lavc: Add per-thread surfaces in get_hw_frame_parameters()
[ffmpeg.git] / libavcodec / mlpdec.c
index 615abf3..b377fd7 100644 (file)
 
 #include <stdint.h>
 
-#include "avcodec.h"
 #include "libavutil/internal.h"
 #include "libavutil/intreadwrite.h"
 #include "libavutil/channel_layout.h"
-#include "get_bits.h"
-#include "internal.h"
 #include "libavutil/crc.h"
+
+#include "avcodec.h"
+#include "bitstream.h"
+#include "internal.h"
 #include "parser.h"
 #include "mlp_parser.h"
 #include "mlpdsp.h"
 #include "mlp.h"
+#include "config.h"
+#include "vlc.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
+#define VLC_STATIC_SIZE     512
+#endif
 
 typedef struct SubStream {
     /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
@@ -59,7 +68,7 @@ typedef struct SubStream {
     /// 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;
+    uint64_t    mask;
     /// The matrix encoding mode for this substream
     enum AVMatrixEncoding matrix_encoding;
 
@@ -125,6 +134,9 @@ typedef struct MLPDecodeContext {
     /// Current access unit being read has a major sync.
     int         is_major_sync_unit;
 
+    /// Size of the major sync unit, in bytes
+    int         major_sync_header_size;
+
     /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
     uint8_t     params_valid;
 
@@ -167,6 +179,11 @@ static const uint64_t thd_channel_order[] = {
     AV_CH_LOW_FREQUENCY_2,                                   // LFE2
 };
 
+static int mlp_channel_layout_subset(uint64_t channel_layout, uint64_t mask)
+{
+    return channel_layout && ((channel_layout & mask) == channel_layout);
+}
+
 static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
                                                    int index)
 {
@@ -190,13 +207,13 @@ static av_cold void init_static(void)
     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, 512);
+                    &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, 512);
+                    &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, 512);
+                    &ff_mlp_huffman_tables[2][0][0], 2, 1, VLC_STATIC_SIZE);
     }
 
     ff_mlp_init_crc();
@@ -223,7 +240,7 @@ static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
  *  and plain LSBs. */
 
-static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
+static inline int read_huff_channels(MLPDecodeContext *m, BitstreamContext *bc,
                                      unsigned int substr, unsigned int pos)
 {
     SubStream *s = &m->substream[substr];
@@ -231,7 +248,7 @@ static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
 
     for (mat = 0; mat < s->num_primitive_matrices; mat++)
         if (s->lsb_bypass[mat])
-            m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
+            m->bypassed_lsbs[pos + s->blockpos][mat] = bitstream_read_bit(bc);
 
     for (channel = s->min_channel; channel <= s->max_channel; channel++) {
         ChannelParams *cp = &s->channel_params[channel];
@@ -241,14 +258,15 @@ static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
         int result = 0;
 
         if (codebook > 0)
-            result = get_vlc2(gbp, huff_vlc[codebook-1].table,
-                            VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
+            result = bitstream_read_vlc(bc, huff_vlc[codebook-1].table,
+                                        VLC_BITS,
+                                        (9 + VLC_BITS - 1) / VLC_BITS);
 
         if (result < 0)
             return AVERROR_INVALIDDATA;
 
         if (lsb_bits > 0)
-            result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
+            result = (result << lsb_bits) + bitstream_read(bc, lsb_bits);
 
         result  += cp->sign_huff_offset;
         result <<= quant_step_size;
@@ -278,12 +296,12 @@ static av_cold int mlp_decode_init(AVCodecContext *avctx)
  *  information is not actually necessary for decoding, only for playback.
  */
 
-static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
+static int read_major_sync(MLPDecodeContext *m, BitstreamContext *bc)
 {
     MLPHeaderInfo mh;
     int substr, ret;
 
-    if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
+    if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, bc)) != 0)
         return ret;
 
     if (mh.group1_bits == 0) {
@@ -339,11 +357,15 @@ static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
         return AVERROR_PATCHWELCOME;
     }
 
+    m->major_sync_header_size = mh.header_size;
+
     m->access_unit_size      = mh.access_unit_size;
     m->access_unit_size_pow2 = mh.access_unit_size_pow2;
 
     m->num_substreams        = mh.num_substreams;
-    m->max_decoded_substream = m->num_substreams - 1;
+
+    /* limit to decoding 3 substreams, as the 4th is used by Dolby Atmos for non-audio data */
+    m->max_decoded_substream = FFMIN(m->num_substreams - 1, 2);
 
     m->avctx->sample_rate    = mh.group1_samplerate;
     m->avctx->frame_size     = mh.access_unit_size;
@@ -353,6 +375,10 @@ static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
         m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
     else
         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++)
@@ -363,17 +389,17 @@ static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
      * 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;
+            m->substream[0].mask = AV_CH_LAYOUT_STEREO;
+        m->substream[substr].mask = mh.channel_layout_mlp;
     } else {
         if ((substr = (mh.num_substreams > 1)))
-            m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
+            m->substream[0].mask = 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;
+                m->substream[2].mask = 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;
+                m->substream[2].mask = mh.channel_layout_thd_stream1;
+        m->substream[substr].mask = mh.channel_layout_thd_stream1;
     }
 
     /* Parse the TrueHD decoder channel modifiers and set each substream's
@@ -423,7 +449,7 @@ static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
  *  required to decode the audio that do not change very often. Generally
  *  (always) present only in blocks following a major sync. */
 
-static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
+static int read_restart_header(MLPDecodeContext *m, BitstreamContext *bc,
                                const uint8_t *buf, unsigned int substr)
 {
     SubStream *s = &m->substream[substr];
@@ -431,13 +457,13 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
     int sync_word, tmp;
     uint8_t checksum;
     uint8_t lossless_check;
-    int start_count = get_bits_count(gbp);
+    int start_count = bitstream_tell(bc);
     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);
+    sync_word = bitstream_read(bc, 13);
 
     if (sync_word != 0x31ea >> 1) {
         av_log(m->avctx, AV_LOG_ERROR,
@@ -445,18 +471,18 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
         return AVERROR_INVALIDDATA;
     }
 
-    s->noise_type = get_bits1(gbp);
+    s->noise_type = bitstream_read_bit(bc);
 
     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 */
+    bitstream_skip(bc, 16); /* Output timestamp */
 
-    min_channel        = get_bits(gbp, 4);
-    max_channel        = get_bits(gbp, 4);
-    max_matrix_channel = get_bits(gbp, 4);
+    min_channel        = bitstream_read(bc, 4);
+    max_channel        = bitstream_read(bc, 4);
+    max_matrix_channel = bitstream_read(bc, 4);
 
     if (max_matrix_channel > std_max_matrix_channel) {
         av_log(m->avctx, AV_LOG_ERROR,
@@ -491,35 +517,22 @@ static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
     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 &&
+    if (mlp_channel_layout_subset(m->avctx->request_channel_layout, s->mask) &&
         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);
+               s->max_channel + 1, s->mask, substr);
         m->max_decoded_substream = substr;
     }
 
-    s->noise_shift   = get_bits(gbp,  4);
-    s->noisegen_seed = get_bits(gbp, 23);
+    s->noise_shift   = bitstream_read(bc,  4);
+    s->noisegen_seed = bitstream_read(bc, 23);
 
-    skip_bits(gbp, 19);
+    bitstream_skip(bc, 19);
 
-    s->data_check_present = get_bits1(gbp);
-    lossless_check = get_bits(gbp, 8);
+    s->data_check_present = bitstream_read_bit(bc);
+    lossless_check = bitstream_read(bc, 8);
     if (substr == m->max_decoded_substream
         && s->lossless_check_data != 0xffffffff) {
         tmp = xor_32_to_8(s->lossless_check_data);
@@ -529,19 +542,19 @@ FF_ENABLE_DEPRECATION_WARNINGS
                    lossless_check, tmp);
     }
 
-    skip_bits(gbp, 16);
+    bitstream_skip(bc, 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);
+        int ch_assign = bitstream_read(bc, 6);
         if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
-            uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
+            uint64_t channel = thd_channel_layout_extract_channel(s->mask,
                                                                   ch_assign);
-            ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
+            ch_assign = av_get_channel_layout_channel_index(s->mask,
                                                             channel);
         }
-        if (ch_assign > s->max_matrix_channel) {
+        if (ch_assign < 0 || ch_assign > s->max_matrix_channel) {
             avpriv_request_sample(m->avctx,
                                   "Assignment of matrix channel %d to invalid output channel %d",
                                   ch, ch_assign);
@@ -550,9 +563,9 @@ FF_ENABLE_DEPRECATION_WARNINGS
         s->ch_assign[ch_assign] = ch;
     }
 
-    checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
+    checksum = ff_mlp_restart_checksum(buf, bitstream_tell(bc) - start_count);
 
-    if (checksum != get_bits(gbp, 8))
+    if (checksum != bitstream_read(bc, 8))
         av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
 
     /* Set default decoding parameters. */
@@ -573,14 +586,18 @@ FF_ENABLE_DEPRECATION_WARNINGS
 
         /* Default audio coding is 24-bit raw PCM. */
         cp->huff_offset      = 0;
-        cp->sign_huff_offset = (-1) << 23;
+        cp->sign_huff_offset = -(1 << 23);
         cp->codebook         = 0;
         cp->huff_lsbs        = 24;
     }
 
     if (substr == m->max_decoded_substream) {
         m->avctx->channels       = s->max_matrix_channel + 1;
-        m->avctx->channel_layout = s->ch_layout;
+        m->avctx->channel_layout = s->mask;
+        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;
@@ -588,7 +605,7 @@ FF_ENABLE_DEPRECATION_WARNINGS
 
 /** Read parameters for one of the prediction filters. */
 
-static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
+static int read_filter_params(MLPDecodeContext *m, BitstreamContext *bc,
                               unsigned int substr, unsigned int channel,
                               unsigned int filter)
 {
@@ -606,7 +623,7 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
         return AVERROR_INVALIDDATA;
     }
 
-    order = get_bits(gbp, 4);
+    order = bitstream_read(bc, 4);
     if (order > max_order) {
         av_log(m->avctx, AV_LOG_ERROR,
                "%cIR filter order %d is greater than maximum %d.\n",
@@ -619,10 +636,10 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
         int32_t *fcoeff = s->channel_params[channel].coeff[filter];
         int coeff_bits, coeff_shift;
 
-        fp->shift = get_bits(gbp, 4);
+        fp->shift = bitstream_read(bc, 4);
 
-        coeff_bits  = get_bits(gbp, 5);
-        coeff_shift = get_bits(gbp, 3);
+        coeff_bits  = bitstream_read(bc, 5);
+        coeff_shift = bitstream_read(bc, 3);
         if (coeff_bits < 1 || coeff_bits > 16) {
             av_log(m->avctx, AV_LOG_ERROR,
                    "%cIR filter coeff_bits must be between 1 and 16.\n",
@@ -637,9 +654,9 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
         }
 
         for (i = 0; i < order; i++)
-            fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
+            fcoeff[i] = bitstream_read_signed(bc, coeff_bits) << coeff_shift;
 
-        if (get_bits1(gbp)) {
+        if (bitstream_read_bit(bc)) {
             int state_bits, state_shift;
 
             if (filter == FIR) {
@@ -648,13 +665,13 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
                 return AVERROR_INVALIDDATA;
             }
 
-            state_bits  = get_bits(gbp, 4);
-            state_shift = get_bits(gbp, 4);
+            state_bits  = bitstream_read(bc, 4);
+            state_shift = bitstream_read(bc, 4);
 
             /* TODO: Check validity of state data. */
 
             for (i = 0; i < order; i++)
-                fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
+                fp->state[i] = bitstream_read_signed(bc, state_bits) << state_shift;
         }
     }
 
@@ -663,7 +680,8 @@ static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
 
 /** Read parameters for primitive matrices. */
 
-static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
+static int read_matrix_params(MLPDecodeContext *m, unsigned int substr,
+                              BitstreamContext *bc)
 {
     SubStream *s = &m->substream[substr];
     unsigned int mat, ch;
@@ -676,7 +694,7 @@ static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitCo
         return AVERROR_INVALIDDATA;
     }
 
-    s->num_primitive_matrices = get_bits(gbp, 4);
+    s->num_primitive_matrices = bitstream_read(bc, 4);
 
     if (s->num_primitive_matrices > max_primitive_matrices) {
         av_log(m->avctx, AV_LOG_ERROR,
@@ -687,9 +705,9 @@ static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitCo
 
     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);
+        s->matrix_out_ch[mat] = bitstream_read(bc, 4);
+        frac_bits             = bitstream_read(bc, 4);
+        s->lsb_bypass[mat]    = bitstream_read_bit(bc);
 
         if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
             av_log(m->avctx, AV_LOG_ERROR,
@@ -709,14 +727,14 @@ static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitCo
 
         for (ch = 0; ch <= max_chan; ch++) {
             int coeff_val = 0;
-            if (get_bits1(gbp))
-                coeff_val = get_sbits(gbp, frac_bits + 2);
+            if (bitstream_read_bit(bc))
+                coeff_val = bitstream_read_signed(bc, 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);
+            s->matrix_noise_shift[mat] = bitstream_read(bc, 4);
         else
             s->matrix_noise_shift[mat] = 0;
     }
@@ -727,7 +745,7 @@ static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitCo
 /** Read channel parameters. */
 
 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
-                               GetBitContext *gbp, unsigned int ch)
+                               BitstreamContext *bc, unsigned int ch)
 {
     SubStream *s = &m->substream[substr];
     ChannelParams *cp = &s->channel_params[ch];
@@ -736,13 +754,13 @@ static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
     int ret;
 
     if (s->param_presence_flags & PARAM_FIR)
-        if (get_bits1(gbp))
-            if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
+        if (bitstream_read_bit(bc))
+            if ((ret = read_filter_params(m, bc, 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)
+        if (bitstream_read_bit(bc))
+            if ((ret = read_filter_params(m, bc, substr, ch, IIR)) < 0)
                 return ret;
 
     if (fir->order + iir->order > 8) {
@@ -765,11 +783,11 @@ static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
         fir->shift = iir->shift;
 
     if (s->param_presence_flags & PARAM_HUFFOFFSET)
-        if (get_bits1(gbp))
-            cp->huff_offset = get_sbits(gbp, 15);
+        if (bitstream_read_bit(bc))
+            cp->huff_offset = bitstream_read_signed(bc, 15);
 
-    cp->codebook  = get_bits(gbp, 2);
-    cp->huff_lsbs = get_bits(gbp, 5);
+    cp->codebook  = bitstream_read(bc, 2);
+    cp->huff_lsbs = bitstream_read(bc, 5);
 
     if (cp->huff_lsbs > 24) {
         av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
@@ -784,7 +802,7 @@ static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
 /** Read decoding parameters that change more often than those in the restart
  *  header. */
 
-static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
+static int read_decoding_params(MLPDecodeContext *m, BitstreamContext *bc,
                                 unsigned int substr)
 {
     SubStream *s = &m->substream[substr];
@@ -792,12 +810,12 @@ static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
     int ret;
 
     if (s->param_presence_flags & PARAM_PRESENCE)
-        if (get_bits1(gbp))
-            s->param_presence_flags = get_bits(gbp, 8);
+        if (bitstream_read_bit(bc))
+            s->param_presence_flags = bitstream_read(bc, 8);
 
     if (s->param_presence_flags & PARAM_BLOCKSIZE)
-        if (get_bits1(gbp)) {
-            s->blocksize = get_bits(gbp, 9);
+        if (bitstream_read_bit(bc)) {
+            s->blocksize = bitstream_read(bc, 9);
             if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
                 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
                 s->blocksize = 0;
@@ -806,28 +824,34 @@ static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
         }
 
     if (s->param_presence_flags & PARAM_MATRIX)
-        if (get_bits1(gbp))
-            if ((ret = read_matrix_params(m, substr, gbp)) < 0)
+        if (bitstream_read_bit(bc))
+            if ((ret = read_matrix_params(m, substr, bc)) < 0)
                 return ret;
 
     if (s->param_presence_flags & PARAM_OUTSHIFT)
-        if (get_bits1(gbp))
+        if (bitstream_read_bit(bc)) {
             for (ch = 0; ch <= s->max_matrix_channel; ch++)
-                s->output_shift[ch] = get_sbits(gbp, 4);
+                s->output_shift[ch] = bitstream_read_signed(bc, 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))
+        if (bitstream_read_bit(bc))
             for (ch = 0; ch <= s->max_channel; ch++) {
                 ChannelParams *cp = &s->channel_params[ch];
 
-                s->quant_step_size[ch] = get_bits(gbp, 4);
+                s->quant_step_size[ch] = bitstream_read(bc, 4);
 
                 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
             }
 
     for (ch = s->min_channel; ch <= s->max_channel; ch++)
-        if (get_bits1(gbp))
-            if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
+        if (bitstream_read_bit(bc))
+            if ((ret = read_channel_params(m, substr, bc, ch)) < 0)
                 return ret;
 
     return 0;
@@ -865,7 +889,7 @@ static void filter_channel(MLPDecodeContext *m, unsigned int substr,
 
 /** Read a block of PCM residual data (or actual if no filtering active). */
 
-static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
+static int read_block_data(MLPDecodeContext *m, BitstreamContext *bc,
                            unsigned int substr)
 {
     SubStream *s = &m->substream[substr];
@@ -873,8 +897,8 @@ static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
     int ret;
 
     if (s->data_check_present) {
-        expected_stream_pos  = get_bits_count(gbp);
-        expected_stream_pos += get_bits(gbp, 16);
+        expected_stream_pos  = bitstream_tell(bc);
+        expected_stream_pos += bitstream_read(bc, 16);
         avpriv_request_sample(m->avctx,
                               "Substreams with VLC block size check info");
     }
@@ -888,7 +912,7 @@ static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
            s->blocksize * sizeof(m->bypassed_lsbs[0]));
 
     for (i = 0; i < s->blocksize; i++)
-        if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
+        if ((ret = read_huff_channels(m, bc, substr, i)) < 0)
             return ret;
 
     for (ch = s->min_channel; ch <= s->max_channel; ch++)
@@ -897,9 +921,9 @@ static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
     s->blockpos += s->blocksize;
 
     if (s->data_check_present) {
-        if (get_bits_count(gbp) != expected_stream_pos)
+        if (bitstream_tell(bc) != expected_stream_pos)
             av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
-        skip_bits(gbp, 8);
+        bitstream_skip(bc, 8);
     }
 
     return 0;
@@ -978,7 +1002,7 @@ 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;
@@ -990,31 +1014,18 @@ 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]);
-        int32_t *coeffs = s->matrix_coeff[mat];
-        int index  = s->num_primitive_matrices - mat;
-        int index2 = 2 * index + 1;
-
-        /* TODO: DSPContext? */
-
-        for (i = 0; i < s->blockpos; i++) {
-            int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
-            int32_t *samples = m->sample_buffer[i];
-            int64_t accum = 0;
-
-            for (src_ch = 0; src_ch <= maxchan; src_ch++)
-                accum += (int64_t) samples[src_ch] * coeffs[src_ch];
-
-            if (matrix_noise_shift) {
-                index &= m->access_unit_size_pow2 - 1;
-                accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
-                index += index2;
-            }
-
-            samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
-        }
+        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]));
     }
 }
 
@@ -1025,9 +1036,6 @@ static int output_data(MLPDecodeContext *m, unsigned int substr,
 {
     AVCodecContext *avctx = m->avctx;
     SubStream *s = &m->substream[substr];
-    unsigned int i, out_ch = 0;
-    int32_t *data_32;
-    int16_t *data_16;
     int ret;
     int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
 
@@ -1047,19 +1055,18 @@ static int output_data(MLPDecodeContext *m, unsigned int substr,
         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
         return ret;
     }
-    data_32 = (int32_t *)frame->data[0];
-    data_16 = (int16_t *)frame->data[0];
-
-    for (i = 0; i < s->blockpos; i++) {
-        for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
-            int mat_ch = s->ch_assign[out_ch];
-            int32_t sample = m->sample_buffer[i][mat_ch]
-                          << s->output_shift[mat_ch];
-            s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
-            if (is32) *data_32++ = sample << 8;
-            else      *data_16++ = sample >> 8;
-        }
-    }
+    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);
+
+    /* Update matrix encoding side data */
+    if ((ret = ff_side_data_update_matrix_encoding(frame, s->matrix_encoding)) < 0)
+        return ret;
 
     *got_frame_ptr = 1;
 
@@ -1076,7 +1083,7 @@ static int read_access_unit(AVCodecContext *avctx, void* data,
     const uint8_t *buf = avpkt->data;
     int buf_size = avpkt->size;
     MLPDecodeContext *m = avctx->priv_data;
-    GetBitContext gb;
+    BitstreamContext bc;
     unsigned int length, substr;
     unsigned int substream_start;
     unsigned int header_size = 4;
@@ -1094,14 +1101,14 @@ static int read_access_unit(AVCodecContext *avctx, void* data,
     if (length < 4 || length > buf_size)
         return AVERROR_INVALIDDATA;
 
-    init_get_bits(&gb, (buf + 4), (length - 4) * 8);
+    bitstream_init8(&bc, buf + 4, length - 4);
 
     m->is_major_sync_unit = 0;
-    if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
-        if (read_major_sync(m, &gb) < 0)
+    if (bitstream_peek(&bc, 31) == (0xf8726fba >> 1)) {
+        if (read_major_sync(m, &bc) < 0)
             goto error;
         m->is_major_sync_unit = 1;
-        header_size += 28;
+        header_size += m->major_sync_header_size;
     }
 
     if (!m->params_valid) {
@@ -1116,12 +1123,12 @@ static int read_access_unit(AVCodecContext *avctx, void* data,
     for (substr = 0; substr < m->num_substreams; substr++) {
         int extraword_present, checkdata_present, end, nonrestart_substr;
 
-        extraword_present = get_bits1(&gb);
-        nonrestart_substr = get_bits1(&gb);
-        checkdata_present = get_bits1(&gb);
-        skip_bits1(&gb);
+        extraword_present = bitstream_read_bit(&bc);
+        nonrestart_substr = bitstream_read_bit(&bc);
+        checkdata_present = bitstream_read_bit(&bc);
+        bitstream_skip(&bc, 1);
 
-        end = get_bits(&gb, 12) * 2;
+        end = bitstream_read(&bc, 12) * 2;
 
         substr_header_size += 2;
 
@@ -1130,7 +1137,7 @@ static int read_access_unit(AVCodecContext *avctx, void* data,
                 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
                 goto error;
             }
-            skip_bits(&gb, 16);
+            bitstream_skip(&bc, 16);
             substr_header_size += 2;
         }
 
@@ -1174,47 +1181,47 @@ static int read_access_unit(AVCodecContext *avctx, void* data,
 
     for (substr = 0; substr <= m->max_decoded_substream; substr++) {
         SubStream *s = &m->substream[substr];
-        init_get_bits(&gb, buf, substream_data_len[substr] * 8);
+        bitstream_init8(&bc, buf, substream_data_len[substr]);
 
         m->matrix_changed = 0;
         memset(m->filter_changed, 0, sizeof(m->filter_changed));
 
         s->blockpos = 0;
         do {
-            if (get_bits1(&gb)) {
-                if (get_bits1(&gb)) {
+            if (bitstream_read_bit(&bc)) {
+                if (bitstream_read_bit(&bc)) {
                     /* A restart header should be present. */
-                    if (read_restart_header(m, &gb, buf, substr) < 0)
+                    if (read_restart_header(m, &bc, buf, substr) < 0)
                         goto next_substr;
                     s->restart_seen = 1;
                 }
 
                 if (!s->restart_seen)
                     goto next_substr;
-                if (read_decoding_params(m, &gb, substr) < 0)
+                if (read_decoding_params(m, &bc, substr) < 0)
                     goto next_substr;
             }
 
             if (!s->restart_seen)
                 goto next_substr;
 
-            if ((ret = read_block_data(m, &gb, substr)) < 0)
+            if ((ret = read_block_data(m, &bc, substr)) < 0)
                 return ret;
 
-            if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
+            if (bitstream_tell(&bc) >= substream_data_len[substr] * 8)
                 goto substream_length_mismatch;
 
-        } while (!get_bits1(&gb));
+        } while (!bitstream_read_bit(&bc));
 
-        skip_bits(&gb, (-get_bits_count(&gb)) & 15);
+        bitstream_skip(&bc, (-bitstream_tell(&bc)) & 15);
 
-        if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
+        if (substream_data_len[substr] * 8 - bitstream_tell(&bc) >= 32) {
             int shorten_by;
 
-            if (get_bits(&gb, 16) != 0xD234)
+            if (bitstream_read(&bc, 16) != 0xD234)
                 return AVERROR_INVALIDDATA;
 
-            shorten_by = get_bits(&gb, 16);
+            shorten_by = bitstream_read(&bc, 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)
@@ -1227,19 +1234,19 @@ static int read_access_unit(AVCodecContext *avctx, void* data,
         if (substream_parity_present[substr]) {
             uint8_t parity, checksum;
 
-            if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
+            if (substream_data_len[substr] * 8 - bitstream_tell(&bc) != 16)
                 goto substream_length_mismatch;
 
             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    )
+            if ((bitstream_read(&bc, 8) ^ parity) != 0xa9)
                 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
-            if ( get_bits(&gb, 8)           != checksum)
+            if (bitstream_read(&bc, 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))
+        if (substream_data_len[substr] * 8 != bitstream_tell(&bc))
             goto substream_length_mismatch;
 
 next_substr:
@@ -1274,7 +1281,7 @@ AVCodec ff_mlp_decoder = {
     .priv_data_size = sizeof(MLPDecodeContext),
     .init           = mlp_decode_init,
     .decode         = read_access_unit,
-    .capabilities   = CODEC_CAP_DR1,
+    .capabilities   = AV_CODEC_CAP_DR1,
 };
 
 #if CONFIG_TRUEHD_DECODER
@@ -1286,6 +1293,6 @@ AVCodec ff_truehd_decoder = {
     .priv_data_size = sizeof(MLPDecodeContext),
     .init           = mlp_decode_init,
     .decode         = read_access_unit,
-    .capabilities   = CODEC_CAP_DR1,
+    .capabilities   = AV_CODEC_CAP_DR1,
 };
 #endif /* CONFIG_TRUEHD_DECODER */