x86: huffyuvdsp: Move inline assembly to init file
[ffmpeg.git] / libavcodec / mlpdec.c
index d3c9c2b..6baf4c1 100644 (file)
@@ -2,72 +2,86 @@
  * MLP decoder
  * Copyright (c) 2007-2008 Ian Caulfield
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 /**
- * @file mlpdec.c
+ * @file
  * MLP decoder
  */
 
 #include <stdint.h>
 
 #include "avcodec.h"
+#include "libavutil/internal.h"
 #include "libavutil/intreadwrite.h"
-#include "bitstream.h"
+#include "libavutil/channel_layout.h"
+#include "get_bits.h"
+#include "internal.h"
 #include "libavutil/crc.h"
 #include "parser.h"
 #include "mlp_parser.h"
+#include "mlpdsp.h"
 #include "mlp.h"
+#include "config.h"
 
 /** number of bits used for VLC lookup - longest Huffman code is 9 */
+#if ARCH_ARM
+#define VLC_BITS            5
+#define VLC_STATIC_SIZE     64
+#else
 #define VLC_BITS            9
-
-
-static const char* sample_message =
-    "Please file a bug report following the instructions at "
-    "http://ffmpeg.mplayerhq.hu/bugreports.html and include "
-    "a sample of this file.";
+#define VLC_STATIC_SIZE     512
+#endif
 
 typedef struct SubStream {
-    //! Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
+    /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
     uint8_t     restart_seen;
 
     //@{
     /** restart header data */
-    //! The type of noise to be used in the rematrix stage.
+    /// The type of noise to be used in the rematrix stage.
     uint16_t    noise_type;
 
-    //! The index of the first channel coded in this substream.
+    /// The index of the first channel coded in this substream.
     uint8_t     min_channel;
-    //! The index of the last channel coded in this substream.
+    /// The index of the last channel coded in this substream.
     uint8_t     max_channel;
-    //! The number of channels input into the rematrix stage.
+    /// The number of channels input into the rematrix stage.
     uint8_t     max_matrix_channel;
+    /// For each channel output by the matrix, the output channel to map it to
+    uint8_t     ch_assign[MAX_CHANNELS];
+    /// The channel layout for this substream
+    uint64_t    ch_layout;
+    /// The matrix encoding mode for this substream
+    enum AVMatrixEncoding matrix_encoding;
+
+    /// Channel coding parameters for channels in the substream
+    ChannelParams channel_params[MAX_CHANNELS];
 
-    //! The left shift applied to random noise in 0x31ea substreams.
+    /// The left shift applied to random noise in 0x31ea substreams.
     uint8_t     noise_shift;
-    //! The current seed value for the pseudorandom noise generator(s).
+    /// The current seed value for the pseudorandom noise generator(s).
     uint32_t    noisegen_seed;
 
-    //! Set if the substream contains extra info to check the size of VLC blocks.
+    /// Set if the substream contains extra info to check the size of VLC blocks.
     uint8_t     data_check_present;
 
-    //! Bitmask of which parameter sets are conveyed in a decoding parameter block.
+    /// Bitmask of which parameter sets are conveyed in a decoding parameter block.
     uint8_t     param_presence_flags;
 #define PARAM_BLOCKSIZE     (1 << 7)
 #define PARAM_MATRIX        (1 << 6)
@@ -76,37 +90,38 @@ typedef struct SubStream {
 #define PARAM_FIR           (1 << 3)
 #define PARAM_IIR           (1 << 2)
 #define PARAM_HUFFOFFSET    (1 << 1)
+#define PARAM_PRESENCE      (1 << 0)
     //@}
 
     //@{
     /** matrix data */
 
-    //! Number of matrices to be applied.
+    /// Number of matrices to be applied.
     uint8_t     num_primitive_matrices;
 
-    //! matrix output channel
+    /// matrix output channel
     uint8_t     matrix_out_ch[MAX_MATRICES];
 
-    //! Whether the LSBs of the matrix output are encoded in the bitstream.
+    /// Whether the LSBs of the matrix output are encoded in the bitstream.
     uint8_t     lsb_bypass[MAX_MATRICES];
-    //! Matrix coefficients, stored as 2.14 fixed point.
-    int32_t     matrix_coeff[MAX_MATRICES][MAX_CHANNELS+2];
-    //! Left shift to apply to noise values in 0x31eb substreams.
+    /// Matrix coefficients, stored as 2.14 fixed point.
+    int32_t     matrix_coeff[MAX_MATRICES][MAX_CHANNELS];
+    /// Left shift to apply to noise values in 0x31eb substreams.
     uint8_t     matrix_noise_shift[MAX_MATRICES];
     //@}
 
-    //! Left shift to apply to Huffman-decoded residuals.
+    /// Left shift to apply to Huffman-decoded residuals.
     uint8_t     quant_step_size[MAX_CHANNELS];
 
-    //! number of PCM samples in current audio block
+    /// number of PCM samples in current audio block
     uint16_t    blocksize;
-    //! Number of PCM samples decoded so far in this frame.
+    /// Number of PCM samples decoded so far in this frame.
     uint16_t    blockpos;
 
-    //! Left shift to apply to decoded PCM values to get final 24-bit output.
+    /// Left shift to apply to decoded PCM values to get final 24-bit output.
     int8_t      output_shift[MAX_CHANNELS];
 
-    //! Running XOR of all output samples.
+    /// Running XOR of all output samples.
     int32_t     lossless_check_data;
 
 } SubStream;
@@ -114,44 +129,82 @@ typedef struct SubStream {
 typedef struct MLPDecodeContext {
     AVCodecContext *avctx;
 
-    //! Set if a valid major sync block has been read. Otherwise no decoding is possible.
+    /// Current access unit being read has a major sync.
+    int         is_major_sync_unit;
+
+    /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
     uint8_t     params_valid;
 
-    //! Number of substreams contained within this stream.
+    /// Number of substreams contained within this stream.
     uint8_t     num_substreams;
 
-    //! Index of the last substream to decode - further substreams are skipped.
+    /// Index of the last substream to decode - further substreams are skipped.
     uint8_t     max_decoded_substream;
 
-    //! number of PCM samples contained in each frame
+    /// number of PCM samples contained in each frame
     int         access_unit_size;
-    //! next power of two above the number of samples in each frame
+    /// next power of two above the number of samples in each frame
     int         access_unit_size_pow2;
 
     SubStream   substream[MAX_SUBSTREAMS];
 
-    ChannelParams channel_params[MAX_CHANNELS];
+    int         matrix_changed;
+    int         filter_changed[MAX_CHANNELS][NUM_FILTERS];
 
     int8_t      noise_buffer[MAX_BLOCKSIZE_POW2];
     int8_t      bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
-    int32_t     sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS+2];
+    int32_t     sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
+
+    MLPDSPContext dsp;
 } MLPDecodeContext;
 
+static const uint64_t thd_channel_order[] = {
+    AV_CH_FRONT_LEFT, AV_CH_FRONT_RIGHT,                     // LR
+    AV_CH_FRONT_CENTER,                                      // C
+    AV_CH_LOW_FREQUENCY,                                     // LFE
+    AV_CH_SIDE_LEFT, AV_CH_SIDE_RIGHT,                       // LRs
+    AV_CH_TOP_FRONT_LEFT, AV_CH_TOP_FRONT_RIGHT,             // LRvh
+    AV_CH_FRONT_LEFT_OF_CENTER, AV_CH_FRONT_RIGHT_OF_CENTER, // LRc
+    AV_CH_BACK_LEFT, AV_CH_BACK_RIGHT,                       // LRrs
+    AV_CH_BACK_CENTER,                                       // Cs
+    AV_CH_TOP_CENTER,                                        // Ts
+    AV_CH_SURROUND_DIRECT_LEFT, AV_CH_SURROUND_DIRECT_RIGHT, // LRsd
+    AV_CH_WIDE_LEFT, AV_CH_WIDE_RIGHT,                       // LRw
+    AV_CH_TOP_FRONT_CENTER,                                  // Cvh
+    AV_CH_LOW_FREQUENCY_2,                                   // LFE2
+};
+
+static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
+                                                   int index)
+{
+    int i;
+
+    if (av_get_channel_layout_nb_channels(channel_layout) <= index)
+        return 0;
+
+    for (i = 0; i < FF_ARRAY_ELEMS(thd_channel_order); i++)
+        if (channel_layout & thd_channel_order[i] && !index--)
+            return thd_channel_order[i];
+    return 0;
+}
+
 static VLC huff_vlc[3];
 
 /** Initialize static data, constant between all invocations of the codec. */
 
-static av_cold void init_static()
+static av_cold void init_static(void)
 {
-    INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
-                &ff_mlp_huffman_tables[0][0][1], 2, 1,
-                &ff_mlp_huffman_tables[0][0][0], 2, 1, 512);
-    INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
-                &ff_mlp_huffman_tables[1][0][1], 2, 1,
-                &ff_mlp_huffman_tables[1][0][0], 2, 1, 512);
-    INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
-                &ff_mlp_huffman_tables[2][0][1], 2, 1,
-                &ff_mlp_huffman_tables[2][0][0], 2, 1, 512);
+    if (!huff_vlc[0].bits) {
+        INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
+                    &ff_mlp_huffman_tables[0][0][1], 2, 1,
+                    &ff_mlp_huffman_tables[0][0][0], 2, 1, VLC_STATIC_SIZE);
+        INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
+                    &ff_mlp_huffman_tables[1][0][1], 2, 1,
+                    &ff_mlp_huffman_tables[1][0][0], 2, 1, VLC_STATIC_SIZE);
+        INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
+                    &ff_mlp_huffman_tables[2][0][1], 2, 1,
+                    &ff_mlp_huffman_tables[2][0][0], 2, 1, VLC_STATIC_SIZE);
+    }
 
     ff_mlp_init_crc();
 }
@@ -159,8 +212,8 @@ static av_cold void init_static()
 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
                                           unsigned int substr, unsigned int ch)
 {
-    ChannelParams *cp = &m->channel_params[ch];
     SubStream *s = &m->substream[substr];
+    ChannelParams *cp = &s->channel_params[ch];
     int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
     int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
     int32_t sign_huff_offset = cp->huff_offset;
@@ -188,7 +241,7 @@ static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
             m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
 
     for (channel = s->min_channel; channel <= s->max_channel; channel++) {
-        ChannelParams *cp = &m->channel_params[channel];
+        ChannelParams *cp = &s->channel_params[channel];
         int codebook = cp->codebook;
         int quant_step_size = s->quant_step_size[channel];
         int lsb_bits = cp->huff_lsbs - quant_step_size;
@@ -199,7 +252,7 @@ static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
                             VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
 
         if (result < 0)
-            return -1;
+            return AVERROR_INVALIDDATA;
 
         if (lsb_bits > 0)
             result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
@@ -222,6 +275,7 @@ static av_cold int mlp_decode_init(AVCodecContext *avctx)
     m->avctx = avctx;
     for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
         m->substream[substr].lossless_check_data = 0xffffffff;
+    ff_mlpdsp_init(&m->dsp);
 
     return 0;
 }
@@ -234,57 +288,62 @@ static av_cold int mlp_decode_init(AVCodecContext *avctx)
 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
 {
     MLPHeaderInfo mh;
-    int substr;
+    int substr, ret;
 
-    if (ff_mlp_read_major_sync(m->avctx, &mh, gb) != 0)
-        return -1;
+    if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
+        return ret;
 
     if (mh.group1_bits == 0) {
         av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
     if (mh.group2_bits > mh.group1_bits) {
         av_log(m->avctx, AV_LOG_ERROR,
                "Channel group 2 cannot have more bits per sample than group 1.\n");
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
     if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
         av_log(m->avctx, AV_LOG_ERROR,
                "Channel groups with differing sample rates are not currently supported.\n");
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
     if (mh.group1_samplerate == 0) {
         av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
     if (mh.group1_samplerate > MAX_SAMPLERATE) {
         av_log(m->avctx, AV_LOG_ERROR,
                "Sampling rate %d is greater than the supported maximum (%d).\n",
                mh.group1_samplerate, MAX_SAMPLERATE);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
     if (mh.access_unit_size > MAX_BLOCKSIZE) {
         av_log(m->avctx, AV_LOG_ERROR,
                "Block size %d is greater than the supported maximum (%d).\n",
                mh.access_unit_size, MAX_BLOCKSIZE);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
     if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
         av_log(m->avctx, AV_LOG_ERROR,
                "Block size pow2 %d is greater than the supported maximum (%d).\n",
                mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
     if (mh.num_substreams == 0)
-        return -1;
+        return AVERROR_INVALIDDATA;
+    if (m->avctx->codec_id == AV_CODEC_ID_MLP && mh.num_substreams > 2) {
+        av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
+        return AVERROR_INVALIDDATA;
+    }
     if (mh.num_substreams > MAX_SUBSTREAMS) {
-        av_log(m->avctx, AV_LOG_ERROR,
-               "Number of substreams %d is larger than the maximum supported "
-               "by the decoder. %s\n", mh.num_substreams, sample_message);
-        return -1;
+        avpriv_request_sample(m->avctx,
+                              "%d substreams (more than the "
+                              "maximum supported by the decoder)",
+                              mh.num_substreams);
+        return AVERROR_PATCHWELCOME;
     }
 
     m->access_unit_size      = mh.access_unit_size;
@@ -298,14 +357,76 @@ static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
 
     m->avctx->bits_per_raw_sample = mh.group1_bits;
     if (mh.group1_bits > 16)
-        m->avctx->sample_fmt = SAMPLE_FMT_S32;
+        m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
     else
-        m->avctx->sample_fmt = SAMPLE_FMT_S16;
+        m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
+    m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(m->substream[m->max_decoded_substream].ch_assign,
+                                                           m->substream[m->max_decoded_substream].output_shift,
+                                                           m->substream[m->max_decoded_substream].max_matrix_channel,
+                                                           m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
 
     m->params_valid = 1;
     for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
         m->substream[substr].restart_seen = 0;
 
+    /* Set the layout for each substream. When there's more than one, the first
+     * substream is Stereo. Subsequent substreams' layouts are indicated in the
+     * major sync. */
+    if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
+        if ((substr = (mh.num_substreams > 1)))
+            m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
+        m->substream[substr].ch_layout = mh.channel_layout_mlp;
+    } else {
+        if ((substr = (mh.num_substreams > 1)))
+            m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
+        if (mh.num_substreams > 2)
+            if (mh.channel_layout_thd_stream2)
+                m->substream[2].ch_layout = mh.channel_layout_thd_stream2;
+            else
+                m->substream[2].ch_layout = mh.channel_layout_thd_stream1;
+        m->substream[substr].ch_layout = mh.channel_layout_thd_stream1;
+    }
+
+    /* Parse the TrueHD decoder channel modifiers and set each substream's
+     * AVMatrixEncoding accordingly.
+     *
+     * The meaning of the modifiers depends on the channel layout:
+     *
+     * - THD_CH_MODIFIER_LTRT, THD_CH_MODIFIER_LBINRBIN only apply to 2-channel
+     *
+     * - THD_CH_MODIFIER_MONO applies to 1-channel or 2-channel (dual mono)
+     *
+     * - THD_CH_MODIFIER_SURROUNDEX, THD_CH_MODIFIER_NOTSURROUNDEX only apply to
+     *   layouts with an Ls/Rs channel pair
+     */
+    for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
+        m->substream[substr].matrix_encoding = AV_MATRIX_ENCODING_NONE;
+    if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
+        if (mh.num_substreams > 2 &&
+            mh.channel_layout_thd_stream2 & AV_CH_SIDE_LEFT &&
+            mh.channel_layout_thd_stream2 & AV_CH_SIDE_RIGHT &&
+            mh.channel_modifier_thd_stream2 == THD_CH_MODIFIER_SURROUNDEX)
+            m->substream[2].matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
+
+        if (mh.num_substreams > 1 &&
+            mh.channel_layout_thd_stream1 & AV_CH_SIDE_LEFT &&
+            mh.channel_layout_thd_stream1 & AV_CH_SIDE_RIGHT &&
+            mh.channel_modifier_thd_stream1 == THD_CH_MODIFIER_SURROUNDEX)
+            m->substream[1].matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
+
+        if (mh.num_substreams > 0)
+            switch (mh.channel_modifier_thd_stream0) {
+            case THD_CH_MODIFIER_LTRT:
+                m->substream[0].matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
+                break;
+            case THD_CH_MODIFIER_LBINRBIN:
+                m->substream[0].matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
+                break;
+            default:
+                break;
+            }
+    }
+
     return 0;
 }
 
@@ -322,36 +443,85 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
     uint8_t checksum;
     uint8_t lossless_check;
     int start_count = get_bits_count(gbp);
+    int min_channel, max_channel, max_matrix_channel;
+    const int std_max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
+                                     ? MAX_MATRIX_CHANNEL_MLP
+                                     : MAX_MATRIX_CHANNEL_TRUEHD;
 
     sync_word = get_bits(gbp, 13);
 
     if (sync_word != 0x31ea >> 1) {
         av_log(m->avctx, AV_LOG_ERROR,
                "restart header sync incorrect (got 0x%04x)\n", sync_word);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
+
     s->noise_type = get_bits1(gbp);
 
+    if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
+        av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
+        return AVERROR_INVALIDDATA;
+    }
+
     skip_bits(gbp, 16); /* Output timestamp */
 
-    s->min_channel        = get_bits(gbp, 4);
-    s->max_channel        = get_bits(gbp, 4);
-    s->max_matrix_channel = get_bits(gbp, 4);
+    min_channel        = get_bits(gbp, 4);
+    max_channel        = get_bits(gbp, 4);
+    max_matrix_channel = get_bits(gbp, 4);
 
-    if (s->min_channel > s->max_channel) {
+    if (max_matrix_channel > std_max_matrix_channel) {
+        av_log(m->avctx, AV_LOG_ERROR,
+               "Max matrix channel cannot be greater than %d.\n",
+               max_matrix_channel);
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (max_channel != max_matrix_channel) {
+        av_log(m->avctx, AV_LOG_ERROR,
+               "Max channel must be equal max matrix channel.\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    /* This should happen for TrueHD streams with >6 channels and MLP's noise
+     * type. It is not yet known if this is allowed. */
+    if (s->max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
+        avpriv_request_sample(m->avctx,
+                              "%d channels (more than the "
+                              "maximum supported by the decoder)",
+                              s->max_channel + 2);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    if (min_channel > max_channel) {
         av_log(m->avctx, AV_LOG_ERROR,
                "Substream min channel cannot be greater than max channel.\n");
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
-    if (m->avctx->request_channels > 0
-        && s->max_channel + 1 >= m->avctx->request_channels
-        && substr < m->max_decoded_substream) {
-        av_log(m->avctx, AV_LOG_INFO,
-               "Extracting %d channel downmix from substream %d. "
+    s->min_channel        = min_channel;
+    s->max_channel        = max_channel;
+    s->max_matrix_channel = max_matrix_channel;
+
+#if FF_API_REQUEST_CHANNELS
+FF_DISABLE_DEPRECATION_WARNINGS
+    if (m->avctx->request_channels > 0 &&
+        m->avctx->request_channels <= s->max_channel + 1 &&
+        m->max_decoded_substream > substr) {
+        av_log(m->avctx, AV_LOG_DEBUG,
+               "Extracting %d-channel downmix from substream %d. "
                "Further substreams will be skipped.\n",
                s->max_channel + 1, substr);
         m->max_decoded_substream = substr;
+    } else
+FF_ENABLE_DEPRECATION_WARNINGS
+#endif
+    if (m->avctx->request_channel_layout && (s->ch_layout & m->avctx->request_channel_layout) ==
+        m->avctx->request_channel_layout && m->max_decoded_substream > substr) {
+        av_log(m->avctx, AV_LOG_DEBUG,
+               "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
+               "Further substreams will be skipped.\n",
+               s->max_channel + 1, s->ch_layout, substr);
+        m->max_decoded_substream = substr;
     }
 
     s->noise_shift   = get_bits(gbp,  4);
@@ -368,23 +538,27 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
             av_log(m->avctx, AV_LOG_WARNING,
                    "Lossless check failed - expected %02x, calculated %02x.\n",
                    lossless_check, tmp);
-        else
-            dprintf(m->avctx, "Lossless check passed for substream %d (%x).\n",
-                    substr, tmp);
     }
 
     skip_bits(gbp, 16);
 
+    memset(s->ch_assign, 0, sizeof(s->ch_assign));
+
     for (ch = 0; ch <= s->max_matrix_channel; ch++) {
         int ch_assign = get_bits(gbp, 6);
-        dprintf(m->avctx, "ch_assign[%d][%d] = %d\n", substr, ch,
-                ch_assign);
-        if (ch_assign != ch) {
-            av_log(m->avctx, AV_LOG_ERROR,
-                   "Non-1:1 channel assignments are used in this stream. %s\n",
-                   sample_message);
-            return -1;
+        if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
+            uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
+                                                                  ch_assign);
+            ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
+                                                            channel);
+        }
+        if (ch_assign > s->max_matrix_channel) {
+            avpriv_request_sample(m->avctx,
+                                  "Assignment of matrix channel %d to invalid output channel %d",
+                                  ch, ch_assign);
+            return AVERROR_PATCHWELCOME;
         }
+        s->ch_assign[ch_assign] = ch;
     }
 
     checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
@@ -402,7 +576,7 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
     memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
 
     for (ch = s->min_channel; ch <= s->max_channel; ch++) {
-        ChannelParams *cp = &m->channel_params[ch];
+        ChannelParams *cp = &s->channel_params[ch];
         cp->filter_params[FIR].order = 0;
         cp->filter_params[IIR].order = 0;
         cp->filter_params[FIR].shift = 0;
@@ -416,7 +590,12 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
     }
 
     if (substr == m->max_decoded_substream) {
-        m->avctx->channels = s->max_channel + 1;
+        m->avctx->channels       = s->max_matrix_channel + 1;
+        m->avctx->channel_layout = s->ch_layout;
+        m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
+                                                               s->output_shift,
+                                                               s->max_matrix_channel,
+                                                               m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
     }
 
     return 0;
@@ -425,25 +604,34 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
 /** Read parameters for one of the prediction filters. */
 
 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
-                              unsigned int channel, unsigned int filter)
+                              unsigned int substr, unsigned int channel,
+                              unsigned int filter)
 {
-    FilterParams *fp = &m->channel_params[channel].filter_params[filter];
+    SubStream *s = &m->substream[substr];
+    FilterParams *fp = &s->channel_params[channel].filter_params[filter];
+    const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
     const char fchar = filter ? 'I' : 'F';
     int i, order;
 
     // Filter is 0 for FIR, 1 for IIR.
     assert(filter < 2);
 
+    if (m->filter_changed[channel][filter]++ > 1) {
+        av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
+        return AVERROR_INVALIDDATA;
+    }
+
     order = get_bits(gbp, 4);
-    if (order > MAX_FILTER_ORDER) {
+    if (order > max_order) {
         av_log(m->avctx, AV_LOG_ERROR,
                "%cIR filter order %d is greater than maximum %d.\n",
-               fchar, order, MAX_FILTER_ORDER);
-        return -1;
+               fchar, order, max_order);
+        return AVERROR_INVALIDDATA;
     }
     fp->order = order;
 
     if (order > 0) {
+        int32_t *fcoeff = s->channel_params[channel].coeff[filter];
         int coeff_bits, coeff_shift;
 
         fp->shift = get_bits(gbp, 4);
@@ -454,17 +642,17 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
             av_log(m->avctx, AV_LOG_ERROR,
                    "%cIR filter coeff_bits must be between 1 and 16.\n",
                    fchar);
-            return -1;
+            return AVERROR_INVALIDDATA;
         }
         if (coeff_bits + coeff_shift > 16) {
             av_log(m->avctx, AV_LOG_ERROR,
                    "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
                    fchar);
-            return -1;
+            return AVERROR_INVALIDDATA;
         }
 
         for (i = 0; i < order; i++)
-            fp->coeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
+            fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
 
         if (get_bits1(gbp)) {
             int state_bits, state_shift;
@@ -472,7 +660,7 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
             if (filter == FIR) {
                 av_log(m->avctx, AV_LOG_ERROR,
                        "FIR filter has state data specified.\n");
-                return -1;
+                return AVERROR_INVALIDDATA;
             }
 
             state_bits  = get_bits(gbp, 4);
@@ -488,6 +676,126 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
     return 0;
 }
 
+/** Read parameters for primitive matrices. */
+
+static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
+{
+    SubStream *s = &m->substream[substr];
+    unsigned int mat, ch;
+    const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
+                                     ? MAX_MATRICES_MLP
+                                     : MAX_MATRICES_TRUEHD;
+
+    if (m->matrix_changed++ > 1) {
+        av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    s->num_primitive_matrices = get_bits(gbp, 4);
+
+    if (s->num_primitive_matrices > max_primitive_matrices) {
+        av_log(m->avctx, AV_LOG_ERROR,
+               "Number of primitive matrices cannot be greater than %d.\n",
+               max_primitive_matrices);
+        return AVERROR_INVALIDDATA;
+    }
+
+    for (mat = 0; mat < s->num_primitive_matrices; mat++) {
+        int frac_bits, max_chan;
+        s->matrix_out_ch[mat] = get_bits(gbp, 4);
+        frac_bits             = get_bits(gbp, 4);
+        s->lsb_bypass   [mat] = get_bits1(gbp);
+
+        if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
+            av_log(m->avctx, AV_LOG_ERROR,
+                    "Invalid channel %d specified as output from matrix.\n",
+                    s->matrix_out_ch[mat]);
+            return AVERROR_INVALIDDATA;
+        }
+        if (frac_bits > 14) {
+            av_log(m->avctx, AV_LOG_ERROR,
+                    "Too many fractional bits specified.\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        max_chan = s->max_matrix_channel;
+        if (!s->noise_type)
+            max_chan+=2;
+
+        for (ch = 0; ch <= max_chan; ch++) {
+            int coeff_val = 0;
+            if (get_bits1(gbp))
+                coeff_val = get_sbits(gbp, frac_bits + 2);
+
+            s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
+        }
+
+        if (s->noise_type)
+            s->matrix_noise_shift[mat] = get_bits(gbp, 4);
+        else
+            s->matrix_noise_shift[mat] = 0;
+    }
+
+    return 0;
+}
+
+/** Read channel parameters. */
+
+static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
+                               GetBitContext *gbp, unsigned int ch)
+{
+    SubStream *s = &m->substream[substr];
+    ChannelParams *cp = &s->channel_params[ch];
+    FilterParams *fir = &cp->filter_params[FIR];
+    FilterParams *iir = &cp->filter_params[IIR];
+    int ret;
+
+    if (s->param_presence_flags & PARAM_FIR)
+        if (get_bits1(gbp))
+            if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
+                return ret;
+
+    if (s->param_presence_flags & PARAM_IIR)
+        if (get_bits1(gbp))
+            if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
+                return ret;
+
+    if (fir->order + iir->order > 8) {
+        av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (fir->order && iir->order &&
+        fir->shift != iir->shift) {
+        av_log(m->avctx, AV_LOG_ERROR,
+                "FIR and IIR filters must use the same precision.\n");
+        return AVERROR_INVALIDDATA;
+    }
+    /* The FIR and IIR filters must have the same precision.
+     * To simplify the filtering code, only the precision of the
+     * FIR filter is considered. If only the IIR filter is employed,
+     * the FIR filter precision is set to that of the IIR filter, so
+     * that the filtering code can use it. */
+    if (!fir->order && iir->order)
+        fir->shift = iir->shift;
+
+    if (s->param_presence_flags & PARAM_HUFFOFFSET)
+        if (get_bits1(gbp))
+            cp->huff_offset = get_sbits(gbp, 15);
+
+    cp->codebook  = get_bits(gbp, 2);
+    cp->huff_lsbs = get_bits(gbp, 5);
+
+    if (cp->huff_lsbs > 24) {
+        av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
+
+    return 0;
+}
+
 /** Read decoding parameters that change more often than those in the restart
  *  header. */
 
@@ -495,123 +803,53 @@ static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
                                 unsigned int substr)
 {
     SubStream *s = &m->substream[substr];
-    unsigned int mat, ch;
+    unsigned int ch;
+    int ret;
 
-    if (get_bits1(gbp))
-        s->param_presence_flags = get_bits(gbp, 8);
+    if (s->param_presence_flags & PARAM_PRESENCE)
+        if (get_bits1(gbp))
+            s->param_presence_flags = get_bits(gbp, 8);
 
     if (s->param_presence_flags & PARAM_BLOCKSIZE)
         if (get_bits1(gbp)) {
             s->blocksize = get_bits(gbp, 9);
-            if (s->blocksize > MAX_BLOCKSIZE) {
-                av_log(m->avctx, AV_LOG_ERROR, "block size too large\n");
+            if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
+                av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
                 s->blocksize = 0;
-                return -1;
+                return AVERROR_INVALIDDATA;
             }
         }
 
     if (s->param_presence_flags & PARAM_MATRIX)
-        if (get_bits1(gbp)) {
-            s->num_primitive_matrices = get_bits(gbp, 4);
-
-            for (mat = 0; mat < s->num_primitive_matrices; mat++) {
-                int frac_bits, max_chan;
-                s->matrix_out_ch[mat] = get_bits(gbp, 4);
-                frac_bits             = get_bits(gbp, 4);
-                s->lsb_bypass   [mat] = get_bits1(gbp);
-
-                if (s->matrix_out_ch[mat] > s->max_channel) {
-                    av_log(m->avctx, AV_LOG_ERROR,
-                           "Invalid channel %d specified as output from matrix.\n",
-                           s->matrix_out_ch[mat]);
-                    return -1;
-                }
-                if (frac_bits > 14) {
-                    av_log(m->avctx, AV_LOG_ERROR,
-                           "Too many fractional bits specified.\n");
-                    return -1;
-                }
-
-                max_chan = s->max_matrix_channel;
-                if (!s->noise_type)
-                    max_chan+=2;
-
-                for (ch = 0; ch <= max_chan; ch++) {
-                    int coeff_val = 0;
-                    if (get_bits1(gbp))
-                        coeff_val = get_sbits(gbp, frac_bits + 2);
-
-                    s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
-                }
-
-                if (s->noise_type)
-                    s->matrix_noise_shift[mat] = get_bits(gbp, 4);
-                else
-                    s->matrix_noise_shift[mat] = 0;
-            }
-        }
+        if (get_bits1(gbp))
+            if ((ret = read_matrix_params(m, substr, gbp)) < 0)
+                return ret;
 
     if (s->param_presence_flags & PARAM_OUTSHIFT)
-        if (get_bits1(gbp))
-            for (ch = 0; ch <= s->max_matrix_channel; ch++) {
-                s->output_shift[ch] = get_bits(gbp, 4);
-                dprintf(m->avctx, "output shift[%d] = %d\n",
-                        ch, s->output_shift[ch]);
-                /* TODO: validate */
-            }
+        if (get_bits1(gbp)) {
+            for (ch = 0; ch <= s->max_matrix_channel; ch++)
+                s->output_shift[ch] = get_sbits(gbp, 4);
+            if (substr == m->max_decoded_substream)
+                m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
+                                                                       s->output_shift,
+                                                                       s->max_matrix_channel,
+                                                                       m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
+        }
 
     if (s->param_presence_flags & PARAM_QUANTSTEP)
         if (get_bits1(gbp))
             for (ch = 0; ch <= s->max_channel; ch++) {
-                ChannelParams *cp = &m->channel_params[ch];
+                ChannelParams *cp = &s->channel_params[ch];
 
                 s->quant_step_size[ch] = get_bits(gbp, 4);
-                /* TODO: validate */
 
                 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
             }
 
     for (ch = s->min_channel; ch <= s->max_channel; ch++)
-        if (get_bits1(gbp)) {
-            ChannelParams *cp = &m->channel_params[ch];
-            FilterParams *fir = &cp->filter_params[FIR];
-            FilterParams *iir = &cp->filter_params[IIR];
-
-            if (s->param_presence_flags & PARAM_FIR)
-                if (get_bits1(gbp))
-                    if (read_filter_params(m, gbp, ch, FIR) < 0)
-                        return -1;
-
-            if (s->param_presence_flags & PARAM_IIR)
-                if (get_bits1(gbp))
-                    if (read_filter_params(m, gbp, ch, IIR) < 0)
-                        return -1;
-
-            if (fir->order && iir->order &&
-                fir->shift != iir->shift) {
-                av_log(m->avctx, AV_LOG_ERROR,
-                       "FIR and IIR filters must use the same precision.\n");
-                return -1;
-            }
-            /* The FIR and IIR filters must have the same precision.
-             * To simplify the filtering code, only the precision of the
-             * FIR filter is considered. If only the IIR filter is employed,
-             * the FIR filter precision is set to that of the IIR filter, so
-             * that the filtering code can use it. */
-            if (!fir->order && iir->order)
-                fir->shift = iir->shift;
-
-            if (s->param_presence_flags & PARAM_HUFFOFFSET)
-                if (get_bits1(gbp))
-                    cp->huff_offset = get_sbits(gbp, 15);
-
-            cp->codebook  = get_bits(gbp, 2);
-            cp->huff_lsbs = get_bits(gbp, 5);
-
-            cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
-
-            /* TODO: validate */
-        }
+        if (get_bits1(gbp))
+            if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
+                return ret;
 
     return 0;
 }
@@ -625,47 +863,25 @@ static void filter_channel(MLPDecodeContext *m, unsigned int substr,
                            unsigned int channel)
 {
     SubStream *s = &m->substream[substr];
-    int32_t filter_state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FILTER_ORDER];
-    FilterParams *fp[NUM_FILTERS] = { &m->channel_params[channel].filter_params[FIR],
-                                      &m->channel_params[channel].filter_params[IIR], };
-    unsigned int filter_shift = fp[FIR]->shift;
+    const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
+    int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
+    int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
+    int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
+    FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
+    FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
+    unsigned int filter_shift = fir->shift;
     int32_t mask = MSB_MASK(s->quant_step_size[channel]);
-    int index = MAX_BLOCKSIZE;
-    int j, i;
-
-    for (j = 0; j < NUM_FILTERS; j++) {
-        memcpy(&filter_state_buffer[j][MAX_BLOCKSIZE], &fp[j]->state[0],
-               MAX_FILTER_ORDER * sizeof(int32_t));
-    }
 
-    for (i = 0; i < s->blocksize; i++) {
-        int32_t residual = m->sample_buffer[i + s->blockpos][channel];
-        unsigned int order;
-        int64_t accum = 0;
-        int32_t result;
+    memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
+    memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
 
-        /* TODO: Move this code to DSPContext? */
+    m->dsp.mlp_filter_channel(firbuf, fircoeff,
+                              fir->order, iir->order,
+                              filter_shift, mask, s->blocksize,
+                              &m->sample_buffer[s->blockpos][channel]);
 
-        for (j = 0; j < NUM_FILTERS; j++)
-            for (order = 0; order < fp[j]->order; order++)
-                accum += (int64_t)filter_state_buffer[j][index + order] *
-                                  fp[j]->coeff[order];
-
-        accum  = accum >> filter_shift;
-        result = (accum + residual) & mask;
-
-        --index;
-
-        filter_state_buffer[FIR][index] = result;
-        filter_state_buffer[IIR][index] = result - accum;
-
-        m->sample_buffer[i + s->blockpos][channel] = result;
-    }
-
-    for (j = 0; j < NUM_FILTERS; j++) {
-        memcpy(&fp[j]->state[0], &filter_state_buffer[j][index],
-               MAX_FILTER_ORDER * sizeof(int32_t));
-    }
+    memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
+    memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
 }
 
 /** Read a block of PCM residual data (or actual if no filtering active). */
@@ -675,30 +891,29 @@ static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
 {
     SubStream *s = &m->substream[substr];
     unsigned int i, ch, expected_stream_pos = 0;
+    int ret;
 
     if (s->data_check_present) {
         expected_stream_pos  = get_bits_count(gbp);
         expected_stream_pos += get_bits(gbp, 16);
-        av_log(m->avctx, AV_LOG_WARNING, "This file contains some features "
-               "we have not tested yet. %s\n", sample_message);
+        avpriv_request_sample(m->avctx,
+                              "Substreams with VLC block size check info");
     }
 
     if (s->blockpos + s->blocksize > m->access_unit_size) {
         av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
     memset(&m->bypassed_lsbs[s->blockpos][0], 0,
            s->blocksize * sizeof(m->bypassed_lsbs[0]));
 
-    for (i = 0; i < s->blocksize; i++) {
-        if (read_huff_channels(m, gbp, substr, i) < 0)
-            return -1;
-    }
+    for (i = 0; i < s->blocksize; i++)
+        if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
+            return ret;
 
-    for (ch = s->min_channel; ch <= s->max_channel; ch++) {
+    for (ch = s->min_channel; ch <= s->max_channel; ch++)
         filter_channel(m, substr, ch);
-    }
 
     s->blockpos += s->blocksize;
 
@@ -784,7 +999,7 @@ static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
 {
     SubStream *s = &m->substream[substr];
-    unsigned int mat, src_ch, i;
+    unsigned int mat;
     unsigned int maxchan;
 
     maxchan = s->max_matrix_channel;
@@ -796,73 +1011,74 @@ static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
     }
 
     for (mat = 0; mat < s->num_primitive_matrices; mat++) {
-        int matrix_noise_shift = s->matrix_noise_shift[mat];
         unsigned int dest_ch = s->matrix_out_ch[mat];
-        int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
-
-        /* TODO: DSPContext? */
-
-        for (i = 0; i < s->blockpos; i++) {
-            int64_t accum = 0;
-            for (src_ch = 0; src_ch <= maxchan; src_ch++) {
-                accum += (int64_t)m->sample_buffer[i][src_ch]
-                                  * s->matrix_coeff[mat][src_ch];
-            }
-            if (matrix_noise_shift) {
-                uint32_t index = s->num_primitive_matrices - mat;
-                index = (i * (index * 2 + 1) + index) & (m->access_unit_size_pow2 - 1);
-                accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
-            }
-            m->sample_buffer[i][dest_ch] = ((accum >> 14) & mask)
-                                             + m->bypassed_lsbs[i][mat];
-        }
+        m->dsp.mlp_rematrix_channel(&m->sample_buffer[0][0],
+                                    s->matrix_coeff[mat],
+                                    &m->bypassed_lsbs[0][mat],
+                                    m->noise_buffer,
+                                    s->num_primitive_matrices - mat,
+                                    dest_ch,
+                                    s->blockpos,
+                                    maxchan,
+                                    s->matrix_noise_shift[mat],
+                                    m->access_unit_size_pow2,
+                                    MSB_MASK(s->quant_step_size[dest_ch]));
     }
 }
 
 /** Write the audio data into the output buffer. */
 
-static int output_data_internal(MLPDecodeContext *m, unsigned int substr,
-                                uint8_t *data, unsigned int *data_size, int is32)
+static int output_data(MLPDecodeContext *m, unsigned int substr,
+                       AVFrame *frame, int *got_frame_ptr)
 {
+    AVCodecContext *avctx = m->avctx;
     SubStream *s = &m->substream[substr];
-    unsigned int i, ch = 0;
-    int32_t *data_32 = (int32_t*) data;
-    int16_t *data_16 = (int16_t*) data;
+    int ret;
+    int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
 
-    if (*data_size < (s->max_channel + 1) * s->blockpos * (is32 ? 4 : 2))
-        return -1;
+    if (m->avctx->channels != s->max_matrix_channel + 1) {
+        av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
+        return AVERROR_INVALIDDATA;
+    }
 
-    for (i = 0; i < s->blockpos; i++) {
-        for (ch = 0; ch <= s->max_channel; ch++) {
-            int32_t sample = m->sample_buffer[i][ch] << s->output_shift[ch];
-            s->lossless_check_data ^= (sample & 0xffffff) << ch;
-            if (is32) *data_32++ = sample << 8;
-            else      *data_16++ = sample >> 8;
-        }
+    if (!s->blockpos) {
+        av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
+        return AVERROR_INVALIDDATA;
     }
 
-    *data_size = i * ch * (is32 ? 4 : 2);
+    /* get output buffer */
+    frame->nb_samples = s->blockpos;
+    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
+        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+        return ret;
+    }
+    s->lossless_check_data = m->dsp.mlp_pack_output(s->lossless_check_data,
+                                                    s->blockpos,
+                                                    m->sample_buffer,
+                                                    frame->data[0],
+                                                    s->ch_assign,
+                                                    s->output_shift,
+                                                    s->max_matrix_channel,
+                                                    is32);
 
-    return 0;
-}
+    /* Update matrix encoding side data */
+    if ((ret = ff_side_data_update_matrix_encoding(frame, s->matrix_encoding)) < 0)
+        return ret;
 
-static int output_data(MLPDecodeContext *m, unsigned int substr,
-                       uint8_t *data, unsigned int *data_size)
-{
-    if (m->avctx->sample_fmt == SAMPLE_FMT_S32)
-        return output_data_internal(m, substr, data, data_size, 1);
-    else
-        return output_data_internal(m, substr, data, data_size, 0);
-}
+    *got_frame_ptr = 1;
 
+    return 0;
+}
 
 /** Read an access unit from the stream.
- *  Returns < 0 on error, 0 if not enough data is present in the input stream
- *  otherwise returns the number of bytes consumed. */
+ *  @return negative on error, 0 if not enough data is present in the input stream,
+ *  otherwise the number of bytes consumed. */
 
-static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
-                            const uint8_t *buf, int buf_size)
+static int read_access_unit(AVCodecContext *avctx, void* data,
+                            int *got_frame_ptr, AVPacket *avpkt)
 {
+    const uint8_t *buf = avpkt->data;
+    int buf_size = avpkt->size;
     MLPDecodeContext *m = avctx->priv_data;
     GetBitContext gb;
     unsigned int length, substr;
@@ -872,38 +1088,40 @@ static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
     uint8_t substream_parity_present[MAX_SUBSTREAMS];
     uint16_t substream_data_len[MAX_SUBSTREAMS];
     uint8_t parity_bits;
+    int ret;
 
     if (buf_size < 4)
         return 0;
 
     length = (AV_RB16(buf) & 0xfff) * 2;
 
-    if (length > buf_size)
-        return -1;
+    if (length < 4 || length > buf_size)
+        return AVERROR_INVALIDDATA;
 
     init_get_bits(&gb, (buf + 4), (length - 4) * 8);
 
+    m->is_major_sync_unit = 0;
     if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
-        dprintf(m->avctx, "Found major sync.\n");
         if (read_major_sync(m, &gb) < 0)
             goto error;
+        m->is_major_sync_unit = 1;
         header_size += 28;
     }
 
     if (!m->params_valid) {
         av_log(m->avctx, AV_LOG_WARNING,
                "Stream parameters not seen; skipping frame.\n");
-        *data_size = 0;
+        *got_frame_ptr = 0;
         return length;
     }
 
     substream_start = 0;
 
     for (substr = 0; substr < m->num_substreams; substr++) {
-        int extraword_present, checkdata_present, end;
+        int extraword_present, checkdata_present, end, nonrestart_substr;
 
         extraword_present = get_bits1(&gb);
-        skip_bits1(&gb);
+        nonrestart_substr = get_bits1(&gb);
         checkdata_present = get_bits1(&gb);
         skip_bits1(&gb);
 
@@ -912,10 +1130,19 @@ static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
         substr_header_size += 2;
 
         if (extraword_present) {
+            if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
+                av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
+                goto error;
+            }
             skip_bits(&gb, 16);
             substr_header_size += 2;
         }
 
+        if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
+            av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
+            goto error;
+        }
+
         if (end + header_size + substr_header_size > length) {
             av_log(m->avctx, AV_LOG_ERROR,
                    "Indicated length of substream %d data goes off end of "
@@ -953,6 +1180,9 @@ static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
         SubStream *s = &m->substream[substr];
         init_get_bits(&gb, buf, substream_data_len[substr] * 8);
 
+        m->matrix_changed = 0;
+        memset(m->filter_changed, 0, sizeof(m->filter_changed));
+
         s->blockpos = 0;
         do {
             if (get_bits1(&gb)) {
@@ -963,90 +1193,103 @@ static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
                     s->restart_seen = 1;
                 }
 
-                if (!s->restart_seen) {
-                    av_log(m->avctx, AV_LOG_ERROR,
-                           "No restart header present in substream %d.\n",
-                           substr);
+                if (!s->restart_seen)
                     goto next_substr;
-                }
-
                 if (read_decoding_params(m, &gb, substr) < 0)
                     goto next_substr;
             }
 
-            if (!s->restart_seen) {
-                av_log(m->avctx, AV_LOG_ERROR,
-                       "No restart header present in substream %d.\n",
-                       substr);
+            if (!s->restart_seen)
                 goto next_substr;
-            }
 
-            if (read_block_data(m, &gb, substr) < 0)
-                return -1;
+            if ((ret = read_block_data(m, &gb, substr)) < 0)
+                return ret;
+
+            if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
+                goto substream_length_mismatch;
 
-        } while ((get_bits_count(&gb) < substream_data_len[substr] * 8)
-                 && get_bits1(&gb) == 0);
+        } while (!get_bits1(&gb));
 
         skip_bits(&gb, (-get_bits_count(&gb)) & 15);
-        if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32 &&
-            (show_bits_long(&gb, 32) == END_OF_STREAM ||
-             show_bits_long(&gb, 20) == 0xd234e)) {
-            skip_bits(&gb, 18);
+
+        if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
+            int shorten_by;
+
+            if (get_bits(&gb, 16) != 0xD234)
+                return AVERROR_INVALIDDATA;
+
+            shorten_by = get_bits(&gb, 16);
+            if      (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by  & 0x2000)
+                s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
+            else if (m->avctx->codec_id == AV_CODEC_ID_MLP    && shorten_by != 0xD234)
+                return AVERROR_INVALIDDATA;
+
             if (substr == m->max_decoded_substream)
                 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
-
-            if (get_bits1(&gb)) {
-                int shorten_by = get_bits(&gb, 13);
-                shorten_by = FFMIN(shorten_by, s->blockpos);
-                s->blockpos -= shorten_by;
-            } else
-                skip_bits(&gb, 13);
         }
-        if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 16 &&
-            substream_parity_present[substr]) {
+
+        if (substream_parity_present[substr]) {
             uint8_t parity, checksum;
 
-            parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
-            if ((parity ^ get_bits(&gb, 8)) != 0xa9)
-                av_log(m->avctx, AV_LOG_ERROR,
-                       "Substream %d parity check failed.\n", substr);
+            if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
+                goto substream_length_mismatch;
 
-            checksum = ff_mlp_checksum8(buf, substream_data_len[substr] - 2);
-            if (checksum != get_bits(&gb, 8))
-                av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n",
-                       substr);
-        }
-        if (substream_data_len[substr] * 8 != get_bits_count(&gb)) {
-            av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n",
-                   substr);
-            return -1;
+            parity   = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
+            checksum = ff_mlp_checksum8       (buf, substream_data_len[substr] - 2);
+
+            if ((get_bits(&gb, 8) ^ parity) != 0xa9    )
+                av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
+            if ( get_bits(&gb, 8)           != checksum)
+                av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n"    , substr);
         }
 
+        if (substream_data_len[substr] * 8 != get_bits_count(&gb))
+            goto substream_length_mismatch;
+
 next_substr:
+        if (!s->restart_seen)
+            av_log(m->avctx, AV_LOG_ERROR,
+                   "No restart header present in substream %d.\n", substr);
+
         buf += substream_data_len[substr];
     }
 
     rematrix_channels(m, m->max_decoded_substream);
 
-    if (output_data(m, m->max_decoded_substream, data, data_size) < 0)
-        return -1;
+    if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
+        return ret;
 
     return length;
 
+substream_length_mismatch:
+    av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
+    return AVERROR_INVALIDDATA;
+
 error:
     m->params_valid = 0;
-    return -1;
+    return AVERROR_INVALIDDATA;
 }
 
-AVCodec mlp_decoder = {
-    "mlp",
-    CODEC_TYPE_AUDIO,
-    CODEC_ID_MLP,
-    sizeof(MLPDecodeContext),
-    mlp_decode_init,
-    NULL,
-    NULL,
-    read_access_unit,
-    .long_name = NULL_IF_CONFIG_SMALL("Meridian Lossless Packing"),
+AVCodec ff_mlp_decoder = {
+    .name           = "mlp",
+    .long_name      = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
+    .type           = AVMEDIA_TYPE_AUDIO,
+    .id             = AV_CODEC_ID_MLP,
+    .priv_data_size = sizeof(MLPDecodeContext),
+    .init           = mlp_decode_init,
+    .decode         = read_access_unit,
+    .capabilities   = CODEC_CAP_DR1,
 };
 
+#if CONFIG_TRUEHD_DECODER
+AVCodec ff_truehd_decoder = {
+    .name           = "truehd",
+    .long_name      = NULL_IF_CONFIG_SMALL("TrueHD"),
+    .type           = AVMEDIA_TYPE_AUDIO,
+    .id             = AV_CODEC_ID_TRUEHD,
+    .priv_data_size = sizeof(MLPDecodeContext),
+    .init           = mlp_decode_init,
+    .decode         = read_access_unit,
+    .capabilities   = CODEC_CAP_DR1,
+};
+#endif /* CONFIG_TRUEHD_DECODER */