;
  174.  
  175.     int channel_coeffs[WMALL_MAX_CHANNELS][WMALL_BLOCK_MAX_SIZE];
  176. } WmallDecodeCtx;
  177.  
  178. /** Get sign of integer (1 for positive, -1 for negative and 0 for zero) */
  179. #define WMASIGN(x) (((x) > 0) - ((x) < 0))
  180.  
  181. static av_cold int decode_init(AVCodecContext *avctx)
  182. {
  183.     WmallDecodeCtx *s  = avctx->priv_data;
  184.     uint8_t *edata_ptr = avctx->extradata;
  185.     unsigned int channel_mask;
  186.     int i, log2_max_num_subframes;
  187.  
  188.     if (!avctx->block_align) {
  189.         av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
  190.         return AVERROR(EINVAL);
  191.     }
  192.  
  193.     s->avctx = avctx;
  194.     ff_llauddsp_init(&s->dsp);
  195.     init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
  196.  
  197.     if (avctx->extradata_size >= 18) {
  198.         s->decode_flags    = AV_RL16(edata_ptr + 14);
  199.         channel_mask       = AV_RL32(edata_ptr +  2);
  200.         s->bits_per_sample = AV_RL16(edata_ptr);
  201.         if (s->bits_per_sample == 16)
  202.             avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
  203.         else if (s->bits_per_sample == 24) {
  204.             avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
  205.             avctx->bits_per_raw_sample = 24;
  206.         } else {
  207.             av_log(avctx, AV_LOG_ERROR, "Unknown bit-depth: %"PRIu8"\n",
  208.                    s->bits_per_sample);
  209.             return AVERROR_INVALIDDATA;
  210.         }
  211.         /* dump the extradata */
  212.         for (i = 0; i < avctx->extradata_size; i++)
  213.             ff_dlog(avctx, "[%x] ", avctx->extradata[i]);
  214.         ff_dlog(avctx, "\n");
  215.  
  216.     } else {
  217.         avpriv_request_sample(avctx, "Unsupported extradata size");
  218.         return AVERROR_PATCHWELCOME;
  219.     }
  220.  
  221.     /* generic init */
  222.     s->log2_frame_size = av_log2(avctx->block_align) + 4;
  223.  
  224.     /* frame info */
  225.     s->skip_frame  = 1; /* skip first frame */
  226.     s->packet_loss = 1;
  227.     s->len_prefix  = s->decode_flags & 0x40;
  228.  
  229.     /* get frame len */
  230.     s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
  231.                                                           3, s->decode_flags);
  232.     av_assert0(s->samples_per_frame <= WMALL_BLOCK_MAX_SIZE);
  233.  
  234.     /* init previous block len */
  235.     for (i = 0; i < avctx->channels; i++)
  236.         s->channel[i].prev_block_len = s->samples_per_frame;
  237.  
  238.     /* subframe info */
  239.     log2_max_num_subframes  = (s->decode_flags & 0x38) >> 3;
  240.     s->max_num_subframes    = 1 << log2_max_num_subframes;
  241.     s->max_subframe_len_bit = 0;
  242.     s->subframe_len_bits    = av_log2(log2_max_num_subframes) + 1;
  243.  
  244.     s->min_samples_per_subframe  = s->samples_per_frame / s->max_num_subframes;
  245.     s->dynamic_range_compression = s->decode_flags & 0x80;
  246.     s->bV3RTM                    = s->decode_flags & 0x100;
  247.  
  248.     if (s->max_num_subframes > MAX_SUBFRAMES) {
  249.         av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRIu8"\n",
  250.                s->max_num_subframes);
  251.         return AVERROR_INVALIDDATA;
  252.     }
  253.  
  254.     s->num_channels = avctx->channels;
  255.  
  256.     /* extract lfe channel position */
  257.     s->lfe_channel = -1;
  258.  
  259.     if (channel_mask & 8) {
  260.         unsigned int mask;
  261.         for (mask = 1; mask < 16; mask <<= 1)
  262.             if (channel_mask & mask)
  263.                 ++s->lfe_channel;
  264.     }
  265.  
  266.     if (s->num_channels < 0) {
  267.         av_log(avctx, AV_LOG_ERROR, "invalid number of channels %"PRId8"\n",
  268.                s->num_channels);
  269.         return AVERROR_INVALIDDATA;
  270.     } else if (s->num_channels > WMALL_MAX_CHANNELS) {
  271.         avpriv_request_sample(avctx,
  272.                               "More than %d channels", WMALL_MAX_CHANNELS);
  273.         return AVERROR_PATCHWELCOME;
  274.     }
  275.  
  276.     s->frame = av_frame_alloc();
  277.     if (!s->frame)
  278.         return AVERROR(ENOMEM);
  279.  
  280.     avctx->channel_layout = channel_mask;
  281.     return 0;
  282. }
  283.  
  284. /**
  285.  * @brief Decode the subframe length.
  286.  * @param s      context
  287.  * @param offset sample offset in the frame
  288.  * @return decoded subframe length on success, < 0 in case of an error
  289.  */
  290. static int decode_subframe_length(WmallDecodeCtx *s, int offset)
  291. {
  292.     int frame_len_ratio, subframe_len, len;
  293.  
  294.     /* no need to read from the bitstream when only one length is possible */
  295.     if (offset == s->samples_per_frame - s->min_samples_per_subframe)
  296.         return s->min_samples_per_subframe;
  297.  
  298.     len             = av_log2(s->max_num_subframes - 1) + 1;
  299.     frame_len_ratio = get_bits(&s->gb, len);
  300.     subframe_len    = s->min_samples_per_subframe * (frame_len_ratio + 1);
  301.  
  302.     /* sanity check the length */
  303.     if (subframe_len < s->min_samples_per_subframe ||
  304.         subframe_len > s->samples_per_frame) {
  305.         av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
  306.                subframe_len);
  307.         return AVERROR_INVALIDDATA;
  308.     }
  309.     return subframe_len;
  310. }
  311.  
  312. /**
  313.  * @brief Decode how the data in the frame is split into subframes.
  314.  *       Every WMA frame contains the encoded data for a fixed number of
  315.  *       samples per channel. The data for every channel might be split
  316.  *       into several subframes. This function will reconstruct the list of
  317.  *       subframes for every channel.
  318.  *
  319.  *       If the subframes are not evenly split, the algorithm estimates the
  320.  *       channels with the lowest number of total samples.
  321.  *       Afterwards, for each of these channels a bit is read from the
  322.  *       bitstream that indicates if the channel contains a subframe with the
  323.  *       next subframe size that is going to be read from the bitstream or not.
  324.  *       If a channel contains such a subframe, the subframe size gets added to
  325.  *       the channel's subframe list.
  326.  *       The algorithm repeats these steps until the frame is properly divided
  327.  *       between the individual channels.
  328.  *
  329.  * @param s context
  330.  * @return 0 on success, < 0 in case of an error
  331.  */
  332. static int decode_tilehdr(WmallDecodeCtx *s)
  333. {
  334.     uint16_t num_samples[WMALL_MAX_CHANNELS] = { 0 }; /* sum of samples for all currently known subframes of a channel */
  335.     uint8_t  contains_subframe[WMALL_MAX_CHANNELS];   /* flag indicating if a channel contains the current subframe */
  336.     int channels_for_cur_subframe = s->num_channels;  /* number of channels that contain the current subframe */
  337.     int fixed_channel_layout = 0;                     /* flag indicating that all channels use the same subfra2me offsets and sizes */
  338.     int min_channel_len = 0;                          /* smallest sum of samples (channels with this length will be processed first) */
  339.     int c, tile_aligned;
  340.  
  341.     /* reset tiling information */
  342.     for (c = 0; c < s->num_channels; c++)
  343.         s->channel[c].num_subframes = 0;
  344.  
  345.     tile_aligned = get_bits1(&s->gb);
  346.     if (s->max_num_subframes == 1 || tile_aligned)
  347.         fixed_channel_layout = 1;
  348.  
  349.     /* loop until the frame data is split between the subframes */
  350.     do {
  351.         int subframe_len, in_use = 0;
  352.  
  353.         /* check which channels contain the subframe */
  354.         for (c = 0; c < s->num_channels; c++) {
  355.             if (num_samples[c] == min_channel_len) {
  356.                 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
  357.                    (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
  358.                     contains_subframe[c] = 1;
  359.                 } else {
  360.                     contains_subframe[c] = get_bits1(&s->gb);
  361.                 }
  362.                 in_use |= contains_subframe[c];
  363.             } else
  364.                 contains_subframe[c] = 0;
  365.         }
  366.  
  367.         if (!in_use) {
  368.             av_log(s->avctx, AV_LOG_ERROR,
  369.                    "Found empty subframe\n");
  370.             return AVERROR_INVALIDDATA;
  371.         }
  372.  
  373.         /* get subframe length, subframe_len == 0 is not allowed */
  374.         if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
  375.             return AVERROR_INVALIDDATA;
  376.         /* add subframes to the individual channels and find new min_channel_len */
  377.         min_channel_len += subframe_len;
  378.         for (c = 0; c < s->num_channels; c++) {
  379.             WmallChannelCtx *chan = &s->channel[c];
  380.  
  381.             if (contains_subframe[c]) {
  382.                 if (chan->num_subframes >= MAX_SUBFRAMES) {
  383.                     av_log(s->avctx, AV_LOG_ERROR,
  384.                            "broken frame: num subframes > 31\n");
  385.                     return AVERROR_INVALIDDATA;
  386.                 }
  387.                 chan->subframe_len[chan->num_subframes] = subframe_len;
  388.                 num_samples[c] += subframe_len;
  389.                 ++chan->num_subframes;
  390.                 if (num_samples[c] > s->samples_per_frame) {
  391.                     av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
  392.                            "channel len(%"PRIu16") > samples_per_frame(%"PRIu16")\n",
  393.                            num_samples[c], s->samples_per_frame);
  394.                     return AVERROR_INVALIDDATA;
  395.                 }
  396.             } else if (num_samples[c] <= min_channel_len) {
  397.                 if (num_samples[c] < min_channel_len) {
  398.                     channels_for_cur_subframe = 0;
  399.                     min_channel_len = num_samples[c];
  400.                 }
  401.                 ++channels_for_cur_subframe;
  402.             }
  403.         }
  404.     } while (min_channel_len < s->samples_per_frame);
  405.  
  406.     for (c = 0; c < s->num_channels; c++) {
  407.         int i, offset = 0;
  408.         for (i = 0; i < s->channel[c].num_subframes; i++) {
  409.             s->channel[c].subframe_offsets[i] = offset;
  410.             offset += s->channel[c].subframe_len[i];
  411.         }
  412.     }
  413.  
  414.     return 0;
  415. }
  416.  
  417. static void decode_ac_filter(WmallDecodeCtx *s)
  418. {
  419.     int i;
  420.     s->acfilter_order   = get_bits(&s->gb, 4) + 1;
  421.     s->acfilter_scaling = get_bits(&s->gb, 4);
  422.  
  423.     for (i = 0; i < s->acfilter_order; i++)
  424.         s->acfilter_coeffs[i] = (s->acfilter_scaling ?
  425.                                  get_bits(&s->gb, s->acfilter_scaling) : 0) + 1;
  426. }
  427.  
  428. static void decode_mclms(WmallDecodeCtx *s)
  429. {
  430.     s->mclms_order   = (get_bits(&s->gb, 4) + 1) * 2;
  431.     s->mclms_scaling = get_bits(&s->gb, 4);
  432.     if (get_bits1(&s->gb)) {
  433.         int i, send_coef_bits;
  434.         int cbits = av_log2(s->mclms_scaling + 1);
  435.         if (1 << cbits < s->mclms_scaling + 1)
  436.             cbits++;
  437.  
  438.         send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
  439.  
  440.         for (i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++)
  441.             s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
  442.  
  443.         for (i = 0; i < s->num_channels; i++) {
  444.             int c;
  445.             for (c = 0; c < i; c++)
  446.                 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
  447.         }
  448.     }
  449. }
  450.  
  451. static int decode_cdlms(WmallDecodeCtx *s)
  452. {
  453.     int c, i;
  454.     int cdlms_send_coef = get_bits1(&s->gb);
  455.  
  456.     for (c = 0; c < s->num_channels; c++) {
  457.         s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
  458.         for (i = 0; i < s->cdlms_ttl[c]; i++) {
  459.             s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
  460.             if (s->cdlms[c][i].order > MAX_ORDER) {
  461.                 av_log(s->avctx, AV_LOG_ERROR,
  462.                        "Order[%d][%d] %d > max (%d), not supported\n",
  463.                        c, i, s->cdlms[c][i].order, MAX_ORDER);
  464.                 s->cdlms[0][0].order = 0;
  465.                 return AVERROR_INVALIDDATA;
  466.             }
  467.             if(s->cdlms[c][i].order & 8) {
  468.                 static int warned;
  469.                 if(!warned)
  470.                     avpriv_request_sample(s->avctx, "CDLMS of order %d",
  471.                                           s->cdlms[c][i].order);
  472.                 warned = 1;
  473.             }
  474.         }
  475.  
  476.         for (i = 0; i < s->cdlms_ttl[c]; i++)
  477.             s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
  478.  
  479.         if (cdlms_send_coef) {
  480.             for (i = 0; i < s->cdlms_ttl[c]; i++) {
  481.                 int cbits, shift_l, shift_r, j;
  482.                 cbits = av_log2(s->cdlms[c][i].order);
  483.                 if ((1 << cbits) < s->cdlms[c][i].order)
  484.                     cbits++;
  485.                 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
  486.  
  487.                 cbits = av_log2(s->cdlms[c][i].scaling + 1);
  488.                 if ((1 << cbits) < s->cdlms[c][i].scaling + 1)
  489.                     cbits++;
  490.  
  491.                 s->cdlms[c][i].bitsend = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
  492.                 shift_l = 32 - s->cdlms[c][i].bitsend;
  493.                 shift_r = 32 - s->cdlms[c][i].scaling - 2;
  494.                 for (j = 0; j < s->cdlms[c][i].coefsend; j++)
  495.                     s->cdlms[c][i].coefs[j] =
  496.                         (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
  497.             }
  498.         }
  499.  
  500.         for (i = 0; i < s->cdlms_ttl[c]; i++)
  501.             memset(s->cdlms[c][i].coefs + s->cdlms[c][i].order,
  502.                    0, WMALL_COEFF_PAD_SIZE);
  503.     }
  504.  
  505.     return 0;
  506. }
  507.  
  508. static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
  509. {
  510.     int i = 0;
  511.     unsigned int ave_mean;
  512.     s->transient[ch] = get_bits1(&s->gb);
  513.     if (s->transient[ch]) {
  514.         s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
  515.         if (s->transient_pos[ch])
  516.             s->transient[ch] = 0;
  517.         s->channel[ch].transient_counter =
  518.             FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
  519.     } else if (s->channel[ch].transient_counter)
  520.         s->transient[ch] = 1;
  521.  
  522.     if (s->seekable_tile) {
  523.         ave_mean = get_bits(&s->gb, s->bits_per_sample);
  524.         s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
  525.     }
  526.  
  527.     if (s->seekable_tile) {
  528.         if (s->do_inter_ch_decorr)
  529.             s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample + 1);
  530.         else
  531.             s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample);
  532.         i++;
  533.     }
  534.     for (; i < tile_size; i++) {
  535.         int quo = 0, rem, rem_bits, residue;
  536.         while(get_bits1(&s->gb)) {
  537.             quo++;
  538.             if (get_bits_left(&s->gb) <= 0)
  539.                 return -1;
  540.         }
  541.         if (quo >= 32)
  542.             quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
  543.  
  544.         ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
  545.         if (ave_mean <= 1)
  546.             residue = quo;
  547.         else {
  548.             rem_bits = av_ceil_log2(ave_mean);
  549.             rem      = get_bits_long(&s->gb, rem_bits);
  550.             residue  = (quo << rem_bits) + rem;
  551.         }
  552.  
  553.         s->ave_sum[ch] = residue + s->ave_sum[ch] -
  554.                          (s->ave_sum[ch] >> s->movave_scaling);
  555.  
  556.         residue = (residue >> 1) ^ -(residue & 1);
  557.         s->channel_residues[ch][i] = residue;
  558.     }
  559.  
  560.     return 0;
  561.  
  562. }
  563.  
  564. static void decode_lpc(WmallDecodeCtx *s)
  565. {
  566.     int ch, i, cbits;
  567.     s->lpc_order   = get_bits(&s->gb, 5) + 1;
  568.     s->lpc_scaling = get_bits(&s->gb, 4);
  569.     s->lpc_intbits = get_bits(&s->gb, 3) + 1;
  570.     cbits = s->lpc_scaling + s->lpc_intbits;
  571.     for (ch = 0; ch < s->num_channels; ch++)
  572.         for (i = 0; i < s->lpc_order; i++)
  573.             s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
  574. }
  575.  
  576. static void clear_codec_buffers(WmallDecodeCtx *s)
  577. {
  578.     int ich, ilms;
  579.  
  580.     memset(s->acfilter_coeffs,     0, sizeof(s->acfilter_coeffs));
  581.     memset(s->acfilter_prevvalues, 0, sizeof(s->acfilter_prevvalues));
  582.     memset(s->lpc_coefs,           0, sizeof(s->lpc_coefs));
  583.  
  584.     memset(s->mclms_coeffs,     0, sizeof(s->mclms_coeffs));
  585.     memset(s->mclms_coeffs_cur, 0, sizeof(s->mclms_coeffs_cur));
  586.     memset(s->mclms_prevvalues, 0, sizeof(s->mclms_prevvalues));
  587.     memset(s->mclms_updates,    0, sizeof(s->mclms_updates));
  588.  
  589.     for (ich = 0; ich < s->num_channels; ich++) {
  590.         for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) {
  591.             memset(s->cdlms[ich][ilms].coefs, 0,
  592.                    sizeof(s->cdlms[ich][ilms].coefs));
  593.             memset(s->cdlms[ich][ilms].lms_prevvalues, 0,
  594.                    sizeof(s->cdlms[ich][ilms].lms_prevvalues));
  595.             memset(s->cdlms[ich][ilms].lms_updates, 0,
  596.                    sizeof(s->cdlms[ich][ilms].lms_updates));
  597.         }
  598.         s->ave_sum[ich] = 0;
  599.     }
  600. }
  601.  
  602. /**
  603.  * @brief Reset filter parameters and transient area at new seekable tile.
  604.  */
  605. static void reset_codec(WmallDecodeCtx *s)
  606. {
  607.     int ich, ilms;
  608.     s->mclms_recent = s->mclms_order * s->num_channels;
  609.     for (ich = 0; ich < s->num_channels; ich++) {
  610.         for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++)
  611.             s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
  612.         /* first sample of a seekable subframe is considered as the starting of
  613.             a transient area which is samples_per_frame samples long */
  614.         s->channel[ich].transient_counter = s->samples_per_frame;
  615.         s->transient[ich]     = 1;
  616.         s->transient_pos[ich] = 0;
  617.     }
  618. }
  619.  
  620. static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred)
  621. {
  622.     int i, j, ich, pred_error;
  623.     int order        = s->mclms_order;
  624.     int num_channels = s->num_channels;
  625.     int range        = 1 << (s->bits_per_sample - 1);
  626.  
  627.     for (ich = 0; ich < num_channels; ich++) {
  628.         pred_error = s->channel_residues[ich][icoef] - pred[ich];
  629.         if (pred_error > 0) {
  630.             for (i = 0; i < order * num_channels; i++)
  631.                 s->mclms_coeffs[i + ich * order * num_channels] +=
  632.                     s->mclms_updates[s->mclms_recent + i];
  633.             for (j = 0; j < ich; j++)
  634.                 s->mclms_coeffs_cur[ich * num_channels + j] += WMASIGN(s->channel_residues[j][icoef]);
  635.         } else if (pred_error < 0) {
  636.             for (i = 0; i < order * num_channels; i++)
  637.                 s->mclms_coeffs[i + ich * order * num_channels] -=
  638.                     s->mclms_updates[s->mclms_recent + i];
  639.             for (j = 0; j < ich; j++)
  640.                 s->mclms_coeffs_cur[ich * num_channels + j] -= WMASIGN(s->channel_residues[j][icoef]);
  641.         }
  642.     }
  643.  
  644.     for (ich = num_channels - 1; ich >= 0; ich--) {
  645.         s->mclms_recent--;
  646.         s->mclms_prevvalues[s->mclms_recent] = av_clip(s->channel_residues[ich][icoef],
  647.             -range, range - 1);
  648.         s->mclms_updates[s->mclms_recent] = WMASIGN(s->channel_residues[ich][icoef]);
  649.     }
  650.  
  651.     if (s->mclms_recent == 0) {
  652.         memcpy(&s->mclms_prevvalues[order * num_channels],
  653.                s->mclms_prevvalues,
  654.                sizeof(int16_t) * order * num_channels);
  655.         memcpy(&s->mclms_updates[order * num_channels],
  656.                s->mclms_updates,
  657.                sizeof(int16_t) * order * num_channels);
  658.         s->mclms_recent = num_channels * order;
  659.     }
  660. }
  661.  
  662. static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred)
  663. {
  664.     int ich, i;
  665.     int order        = s->mclms_order;
  666.     int num_channels = s->num_channels;
  667.  
  668.     for (ich = 0; ich < num_channels; ich++) {
  669.         pred[ich] = 0;
  670.         if (!s->is_channel_coded[ich])
  671.             continue;
  672.         for (i = 0; i < order * num_channels; i++)
  673.             pred[ich] += s->mclms_prevvalues[i + s->mclms_recent] *
  674.                          s->mclms_coeffs[i + order * num_channels * ich];
  675.         for (i = 0; i < ich; i++)
  676.             pred[ich] += s->channel_residues[i][icoef] *
  677.                          s->mclms_coeffs_cur[i + num_channels * ich];
  678.         pred[ich] += 1 << s->mclms_scaling - 1;
  679.         pred[ich] >>= s->mclms_scaling;
  680.         s->channel_residues[ich][icoef] += pred[ich];
  681.     }
  682. }
  683.  
  684. static void revert_mclms(WmallDecodeCtx *s, int tile_size)
  685. {
  686.     int icoef, pred[WMALL_MAX_CHANNELS] = { 0 };
  687.     for (icoef = 0; icoef < tile_size; icoef++) {
  688.         mclms_predict(s, icoef, pred);
  689.         mclms_update(s, icoef, pred);
  690.     }
  691. }
  692.  
  693. static void lms_update(WmallDecodeCtx *s, int ich, int ilms, int input)
  694. {
  695.     int recent = s->cdlms[ich][ilms].recent;
  696.     int range  = 1 << s->bits_per_sample - 1;
  697.     int order  = s->cdlms[ich][ilms].order;
  698.  
  699.     if (recent)
  700.         recent--;
  701.     else {
  702.         memcpy(s->cdlms[ich][ilms].lms_prevvalues + order,
  703.                s->cdlms[ich][ilms].lms_prevvalues, sizeof(*s->cdlms[ich][ilms].lms_prevvalues) * order);
  704.         memcpy(s->cdlms[ich][ilms].lms_updates + order,
  705.                s->cdlms[ich][ilms].lms_updates, sizeof(*s->cdlms[ich][ilms].lms_updates) * order);
  706.         recent = order - 1;
  707.     }
  708.  
  709.     s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1);
  710.     s->cdlms[ich][ilms].lms_updates[recent] = WMASIGN(input) * s->update_speed[ich];
  711.  
  712.     s->cdlms[ich][ilms].lms_updates[recent + (order >> 4)] >>= 2;
  713.     s->cdlms[ich][ilms].lms_updates[recent + (order >> 3)] >>= 1;
  714.     s->cdlms[ich][ilms].recent = recent;
  715.     memset(s->cdlms[ich][ilms].lms_updates + recent + order, 0,
  716.            sizeof(s->cdlms[ich][ilms].lms_updates) - 2*(recent+order));
  717. }
  718.  
  719. static void use_high_update_speed(WmallDecodeCtx *s, int ich)
  720. {
  721.     int ilms, recent, icoef;
  722.     for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
  723.         recent = s->cdlms[ich][ilms].recent;
  724.         if (s->update_speed[ich] == 16)
  725.             continue;
  726.         if (s->bV3RTM) {
  727.             for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
  728.                 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
  729.         } else {
  730.             for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
  731.                 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
  732.         }
  733.     }
  734.     s->update_speed[ich] = 16;
  735. }
  736.  
  737. static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
  738. {
  739.     int ilms, recent, icoef;
  740.     for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
  741.         recent = s->cdlms[ich][ilms].recent;
  742.         if (s->update_speed[ich] == 8)
  743.             continue;
  744.         if (s->bV3RTM)
  745.             for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
  746.                 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
  747.         else
  748.             for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
  749.                 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
  750.     }
  751.     s->update_speed[ich] = 8;
  752. }
  753.  
  754. static void revert_cdlms(WmallDecodeCtx *s, int ch,
  755.                          int coef_begin, int coef_end)
  756. {
  757.     int icoef, pred, ilms, num_lms, residue, input;
  758.  
  759.     num_lms = s->cdlms_ttl[ch];
  760.     for (ilms = num_lms - 1; ilms >= 0; ilms--) {
  761.         for (icoef = coef_begin; icoef < coef_end; icoef++) {
  762.             pred = 1 << (s->cdlms[ch][ilms].scaling - 1);
  763.             residue = s->channel_residues[ch][icoef];
  764.             pred += s->dsp.scalarproduct_and_madd_int16(s->cdlms[ch][ilms].coefs,
  765.                                                         s->cdlms[ch][ilms].lms_prevvalues
  766.                                                             + s->cdlms[ch][ilms].recent,
  767.                                                         s->cdlms[ch][ilms].lms_updates
  768.                                                             + s->cdlms[ch][ilms].recent,
  769.                                                         FFALIGN(s->cdlms[ch][ilms].order,
  770.                                                                 WMALL_COEFF_PAD_SIZE),
  771.                                                         WMASIGN(residue));
  772.             input = residue + (pred >> s->cdlms[ch][ilms].scaling);
  773.             lms_update(s, ch, ilms, input);
  774.             s->channel_residues[ch][icoef] = input;
  775.         }
  776.     }
  777.     emms_c();
  778. }
  779.  
  780. static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size)
  781. {
  782.     if (s->num_channels != 2)
  783.         return;
  784.     else if (s->is_channel_coded[0] || s->is_channel_coded[1]) {
  785.         int icoef;
  786.         for (icoef = 0; icoef < tile_size; icoef++) {
  787.             s->channel_residues[0][icoef] -= s->channel_residues[1][icoef] >> 1;
  788.             s->channel_residues[1][icoef] += s->channel_residues[0][icoef];
  789.         }
  790.     }
  791. }
  792.  
  793. static void revert_acfilter(WmallDecodeCtx *s, int tile_size)
  794. {
  795.     int ich, pred, i, j;
  796.     int16_t *filter_coeffs = s->acfilter_coeffs;
  797.     int scaling            = s->acfilter_scaling;
  798.     int order              = s->acfilter_order;
  799.  
  800.     for (ich = 0; ich < s->num_channels; ich++) {
  801.         int *prevvalues = s->acfilter_prevvalues[ich];
  802.         for (i = 0; i < order; i++) {
  803.             pred = 0;
  804.             for (j = 0; j < order; j++) {
  805.                 if (i <= j)
  806.                     pred += filter_coeffs[j] * prevvalues[j - i];
  807.                 else
  808.                     pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
  809.             }
  810.             pred >>= scaling;
  811.             s->channel_residues[ich][i] += pred;
  812.         }
  813.         for (i = order; i < tile_size; i++) {
  814.             pred = 0;
  815.             for (j = 0; j < order; j++)
  816.                 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
  817.             pred >>= scaling;
  818.             s->channel_residues[ich][i] += pred;
  819.         }
  820.         for (j = 0; j < order; j++)
  821.             prevvalues[j] = s->channel_residues[ich][tile_size - j - 1];
  822.     }
  823. }
  824.  
  825. static int decode_subframe(WmallDecodeCtx *s)
  826. {
  827.     int offset        = s->samples_per_frame;
  828.     int subframe_len  = s->samples_per_frame;
  829.     int total_samples = s->samples_per_frame * s->num_channels;
  830.     int i, j, rawpcm_tile, padding_zeroes, res;
  831.  
  832.     s->subframe_offset = get_bits_count(&s->gb);
  833.  
  834.     /* reset channel context and find the next block offset and size
  835.         == the next block of the channel with the smallest number of
  836.         decoded samples */
  837.     for (i = 0; i < s->num_channels; i++) {
  838.         if (offset > s->channel[i].decoded_samples) {
  839.             offset = s->channel[i].decoded_samples;
  840.             subframe_len =
  841.                 s->channel[i].subframe_len[s->channel[i].cur_subframe];
  842.         }
  843.     }
  844.  
  845.     /* get a list of all channels that contain the estimated block */
  846.     s->channels_for_cur_subframe = 0;
  847.     for (i = 0; i < s->num_channels; i++) {
  848.         const int cur_subframe = s->channel[i].cur_subframe;
  849.         /* subtract already processed samples */
  850.         total_samples -= s->channel[i].decoded_samples;
  851.  
  852.         /* and count if there are multiple subframes that match our profile */
  853.         if (offset == s->channel[i].decoded_samples &&
  854.             subframe_len == s->channel[i].subframe_len[cur_subframe]) {
  855.             total_samples -= s->channel[i].subframe_len[cur_subframe];
  856.             s->channel[i].decoded_samples +=
  857.                 s->channel[i].subframe_len[cur_subframe];
  858.             s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
  859.             ++s->channels_for_cur_subframe;
  860.         }
  861.     }
  862.  
  863.     /* check if the frame will be complete after processing the
  864.         estimated block */
  865.     if (!total_samples)
  866.         s->parsed_all_subframes = 1;
  867.  
  868.  
  869.     s->seekable_tile = get_bits1(&s->gb);
  870.     if (s->seekable_tile) {
  871.         clear_codec_buffers(s);
  872.  
  873.         s->do_arith_coding    = get_bits1(&s->gb);
  874.         if (s->do_arith_coding) {
  875.             avpriv_request_sample(s->avctx, "Arithmetic coding");
  876.             return AVERROR_PATCHWELCOME;
  877.         }
  878.         s->do_ac_filter       = get_bits1(&s->gb);
  879.         s->do_inter_ch_decorr = get_bits1(&s->gb);
  880.         s->do_mclms           = get_bits1(&s->gb);
  881.  
  882.         if (s->do_ac_filter)
  883.             decode_ac_filter(s);
  884.  
  885.         if (s->do_mclms)
  886.             decode_mclms(s);
  887.  
  888.         if ((res = decode_cdlms(s)) < 0)
  889.             return res;
  890.         s->movave_scaling = get_bits(&s->gb, 3);
  891.         s->quant_stepsize = get_bits(&s->gb, 8) + 1;
  892.  
  893.         reset_codec(s);
  894.     } else if (!s->cdlms[0][0].order) {
  895.         av_log(s->avctx, AV_LOG_DEBUG,
  896.                "Waiting for seekable tile\n");
  897.         av_frame_unref(s->frame);
  898.         return -1;
  899.     }
  900.  
  901.     rawpcm_tile = get_bits1(&s->gb);
  902.  
  903.     for (i = 0; i < s->num_channels; i++)
  904.         s->is_channel_coded[i] = 1;
  905.  
  906.     if (!rawpcm_tile) {
  907.         for (i = 0; i < s->num_channels; i++)
  908.             s->is_channel_coded[i] = get_bits1(&s->gb);
  909.  
  910.         if (s->bV3RTM) {
  911.             // LPC
  912.             s->do_lpc = get_bits1(&s->gb);
  913.             if (s->do_lpc) {
  914.                 decode_lpc(s);
  915.                 avpriv_request_sample(s->avctx, "Expect wrong output since "
  916.                                       "inverse LPC filter");
  917.             }
  918.         } else
  919.             s->do_lpc = 0;
  920.     }
  921.  
  922.  
  923.     if (get_bits1(&s->gb))
  924.         padding_zeroes = get_bits(&s->gb, 5);
  925.     else
  926.         padding_zeroes = 0;
  927.  
  928.     if (rawpcm_tile) {
  929.         int bits = s->bits_per_sample - padding_zeroes;
  930.         if (bits <= 0) {
  931.             av_log(s->avctx, AV_LOG_ERROR,
  932.                    "Invalid number of padding bits in raw PCM tile\n");
  933.             return AVERROR_INVALIDDATA;
  934.         }
  935.         ff_dlog(s->avctx, "RAWPCM %d bits per sample. "
  936.                 "total %d bits, remain=%d\n", bits,
  937.                 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
  938.         for (i = 0; i < s->num_channels; i++)
  939.             for (j = 0; j < subframe_len; j++)
  940.                 s->channel_coeffs[i][j] = get_sbits_long(&s->gb, bits);
  941.     } else {
  942.         for (i = 0; i < s->num_channels; i++)
  943.             if (s->is_channel_coded[i]) {
  944.                 decode_channel_residues(s, i, subframe_len);
  945.                 if (s->seekable_tile)
  946.                     use_high_update_speed(s, i);
  947.                 else
  948.                     use_normal_update_speed(s, i);
  949.                 revert_cdlms(s, i, 0, subframe_len);
  950.             } else {
  951.                 memset(s->channel_residues[i], 0, sizeof(**s->channel_residues) * subframe_len);
  952.             }
  953.     }
  954.     if (s->do_mclms)
  955.         revert_mclms(s, subframe_len);
  956.     if (s->do_inter_ch_decorr)
  957.         revert_inter_ch_decorr(s, subframe_len);
  958.     if (s->do_ac_filter)
  959.         revert_acfilter(s, subframe_len);
  960.  
  961.     /* Dequantize */
  962.     if (s->quant_stepsize != 1)
  963.         for (i = 0; i < s->num_channels; i++)
  964.             for (j = 0; j < subframe_len; j++)
  965.                 s->channel_residues[i][j] *= s->quant_stepsize;
  966.  
  967.     /* Write to proper output buffer depending on bit-depth */
  968.     for (i = 0; i < s->channels_for_cur_subframe; i++) {
  969.         int c = s->channel_indexes_for_cur_subframe[i];
  970.         int subframe_len = s->channel[c].subframe_len[s->channel[c].cur_subframe];
  971.  
  972.         for (j = 0; j < subframe_len; j++) {
  973.             if (s->bits_per_sample == 16) {
  974.                 *s->samples_16[c]++ = (int16_t) s->channel_residues[c][j] << padding_zeroes;
  975.             } else {
  976.                 *s->samples_32[c]++ = s->channel_residues[c][j] << (padding_zeroes + 8);
  977.             }
  978.         }
  979.     }
  980.  
  981.     /* handled one subframe */
  982.     for (i = 0; i < s->channels_for_cur_subframe; i++) {
  983.         int c = s->channel_indexes_for_cur_subframe[i];
  984.         if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
  985.             av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
  986.             return AVERROR_INVALIDDATA;
  987.         }
  988.         ++s->channel[c].cur_subframe;
  989.     }
  990.     return 0;
  991. }
  992.  
  993. /**
  994.  * @brief Decode one WMA frame.
  995.  * @param s codec context
  996.  * @return 0 if the trailer bit indicates that this is the last frame,
  997.  *         1 if there are additional frames
  998.  */
  999. static int decode_frame(WmallDecodeCtx *s)
  1000. {
  1001.     GetBitContext* gb = &s->gb;
  1002.     int more_frames = 0, len = 0, i, ret;
  1003.  
  1004.     s->frame->nb_samples = s->samples_per_frame;
  1005.     if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {
  1006.         /* return an error if no frame could be decoded at all */
  1007.         s->packet_loss = 1;
  1008.         s->frame->nb_samples = 0;
  1009.         return ret;
  1010.     }
  1011.     for (i = 0; i < s->num_channels; i++) {
  1012.         s->samples_16[i] = (int16_t *)s->frame->extended_data[i];
  1013.         s->samples_32[i] = (int32_t *)s->frame->extended_data[i];
  1014.     }
  1015.  
  1016.     /* get frame length */
  1017.     if (s->len_prefix)
  1018.         len = get_bits(gb, s->log2_frame_size);
  1019.  
  1020.     /* decode tile information */
  1021.     if ((ret = decode_tilehdr(s))) {
  1022.         s->packet_loss = 1;
  1023.         av_frame_unref(s->frame);
  1024.         return ret;
  1025.     }
  1026.  
  1027.     /* read drc info */
  1028.     if (s->dynamic_range_compression)
  1029.         s->drc_gain = get_bits(gb, 8);
  1030.  
  1031.     /* no idea what these are for, might be the number of samples
  1032.        that need to be skipped at the beginning or end of a stream */
  1033.     if (get_bits1(gb)) {
  1034.         int av_unused skip;
  1035.  
  1036.         /* usually true for the first frame */
  1037.         if (get_bits1(gb)) {
  1038.             skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
  1039.             ff_dlog(s->avctx, "start skip: %i\n", skip);
  1040.         }
  1041.  
  1042.         /* sometimes true for the last frame */
  1043.         if (get_bits1(gb)) {
  1044.             skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
  1045.             ff_dlog(s->avctx, "end skip: %i\n", skip);
  1046.         }
  1047.  
  1048.     }
  1049.  
  1050.     /* reset subframe states */
  1051.     s->parsed_all_subframes = 0;
  1052.     for (i = 0; i < s->num_channels; i++) {
  1053.         s->channel[i].decoded_samples = 0;
  1054.         s->channel[i].cur_subframe    = 0;
  1055.     }
  1056.  
  1057.     /* decode all subframes */
  1058.     while (!s->parsed_all_subframes) {
  1059.         int decoded_samples = s->channel[0].decoded_samples;
  1060.         if (decode_subframe(s) < 0) {
  1061.             s->packet_loss = 1;
  1062.             if (s->frame->nb_samples)
  1063.                 s->frame->nb_samples = decoded_samples;
  1064.             return 0;
  1065.         }
  1066.     }
  1067.  
  1068.     ff_dlog(s->avctx, "Frame done\n");
  1069.  
  1070.     s->skip_frame = 0;
  1071.  
  1072.     if (s->len_prefix) {
  1073.         if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
  1074.             /* FIXME: not sure if this is always an error */
  1075.             av_log(s->avctx, AV_LOG_ERROR,
  1076.                    "frame[%"PRIu32"] would have to skip %i bits\n",
  1077.                    s->frame_num,
  1078.                    len - (get_bits_count(gb) - s->frame_offset) - 1);
  1079.             s->packet_loss = 1;
  1080.             return 0;
  1081.         }
  1082.  
  1083.         /* skip the rest of the frame data */
  1084.         skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
  1085.     }
  1086.  
  1087.     /* decode trailer bit */
  1088.     more_frames = get_bits1(gb);
  1089.     ++s->frame_num;
  1090.     return more_frames;
  1091. }
  1092.  
  1093. /**
  1094.  * @brief Calculate remaining input buffer length.
  1095.  * @param s  codec context
  1096.  * @param gb bitstream reader context
  1097.  * @return remaining size in bits
  1098.  */
  1099. static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
  1100. {
  1101.     return s->buf_bit_size - get_bits_count(gb);
  1102. }
  1103.  
  1104. /**
  1105.  * @brief Fill the bit reservoir with a (partial) frame.
  1106.  * @param s      codec context
  1107.  * @param gb     bitstream reader context
  1108.  * @param len    length of the partial frame
  1109.  * @param append decides whether to reset the buffer or not
  1110.  */
  1111. static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
  1112.                       int append)
  1113. {
  1114.     int buflen;
  1115.     PutBitContext tmp;
  1116.  
  1117.     /* when the frame data does not need to be concatenated, the input buffer
  1118.         is reset and additional bits from the previous frame are copied
  1119.         and skipped later so that a fast byte copy is possible */
  1120.  
  1121.     if (!append) {
  1122.         s->frame_offset   = get_bits_count(gb) & 7;
  1123.         s->num_saved_bits = s->frame_offset;
  1124.         init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
  1125.     }
  1126.  
  1127.     buflen = (s->num_saved_bits + len + 8) >> 3;
  1128.  
  1129.     if (len <= 0 || buflen > MAX_FRAMESIZE) {
  1130.         avpriv_request_sample(s->avctx, "Too small input buffer");
  1131.         s->packet_loss = 1;
  1132.         return;
  1133.     }
  1134.  
  1135.     s->num_saved_bits += len;
  1136.     if (!append) {
  1137.         avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
  1138.                          s->num_saved_bits);
  1139.     } else {
  1140.         int align = 8 - (get_bits_count(gb) & 7);
  1141.         align = FFMIN(align, len);
  1142.         put_bits(&s->pb, align, get_bits(gb, align));
  1143.         len -= align;
  1144.         avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
  1145.     }
  1146.     skip_bits_long(gb, len);
  1147.  
  1148.     tmp = s->pb;
  1149.     flush_put_bits(&tmp);
  1150.  
  1151.     init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
  1152.     skip_bits(&s->gb, s->frame_offset);
  1153. }
  1154.  
  1155. static int decode_packet(AVCodecContext *avctx, void *data, int *got_frame_ptr,
  1156.                          AVPacket* avpkt)
  1157. {
  1158.     WmallDecodeCtx *s = avctx->priv_data;
  1159.     GetBitContext* gb  = &s->pgb;
  1160.     const uint8_t* buf = avpkt->data;
  1161.     int buf_size       = avpkt->size;
  1162.     int num_bits_prev_frame, packet_sequence_number, spliced_packet;
  1163.  
  1164.     s->frame->nb_samples = 0;
  1165.  
  1166.     if (s->packet_done || s->packet_loss) {
  1167.         s->packet_done = 0;
  1168.  
  1169.         if (!buf_size)
  1170.             return 0;
  1171.         /* sanity check for the buffer length */
  1172.         if (buf_size < avctx->block_align) {
  1173.             av_log(avctx, AV_LOG_ERROR, "buf size %d invalid\n", buf_size);
  1174.             return AVERROR_INVALIDDATA;
  1175.         }
  1176.  
  1177.         s->next_packet_start = buf_size - avctx->block_align;
  1178.         buf_size             = avctx->block_align;
  1179.         s->buf_bit_size      = buf_size << 3;
  1180.  
  1181.         /* parse packet header */
  1182.         init_get_bits(gb, buf, s->buf_bit_size);
  1183.         packet_sequence_number = get_bits(gb, 4);
  1184.         skip_bits(gb, 1);   // Skip seekable_frame_in_packet, currently ununused
  1185.         spliced_packet = get_bits1(gb);
  1186.         if (spliced_packet)
  1187.             avpriv_request_sample(avctx, "Bitstream splicing");
  1188.  
  1189.         /* get number of bits that need to be added to the previous frame */
  1190.         num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
  1191.  
  1192.         /* check for packet loss */
  1193.         if (!s->packet_loss &&
  1194.             ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
  1195.             s->packet_loss = 1;
  1196.             av_log(avctx, AV_LOG_ERROR,
  1197.                    "Packet loss detected! seq %"PRIx8" vs %x\n",
  1198.                    s->packet_sequence_number, packet_sequence_number);
  1199.         }
  1200.         s->packet_sequence_number = packet_sequence_number;
  1201.  
  1202.         if (num_bits_prev_frame > 0) {
  1203.             int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
  1204.             if (num_bits_prev_frame >= remaining_packet_bits) {
  1205.                 num_bits_prev_frame = remaining_packet_bits;
  1206.                 s->packet_done = 1;
  1207.             }
  1208.  
  1209.             /* Append the previous frame data to the remaining data from the
  1210.              * previous packet to create a full frame. */
  1211.             save_bits(s, gb, num_bits_prev_frame, 1);
  1212.  
  1213.             /* decode the cross packet frame if it is valid */
  1214.             if (num_bits_prev_frame < remaining_packet_bits && !s->packet_loss)
  1215.                 decode_frame(s);
  1216.         } else if (s->num_saved_bits - s->frame_offset) {
  1217.             ff_dlog(avctx, "ignoring %x previously saved bits\n",
  1218.                     s->num_saved_bits - s->frame_offset);
  1219.         }
  1220.  
  1221.         if (s->packet_loss) {
  1222.             /* Reset number of saved bits so that the decoder does not start
  1223.              * to decode incomplete frames in the s->len_prefix == 0 case. */
  1224.             s->num_saved_bits = 0;
  1225.             s->packet_loss    = 0;
  1226.             init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
  1227.         }
  1228.  
  1229.     } else {
  1230.         int frame_size;
  1231.  
  1232.         s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
  1233.         init_get_bits(gb, avpkt->data, s->buf_bit_size);
  1234.         skip_bits(gb, s->packet_offset);
  1235.  
  1236.         if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
  1237.             (frame_size = show_bits(gb, s->log2_frame_size)) &&
  1238.             frame_size <= remaining_bits(s, gb)) {
  1239.             save_bits(s, gb, frame_size, 0);
  1240.             s->packet_done = !decode_frame(s);
  1241.         } else if (!s->len_prefix
  1242.                    && s->num_saved_bits > get_bits_count(&s->gb)) {
  1243.             /* when the frames do not have a length prefix, we don't know the
  1244.              * compressed length of the individual frames however, we know what
  1245.              * part of a new packet belongs to the previous frame therefore we
  1246.              * save the incoming packet first, then we append the "previous
  1247.              * frame" data from the next packet so that we get a buffer that
  1248.              * only contains full frames */
  1249.             s->packet_done = !decode_frame(s);
  1250.         } else {
  1251.             s->packet_done = 1;
  1252.         }
  1253.     }
  1254.  
  1255.     if (s->packet_done && !s->packet_loss &&
  1256.         remaining_bits(s, gb) > 0) {
  1257.         /* save the rest of the data so that it can be decoded
  1258.          * with the next packet */
  1259.         save_bits(s, gb, remaining_bits(s, gb), 0);
  1260.     }
  1261.  
  1262.     *got_frame_ptr   = s->frame->nb_samples > 0;
  1263.     av_frame_move_ref(data, s->frame);
  1264.  
  1265.     s->packet_offset = get_bits_count(gb) & 7;
  1266.  
  1267.     return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
  1268. }
  1269.  
  1270. static void flush(AVCodecContext *avctx)
  1271. {
  1272.     WmallDecodeCtx *s    = avctx->priv_data;
  1273.     s->packet_loss       = 1;
  1274.     s->packet_done       = 0;
  1275.     s->num_saved_bits    = 0;
  1276.     s->frame_offset      = 0;
  1277.     s->next_packet_start = 0;
  1278.     s->cdlms[0][0].order = 0;
  1279.     s->frame->nb_samples = 0;
  1280.     init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
  1281. }
  1282.  
  1283. static av_cold int decode_close(AVCodecContext *avctx)
  1284. {
  1285.     WmallDecodeCtx *s = avctx->priv_data;
  1286.  
  1287.     av_frame_free(&s->frame);
  1288.  
  1289.     return 0;
  1290. }
  1291.  
  1292. AVCodec ff_wmalossless_decoder = {
  1293.     .name           = "wmalossless",
  1294.     .long_name      = NULL_IF_CONFIG_SMALL("Windows Media Audio Lossless"),
  1295.     .type           = AVMEDIA_TYPE_AUDIO,
  1296.     .id             = AV_CODEC_ID_WMALOSSLESS,
  1297.     .priv_data_size = sizeof(WmallDecodeCtx),
  1298.     .init           = decode_init,
  1299.     .close          = decode_close,
  1300.     .decode         = decode_packet,
  1301.     .flush          = flush,
  1302.     .capabilities   = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
  1303.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
  1304.                                                       AV_SAMPLE_FMT_S32P,
  1305.                                                       AV_SAMPLE_FMT_NONE },
  1306. };
  1307.