avcodec/idctdsp: Clear idct/idct_add for studio profile
[ffmpeg.git] / libavcodec / dca_xll.c
index 9e1085c..d265cab 100644 (file)
@@ -1,8 +1,5 @@
 /*
- * DCA XLL extension
- *
- * Copyright (C) 2012 Paul B Mahol
- * Copyright (C) 2014 Niels Möller
+ * Copyright (C) 2016 foo86
  *
  * This file is part of FFmpeg.
  *
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-#include "libavutil/attributes.h"
-#include "libavutil/common.h"
-#include "libavutil/internal.h"
-
-#include "avcodec.h"
-#include "dca.h"
+#include "dcadec.h"
 #include "dcadata.h"
-#include "get_bits.h"
+#include "dcamath.h"
+#include "dca_syncwords.h"
 #include "unary.h"
 
-/* Sign as bit 0 */
-static inline int get_bits_sm(GetBitContext *s, unsigned n)
+static int get_linear(GetBitContext *gb, int n)
 {
-    int x = get_bits(s, n);
-    if (x & 1)
-        return -(x >> 1) - 1;
-    else
-        return x >> 1;
-}
-
-/* Return -1 on error. */
-static int32_t get_dmix_coeff(DCAContext *s, int inverse)
-{
-    unsigned code = get_bits(&s->gb, 9);
-    int32_t sign = (int32_t) (code >> 8) - 1;
-    unsigned idx = code & 0xff;
-    int inv_offset = FF_DCA_DMIXTABLE_SIZE -FF_DCA_INV_DMIXTABLE_SIZE;
-    if (idx >= FF_DCA_DMIXTABLE_SIZE) {
-        av_log(s->avctx, AV_LOG_ERROR,
-               "XLL: Invalid channel set downmix code %x\n", code);
-        return -1;
-    } else if (!inverse) {
-        return (ff_dca_dmixtable[idx] ^ sign) - sign;
-    } else if (idx < inv_offset) {
-        av_log(s->avctx, AV_LOG_ERROR,
-               "XLL: Invalid channel set inverse downmix code %x\n", code);
-        return -1;
-    } else {
-        return (ff_dca_inv_dmixtable[idx - inv_offset] ^ sign) - sign;
-    }
+    unsigned int v = get_bits_long(gb, n);
+    return (v >> 1) ^ -(v & 1);
+}
+
+static int get_rice_un(GetBitContext *gb, int k)
+{
+    unsigned int v = get_unary(gb, 1, get_bits_left(gb));
+    return (v << k) | get_bits_long(gb, k);
 }
 
-static int32_t dca_get_dmix_coeff(DCAContext *s)
+static int get_rice(GetBitContext *gb, int k)
 {
-    return get_dmix_coeff(s, 0);
+    unsigned int v = get_rice_un(gb, k);
+    return (v >> 1) ^ -(v & 1);
 }
 
-static int32_t dca_get_inv_dmix_coeff(DCAContext *s)
+static void get_array(GetBitContext *gb, int32_t *array, int size, int n)
 {
-    return get_dmix_coeff(s, 1);
+    int i;
+
+    for (i = 0; i < size; i++)
+        array[i] = get_bits(gb, n);
 }
 
-/* parse XLL header */
-int ff_dca_xll_decode_header(DCAContext *s)
+static void get_linear_array(GetBitContext *gb, int32_t *array, int size, int n)
 {
-    int hdr_pos, hdr_size;
-    av_unused int version, frame_size;
-    int i, chset_index;
+    int i;
 
-    /* get bit position of sync header */
-    hdr_pos    = get_bits_count(&s->gb) - 32;
+    if (n == 0)
+        memset(array, 0, sizeof(*array) * size);
+    else for (i = 0; i < size; i++)
+        array[i] = get_linear(gb, n);
+}
 
-    version    = get_bits(&s->gb, 4) + 1;
-    hdr_size   = get_bits(&s->gb, 8) + 1;
+static void get_rice_array(GetBitContext *gb, int32_t *array, int size, int k)
+{
+    int i;
 
-    frame_size = get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1) + 1;
+    for (i = 0; i < size; i++)
+        array[i] = get_rice(gb, k);
+}
 
-    s->xll_channels          =
-    s->xll_residual_channels = 0;
-    s->xll_nch_sets          = get_bits(&s->gb, 4) + 1;
-    s->xll_segments          = 1 << get_bits(&s->gb, 4);
-    s->xll_log_smpl_in_seg   = get_bits(&s->gb, 4);
-    s->xll_smpl_in_seg       = 1 << s->xll_log_smpl_in_seg;
-    s->xll_bits4seg_size     = get_bits(&s->gb, 5) + 1;
-    s->xll_banddata_crc      = get_bits(&s->gb, 2);
-    s->xll_scalable_lsb      = get_bits1(&s->gb);
-    s->xll_bits4ch_mask      = get_bits(&s->gb, 5) + 1;
+static int parse_dmix_coeffs(DCAXllDecoder *s, DCAXllChSet *c)
+{
+    // Size of downmix coefficient matrix
+    int m = c->primary_chset ? ff_dca_dmix_primary_nch[c->dmix_type] : c->hier_ofs;
+    int i, j, *coeff_ptr = c->dmix_coeff;
+
+    for (i = 0; i < m; i++) {
+        int code, sign, coeff, scale, scale_inv = 0;
+        unsigned int index;
+
+        // Downmix scale (only for non-primary channel sets)
+        if (!c->primary_chset) {
+            code = get_bits(&s->gb, 9);
+            sign = (code >> 8) - 1;
+            index = (code & 0xff) - FF_DCA_DMIXTABLE_OFFSET;
+            if (index >= FF_DCA_INV_DMIXTABLE_SIZE) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL downmix scale index\n");
+                return AVERROR_INVALIDDATA;
+            }
+            scale = ff_dca_dmixtable[index + FF_DCA_DMIXTABLE_OFFSET];
+            scale_inv = ff_dca_inv_dmixtable[index];
+            c->dmix_scale[i] = (scale ^ sign) - sign;
+            c->dmix_scale_inv[i] = (scale_inv ^ sign) - sign;
+        }
 
-    if (s->xll_scalable_lsb) {
-        s->xll_fixed_lsb_width = get_bits(&s->gb, 4);
-        if (s->xll_fixed_lsb_width)
-            av_log(s->avctx, AV_LOG_WARNING,
-                   "XLL: fixed lsb width = %d, non-zero not supported.\n",
-                   s->xll_fixed_lsb_width);
+        // Downmix coefficients
+        for (j = 0; j < c->nchannels; j++) {
+            code = get_bits(&s->gb, 9);
+            sign = (code >> 8) - 1;
+            index = code & 0xff;
+            if (index >= FF_DCA_DMIXTABLE_SIZE) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL downmix coefficient index\n");
+                return AVERROR_INVALIDDATA;
+            }
+            coeff = ff_dca_dmixtable[index];
+            if (!c->primary_chset)
+                // Multiply by |InvDmixScale| to get |UndoDmixScale|
+                coeff = mul16(scale_inv, coeff);
+            *coeff_ptr++ = (coeff ^ sign) - sign;
+        }
     }
-    /* skip to the end of the common header */
-    i = get_bits_count(&s->gb);
-    if (hdr_pos + hdr_size * 8 > i)
-        skip_bits_long(&s->gb, hdr_pos + hdr_size * 8 - i);
 
-    for (chset_index = 0; chset_index < s->xll_nch_sets; chset_index++) {
-        XllChSetSubHeader *chset = &s->xll_chsets[chset_index];
-        hdr_pos  = get_bits_count(&s->gb);
-        hdr_size = get_bits(&s->gb, 10) + 1;
+    return 0;
+}
+
+static int chs_parse_header(DCAXllDecoder *s, DCAXllChSet *c, DCAExssAsset *asset)
+{
+    int i, j, k, ret, band, header_size, header_pos = get_bits_count(&s->gb);
+    DCAXllChSet *p = &s->chset[0];
+    DCAXllBand *b;
 
-        chset->channels           = get_bits(&s->gb, 4) + 1;
-        chset->residual_encode    = get_bits(&s->gb, chset->channels);
-        chset->bit_resolution     = get_bits(&s->gb, 5) + 1;
-        chset->bit_width          = get_bits(&s->gb, 5) + 1;
-        chset->sampling_frequency = ff_dca_sampling_freqs[get_bits(&s->gb, 4)];
-        chset->samp_freq_interp   = get_bits(&s->gb, 2);
-        chset->replacement_set    = get_bits(&s->gb, 2);
-        if (chset->replacement_set)
-            chset->active_replace_set = get_bits(&s->gb, 1);
+    // Size of channel set sub-header
+    header_size = get_bits(&s->gb, 10) + 1;
 
-        if (s->one2one_map_chtospkr) {
-            chset->primary_ch_set              = get_bits(&s->gb, 1);
-            chset->downmix_coeff_code_embedded = get_bits(&s->gb, 1);
-            if (chset->downmix_coeff_code_embedded) {
-                chset->downmix_embedded = get_bits(&s->gb, 1);
-                if (chset->primary_ch_set) {
-                    chset->downmix_type = get_bits(&s->gb, 3);
-                    if (chset->downmix_type > 6) {
-                        av_log(s->avctx, AV_LOG_ERROR,
-                               "XLL: Invalid channel set downmix type\n");
-                        return AVERROR_INVALIDDATA;
-                    }
-                }
+    // Check CRC
+    if (ff_dca_check_crc(s->avctx, &s->gb, header_pos, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL sub-header checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Number of channels in the channel set
+    c->nchannels = get_bits(&s->gb, 4) + 1;
+    if (c->nchannels > DCA_XLL_CHANNELS_MAX) {
+        avpriv_request_sample(s->avctx, "%d XLL channels", c->nchannels);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Residual type
+    c->residual_encode = get_bits(&s->gb, c->nchannels);
+
+    // PCM bit resolution
+    c->pcm_bit_res = get_bits(&s->gb, 5) + 1;
+
+    // Storage unit width
+    c->storage_bit_res = get_bits(&s->gb, 5) + 1;
+    if (c->storage_bit_res != 16 && c->storage_bit_res != 20 && c->storage_bit_res != 24) {
+        avpriv_request_sample(s->avctx, "%d-bit XLL storage resolution", c->storage_bit_res);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    if (c->pcm_bit_res > c->storage_bit_res) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid PCM bit resolution for XLL channel set (%d > %d)\n", c->pcm_bit_res, c->storage_bit_res);
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Original sampling frequency
+    c->freq = ff_dca_sampling_freqs[get_bits(&s->gb, 4)];
+    if (c->freq > 192000) {
+        avpriv_request_sample(s->avctx, "%d Hz XLL sampling frequency", c->freq);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Sampling frequency modifier
+    if (get_bits(&s->gb, 2)) {
+        avpriv_request_sample(s->avctx, "XLL sampling frequency modifier");
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Which replacement set this channel set is member of
+    if (get_bits(&s->gb, 2)) {
+        avpriv_request_sample(s->avctx, "XLL replacement set");
+        return AVERROR_PATCHWELCOME;
+    }
+
+    if (asset->one_to_one_map_ch_to_spkr) {
+        // Primary channel set flag
+        c->primary_chset = get_bits1(&s->gb);
+        if (c->primary_chset != (c == p)) {
+            av_log(s->avctx, AV_LOG_ERROR, "The first (and only) XLL channel set must be primary\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Downmix coefficients present in stream
+        c->dmix_coeffs_present = get_bits1(&s->gb);
+
+        // Downmix already performed by encoder
+        c->dmix_embedded = c->dmix_coeffs_present && get_bits1(&s->gb);
+
+        // Downmix type
+        if (c->dmix_coeffs_present && c->primary_chset) {
+            c->dmix_type = get_bits(&s->gb, 3);
+            if (c->dmix_type >= DCA_DMIX_TYPE_COUNT) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL primary channel set downmix type\n");
+                return AVERROR_INVALIDDATA;
             }
-            chset->hier_chset = get_bits(&s->gb, 1);
-
-            if (chset->downmix_coeff_code_embedded) {
-                /* nDownmixCoeffs is specified as N * M. For a primary
-                 * channel set, it appears that N = number of
-                 * channels, and M is the number of downmix channels.
-                 *
-                 * For a non-primary channel set, N is specified as
-                 * number of channels + 1, and M is derived from the
-                 * channel set hierarchy, and at least in simple cases
-                 * M is the number of channels in preceding channel
-                 * sets. */
-                if (chset->primary_ch_set) {
-                    static const char dmix_table[7] = { 1, 2, 2, 3, 3, 4, 4 };
-                    chset->downmix_ncoeffs = chset->channels * dmix_table[chset->downmix_type];
-                } else
-                    chset->downmix_ncoeffs = (chset->channels + 1) * s->xll_channels;
-
-                if (chset->downmix_ncoeffs > DCA_XLL_DMIX_NCOEFFS_MAX) {
-                    avpriv_request_sample(s->avctx,
-                                          "XLL: More than %d downmix coefficients",
-                                          DCA_XLL_DMIX_NCOEFFS_MAX);
-                    return AVERROR_PATCHWELCOME;
-                } else if (chset->primary_ch_set) {
-                    for (i = 0; i < chset->downmix_ncoeffs; i++)
-                        if ((chset->downmix_coeffs[i] = dca_get_dmix_coeff(s)) == -1)
-                            return AVERROR_INVALIDDATA;
-                } else {
-                    unsigned c, r;
-                    for (c = 0, i = 0; c < s->xll_channels; c++, i += chset->channels + 1) {
-                        if ((chset->downmix_coeffs[i] = dca_get_inv_dmix_coeff(s)) == -1)
-                            return AVERROR_INVALIDDATA;
-                        for (r = 1; r <= chset->channels; r++) {
-                            int32_t coeff = dca_get_dmix_coeff(s);
-                            if (coeff == -1)
-                                return AVERROR_INVALIDDATA;
-                            chset->downmix_coeffs[i + r] =
-                                (chset->downmix_coeffs[i] * (int64_t) coeff + (1 << 15)) >> 16;
-                        }
-                    }
+        }
+
+        // Whether the channel set is part of a hierarchy
+        c->hier_chset = get_bits1(&s->gb);
+        if (!c->hier_chset && s->nchsets != 1) {
+            avpriv_request_sample(s->avctx, "XLL channel set outside of hierarchy");
+            return AVERROR_PATCHWELCOME;
+        }
+
+        // Downmix coefficients
+        if (c->dmix_coeffs_present && (ret = parse_dmix_coeffs(s, c)) < 0)
+            return ret;
+
+        // Channel mask enabled
+        if (!get_bits1(&s->gb)) {
+            avpriv_request_sample(s->avctx, "Disabled XLL channel mask");
+            return AVERROR_PATCHWELCOME;
+        }
+
+        // Channel mask for set
+        c->ch_mask = get_bits_long(&s->gb, s->ch_mask_nbits);
+        if (av_popcount(c->ch_mask) != c->nchannels) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL channel mask\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Build the channel to speaker map
+        for (i = 0, j = 0; i < s->ch_mask_nbits; i++)
+            if (c->ch_mask & (1U << i))
+                c->ch_remap[j++] = i;
+    } else {
+        // Mapping coeffs present flag
+        if (c->nchannels != 2 || s->nchsets != 1 || get_bits1(&s->gb)) {
+            avpriv_request_sample(s->avctx, "Custom XLL channel to speaker mapping");
+            return AVERROR_PATCHWELCOME;
+        }
+
+        // Setup for LtRt decoding
+        c->primary_chset = 1;
+        c->dmix_coeffs_present = 0;
+        c->dmix_embedded = 0;
+        c->hier_chset = 0;
+        c->ch_mask = DCA_SPEAKER_LAYOUT_STEREO;
+        c->ch_remap[0] = DCA_SPEAKER_L;
+        c->ch_remap[1] = DCA_SPEAKER_R;
+    }
+
+    if (c->freq > 96000) {
+        // Extra frequency bands flag
+        if (get_bits1(&s->gb)) {
+            avpriv_request_sample(s->avctx, "Extra XLL frequency bands");
+            return AVERROR_PATCHWELCOME;
+        }
+        c->nfreqbands = 2;
+    } else {
+        c->nfreqbands = 1;
+    }
+
+    // Set the sampling frequency to that of the first frequency band.
+    // Frequency will be doubled again after bands assembly.
+    c->freq >>= c->nfreqbands - 1;
+
+    // Verify that all channel sets have the same audio characteristics
+    if (c != p && (c->nfreqbands != p->nfreqbands || c->freq != p->freq
+                   || c->pcm_bit_res != p->pcm_bit_res
+                   || c->storage_bit_res != p->storage_bit_res)) {
+        avpriv_request_sample(s->avctx, "Different XLL audio characteristics");
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Determine number of bits to read bit allocation coding parameter
+    if (c->storage_bit_res > 16)
+        c->nabits = 5;
+    else if (c->storage_bit_res > 8)
+        c->nabits = 4;
+    else
+        c->nabits = 3;
+
+    // Account for embedded downmix and decimator saturation
+    if ((s->nchsets > 1 || c->nfreqbands > 1) && c->nabits < 5)
+        c->nabits++;
+
+    for (band = 0, b = c->bands; band < c->nfreqbands; band++, b++) {
+        // Pairwise channel decorrelation
+        if ((b->decor_enabled = get_bits1(&s->gb)) && c->nchannels > 1) {
+            int ch_nbits = av_ceil_log2(c->nchannels);
+
+            // Original channel order
+            for (i = 0; i < c->nchannels; i++) {
+                b->orig_order[i] = get_bits(&s->gb, ch_nbits);
+                if (b->orig_order[i] >= c->nchannels) {
+                    av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL original channel order\n");
+                    return AVERROR_INVALIDDATA;
                 }
             }
-            chset->ch_mask_enabled = get_bits(&s->gb, 1);
-            if (chset->ch_mask_enabled)
-                chset->ch_mask = get_bits(&s->gb, s->xll_bits4ch_mask);
-            else
-                /* Skip speaker configuration bits */
-                skip_bits_long(&s->gb, 25 * chset->channels);
+
+            // Pairwise channel coefficients
+            for (i = 0; i < c->nchannels / 2; i++)
+                b->decor_coeff[i] = get_bits1(&s->gb) ? get_linear(&s->gb, 7) : 0;
         } else {
-            chset->primary_ch_set              = 1;
-            chset->downmix_coeff_code_embedded = 0;
-            /* Spec: NumChHierChSet = 0, NumDwnMixCodeCoeffs = 0, whatever that means. */
-            chset->mapping_coeffs_present = get_bits(&s->gb, 1);
-            if (chset->mapping_coeffs_present) {
-                avpriv_report_missing_feature(s->avctx, "XLL: mapping coefficients");
-                return AVERROR_PATCHWELCOME;
-            }
+            for (i = 0; i < c->nchannels; i++)
+                b->orig_order[i] = i;
+            for (i = 0; i < c->nchannels / 2; i++)
+                b->decor_coeff[i] = 0;
         }
-        if (chset->sampling_frequency > 96000)
-            chset->num_freq_bands = 2 * (1 + get_bits(&s->gb, 1));
-        else
-            chset->num_freq_bands = 1;
 
-        if (chset->num_freq_bands > 1) {
-            avpriv_report_missing_feature(s->avctx, "XLL: num_freq_bands > 1");
-            return AVERROR_PATCHWELCOME;
+        // Adaptive predictor order
+        b->highest_pred_order = 0;
+        for (i = 0; i < c->nchannels; i++) {
+            b->adapt_pred_order[i] = get_bits(&s->gb, 4);
+            if (b->adapt_pred_order[i] > b->highest_pred_order)
+                b->highest_pred_order = b->adapt_pred_order[i];
+        }
+        if (b->highest_pred_order > s->nsegsamples) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL adaptive predicition order\n");
+            return AVERROR_INVALIDDATA;
         }
 
-        if (get_bits(&s->gb, 1)) { /* pw_ch_decor_enabled */
-            int bits = av_ceil_log2(chset->channels);
-            for (i = 0; i < chset->channels; i++) {
-                unsigned j = get_bits(&s->gb, bits);
-                if (j >= chset->channels) {
-                    av_log(s->avctx, AV_LOG_ERROR,
-                           "Original channel order value %u too large, only %d channels.\n",
-                           j, chset->channels);
+        // Fixed predictor order
+        for (i = 0; i < c->nchannels; i++)
+            b->fixed_pred_order[i] = b->adapt_pred_order[i] ? 0 : get_bits(&s->gb, 2);
+
+        // Adaptive predictor quantized reflection coefficients
+        for (i = 0; i < c->nchannels; i++) {
+            for (j = 0; j < b->adapt_pred_order[i]; j++) {
+                k = get_linear(&s->gb, 8);
+                if (k == -128) {
+                    av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL reflection coefficient index\n");
                     return AVERROR_INVALIDDATA;
                 }
-                chset->orig_chan_order[0][i]     = j;
-                chset->orig_chan_order_inv[0][j] = i;
-            }
-            for (i = 0; i < chset->channels / 2; i++) {
-                if (get_bits(&s->gb, 1)) /* bChPFlag */
-                    chset->pw_ch_pairs_coeffs[0][i] = get_bits_sm(&s->gb, 7);
+                if (k < 0)
+                    b->adapt_refl_coeff[i][j] = -(int)ff_dca_xll_refl_coeff[-k];
                 else
-                    chset->pw_ch_pairs_coeffs[0][i] = 0;
+                    b->adapt_refl_coeff[i][j] =  (int)ff_dca_xll_refl_coeff[ k];
             }
-        } else {
-            for (i = 0; i < chset->channels; i++)
-                chset->orig_chan_order[0][i]     =
-                chset->orig_chan_order_inv[0][i] = i;
-            for (i = 0; i < chset->channels / 2; i++)
-                chset->pw_ch_pairs_coeffs[0][i] = 0;
-        }
-        /* Adaptive prediction order */
-        chset->adapt_order_max[0] = 0;
-        for (i = 0; i < chset->channels; i++) {
-            chset->adapt_order[0][i] = get_bits(&s->gb, 4);
-            if (chset->adapt_order_max[0] < chset->adapt_order[0][i])
-                chset->adapt_order_max[0] = chset->adapt_order[0][i];
-        }
-        /* Fixed prediction order, used in case the adaptive order
-         * above is zero */
-        for (i = 0; i < chset->channels; i++)
-            chset->fixed_order[0][i] =
-                chset->adapt_order[0][i] ? 0 : get_bits(&s->gb, 2);
-
-        for (i = 0; i < chset->channels; i++) {
-            unsigned j;
-            for (j = 0; j < chset->adapt_order[0][i]; j++)
-                chset->lpc_refl_coeffs_q_ind[0][i][j] = get_bits(&s->gb, 8);
-        }
-
-        if (s->xll_scalable_lsb) {
-            chset->lsb_fsize[0] = get_bits(&s->gb, s->xll_bits4seg_size);
-
-            for (i = 0; i < chset->channels; i++)
-                chset->scalable_lsbs[0][i] = get_bits(&s->gb, 4);
-            for (i = 0; i < chset->channels; i++)
-                chset->bit_width_adj_per_ch[0][i] = get_bits(&s->gb, 4);
-        } else {
-            memset(chset->scalable_lsbs[0], 0,
-                   chset->channels * sizeof(chset->scalable_lsbs[0][0]));
-            memset(chset->bit_width_adj_per_ch[0], 0,
-                   chset->channels * sizeof(chset->bit_width_adj_per_ch[0][0]));
         }
 
-        s->xll_channels          += chset->channels;
-        s->xll_residual_channels += chset->channels -
-                                    av_popcount(chset->residual_encode);
+        // Downmix performed by encoder in extension frequency band
+        b->dmix_embedded = c->dmix_embedded && (band == 0 || get_bits1(&s->gb));
+
+        // MSB/LSB split flag in extension frequency band
+        if ((band == 0 && s->scalable_lsbs) || (band != 0 && get_bits1(&s->gb))) {
+            // Size of LSB section in any segment
+            b->lsb_section_size = get_bits_long(&s->gb, s->seg_size_nbits);
+            if (b->lsb_section_size < 0 || b->lsb_section_size > s->frame_size) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid LSB section size\n");
+                return AVERROR_INVALIDDATA;
+            }
 
-        /* FIXME: Parse header data for extra frequency bands. */
+            // Account for optional CRC bytes after LSB section
+            if (b->lsb_section_size && (s->band_crc_present > 2 ||
+                                        (band == 0 && s->band_crc_present > 1)))
+                b->lsb_section_size += 2;
 
-        /* Skip to end of channel set sub header. */
-        i = get_bits_count(&s->gb);
-        if (hdr_pos + 8 * hdr_size < i) {
-            av_log(s->avctx, AV_LOG_ERROR,
-                   "chset header too large, %d bits, should be <= %d bits\n",
-                   i - hdr_pos, 8 * hdr_size);
-            return AVERROR_INVALIDDATA;
+            // Number of bits to represent the samples in LSB part
+            for (i = 0; i < c->nchannels; i++) {
+                b->nscalablelsbs[i] = get_bits(&s->gb, 4);
+                if (b->nscalablelsbs[i] && !b->lsb_section_size) {
+                    av_log(s->avctx, AV_LOG_ERROR, "LSB section missing with non-zero LSB width\n");
+                    return AVERROR_INVALIDDATA;
+                }
+            }
+        } else {
+            b->lsb_section_size = 0;
+            for (i = 0; i < c->nchannels; i++)
+                b->nscalablelsbs[i] = 0;
+        }
+
+        // Scalable resolution flag in extension frequency band
+        if ((band == 0 && s->scalable_lsbs) || (band != 0 && get_bits1(&s->gb))) {
+            // Number of bits discarded by authoring
+            for (i = 0; i < c->nchannels; i++)
+                b->bit_width_adjust[i] = get_bits(&s->gb, 4);
+        } else {
+            for (i = 0; i < c->nchannels; i++)
+                b->bit_width_adjust[i] = 0;
         }
-        if (hdr_pos + 8 * hdr_size > i)
-            skip_bits_long(&s->gb, hdr_pos + 8 * hdr_size - i);
     }
+
+    // Reserved
+    // Byte align
+    // CRC16 of channel set sub-header
+    if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL sub-header\n");
+        return AVERROR_INVALIDDATA;
+    }
+
     return 0;
 }
 
-/* parse XLL navigation table */
-int ff_dca_xll_decode_navi(DCAContext *s, int asset_end)
+static int chs_alloc_msb_band_data(DCAXllDecoder *s, DCAXllChSet *c)
 {
-    int nbands, band, chset, seg, data_start;
-
-    /* FIXME: Supports only a single frequency band */
-    nbands = 1;
-
-    for (band = 0; band < nbands; band++) {
-        s->xll_navi.band_size[band] = 0;
-        for (seg = 0; seg < s->xll_segments; seg++) {
-            /* Note: The spec, ETSI TS 102 114 V1.4.1 (2012-09), says
-             * we should read a base value for segment_size from the
-             * stream, before reading the sizes of the channel sets.
-             * But that's apparently incorrect. */
-            s->xll_navi.segment_size[band][seg] = 0;
+    int ndecisamples = c->nfreqbands > 1 ? DCA_XLL_DECI_HISTORY_MAX : 0;
+    int nchsamples = s->nframesamples + ndecisamples;
+    int i, j, nsamples = nchsamples * c->nchannels * c->nfreqbands;
+    int32_t *ptr;
+
+    // Reallocate MSB sample buffer
+    av_fast_malloc(&c->sample_buffer[0], &c->sample_size[0], nsamples * sizeof(int32_t));
+    if (!c->sample_buffer[0])
+        return AVERROR(ENOMEM);
 
-            for (chset = 0; chset < s->xll_nch_sets; chset++)
-                if (band < s->xll_chsets[chset].num_freq_bands) {
-                    s->xll_navi.chset_size[band][seg][chset] =
-                        get_bits(&s->gb, s->xll_bits4seg_size) + 1;
-                    s->xll_navi.segment_size[band][seg] +=
-                        s->xll_navi.chset_size[band][seg][chset];
-                }
-            s->xll_navi.band_size[band] += s->xll_navi.segment_size[band][seg];
+    ptr = c->sample_buffer[0] + ndecisamples;
+    for (i = 0; i < c->nfreqbands; i++) {
+        for (j = 0; j < c->nchannels; j++) {
+            c->bands[i].msb_sample_buffer[j] = ptr;
+            ptr += nchsamples;
         }
     }
-    /* Align to 8 bits and skip 16-bit CRC. */
-    skip_bits_long(&s->gb, 16 + ((-get_bits_count(&s->gb)) & 7));
 
-    data_start = get_bits_count(&s->gb);
-    if (data_start + 8 * s->xll_navi.band_size[0] > asset_end) {
-        av_log(s->avctx, AV_LOG_ERROR,
-               "XLL: Data in NAVI table exceeds containing asset\n"
-               "start: %d (bit), size %u (bytes), end %d (bit), error %u\n",
-               data_start, s->xll_navi.band_size[0], asset_end,
-               data_start + 8 * s->xll_navi.band_size[0] - asset_end);
-        return AVERROR_INVALIDDATA;
-    }
-    init_get_bits(&s->xll_navi.gb, s->gb.buffer + data_start / 8,
-                  8 * s->xll_navi.band_size[0]);
     return 0;
 }
 
-static void dca_xll_inv_adapt_pred(int *samples, int nsamples, unsigned order,
-                                   const int *prev, const uint8_t *q_ind)
-{
-    static const uint16_t table[0x81] = {
-            0,  3070,  5110,  7140,  9156, 11154, 13132, 15085,
-        17010, 18904, 20764, 22588, 24373, 26117, 27818, 29474,
-        31085, 32648, 34164, 35631, 37049, 38418, 39738, 41008,
-        42230, 43404, 44530, 45609, 46642, 47630, 48575, 49477,
-        50337, 51157, 51937, 52681, 53387, 54059, 54697, 55302,
-        55876, 56421, 56937, 57426, 57888, 58326, 58741, 59132,
-        59502, 59852, 60182, 60494, 60789, 61066, 61328, 61576,
-        61809, 62029, 62236, 62431, 62615, 62788, 62951, 63105,
-        63250, 63386, 63514, 63635, 63749, 63855, 63956, 64051,
-        64140, 64224, 64302, 64376, 64446, 64512, 64573, 64631,
-        64686, 64737, 64785, 64830, 64873, 64913, 64950, 64986,
-        65019, 65050, 65079, 65107, 65133, 65157, 65180, 65202,
-        65222, 65241, 65259, 65275, 65291, 65306, 65320, 65333,
-        65345, 65357, 65368, 65378, 65387, 65396, 65405, 65413,
-        65420, 65427, 65434, 65440, 65446, 65451, 65456, 65461,
-        65466, 65470, 65474, 65478, 65481, 65485, 65488, 65491,
-        65535, /* Final value is for the -128 corner case, see below. */
-    };
-    int c[DCA_XLL_AORDER_MAX];
-    int64_t s;
-    unsigned i, j;
-
-    for (i = 0; i < order; i++) {
-        if (q_ind[i] & 1)
-            /* The index value 0xff corresponds to a lookup of entry 0x80 in
-             * the table, and no value is provided in the specification. */
-            c[i] = -table[(q_ind[i] >> 1) + 1];
-        else
-            c[i] = table[q_ind[i] >> 1];
-    }
-    /* The description in the spec is a bit convoluted. We can convert
-     * the reflected values to direct values in place, using a
-     * sequence of reflections operating on two values. */
-    for (i = 1; i < order; i++) {
-        /* i = 1: scale c[0]
-         * i = 2: reflect c[0] <-> c[1]
-         * i = 3: scale c[1], reflect c[0] <-> c[2]
-         * i = 4: reflect c[0] <-> c[3] reflect c[1] <-> c[2]
-         * ... */
-        if (i & 1)
-            c[i / 2] += ((int64_t) c[i] * c[i / 2] + 0x8000) >> 16;
-        for (j = 0; j < i / 2; j++) {
-            int r0 = c[j];
-            int r1 = c[i - j - 1];
-            c[j]         += ((int64_t) c[i] * r1 + 0x8000) >> 16;
-            c[i - j - 1] += ((int64_t) c[i] * r0 + 0x8000) >> 16;
-        }
-    }
-    /* Apply predictor. */
-    /* NOTE: Processing samples in this order means that the
-     * predictor is applied to the newly reconstructed samples. */
-    if (prev) {
-        for (i = 0; i < order; i++) {
-            for (j = s = 0; j < i; j++)
-                s += (int64_t) c[j] * samples[i - 1 - j];
-            for (; j < order; j++)
-                s += (int64_t) c[j] * prev[DCA_XLL_AORDER_MAX + i - 1 - j];
-
-            samples[i] -= av_clip((s + 0x8000) >> 16, -0x1000000, 0xffffff);
-        }
-    }
-    for (i = order; i < nsamples; i++) {
-        for (j = s = 0; j < order; j++)
-            s += (int64_t) c[j] * samples[i - 1 - j];
-
-        /* NOTE: Equations seem to imply addition, while the
-         * pseudocode seems to use subtraction.*/
-        samples[i] -= av_clip((s + 0x8000) >> 16, -0x1000000, 0xffffff);
-    }
-}
-
-int ff_dca_xll_decode_audio(DCAContext *s, AVFrame *frame)
-{
-    /* FIXME: Decodes only the first frequency band. */
-    int seg, chset_i;
-
-    /* Coding parameters for each channel set. */
-    struct coding_params {
-        int seg_type;
-        int rice_code_flag[16];
-        int pancAuxABIT[16];
-        int pancABIT0[16];  /* Not sure what this is */
-        int pancABIT[16];   /* Not sure what this is */
-        int nSamplPart0[16];
-    } param_state[16];
-
-    GetBitContext *gb = &s->xll_navi.gb;
-    int *history;
-
-    /* Layout: First the sample buffer for one segment per channel,
-     * followed by history buffers of DCA_XLL_AORDER_MAX samples for
-     * each channel. */
-    av_fast_malloc(&s->xll_sample_buf, &s->xll_sample_buf_size,
-                   (s->xll_smpl_in_seg + DCA_XLL_AORDER_MAX) *
-                   s->xll_channels * sizeof(*s->xll_sample_buf));
-    if (!s->xll_sample_buf)
+static int chs_alloc_lsb_band_data(DCAXllDecoder *s, DCAXllChSet *c)
+{
+    int i, j, nsamples = 0;
+    int32_t *ptr;
+
+    // Determine number of frequency bands that have MSB/LSB split
+    for (i = 0; i < c->nfreqbands; i++)
+        if (c->bands[i].lsb_section_size)
+            nsamples += s->nframesamples * c->nchannels;
+    if (!nsamples)
+        return 0;
+
+    // Reallocate LSB sample buffer
+    av_fast_malloc(&c->sample_buffer[1], &c->sample_size[1], nsamples * sizeof(int32_t));
+    if (!c->sample_buffer[1])
         return AVERROR(ENOMEM);
 
-    history = s->xll_sample_buf + s->xll_smpl_in_seg * s->xll_channels;
-
-    for (seg = 0; seg < s->xll_segments; seg++) {
-        unsigned in_channel;
-
-        for (chset_i = in_channel = 0; chset_i < s->xll_nch_sets; chset_i++) {
-            /* The spec isn't very explicit, but I think the NAVI sizes are in bytes. */
-            int end_pos = get_bits_count(gb) +
-                          8 * s->xll_navi.chset_size[0][seg][chset_i];
-            int i, j;
-            struct coding_params *params = &param_state[chset_i];
-            /* I think this flag means that we should keep seg_type and
-             * other parameters from the previous segment. */
-            int use_seg_state_code_param;
-            XllChSetSubHeader *chset = &s->xll_chsets[chset_i];
-            if (in_channel >= s->avctx->channels)
-                /* FIXME: Could go directly to next segment */
-                goto next_chset;
-
-            if (s->avctx->sample_rate != chset->sampling_frequency) {
-                av_log(s->avctx, AV_LOG_WARNING,
-                       "XLL: unexpected chset sample rate %d, expected %d\n",
-                       chset->sampling_frequency, s->avctx->sample_rate);
-                goto next_chset;
+    ptr = c->sample_buffer[1];
+    for (i = 0; i < c->nfreqbands; i++) {
+        if (c->bands[i].lsb_section_size) {
+            for (j = 0; j < c->nchannels; j++) {
+                c->bands[i].lsb_sample_buffer[j] = ptr;
+                ptr += s->nframesamples;
             }
-            if (seg != 0)
-                use_seg_state_code_param = get_bits(gb, 1);
+        } else {
+            for (j = 0; j < c->nchannels; j++)
+                c->bands[i].lsb_sample_buffer[j] = NULL;
+        }
+    }
+
+    return 0;
+}
+
+static int chs_parse_band_data(DCAXllDecoder *s, DCAXllChSet *c, int band, int seg, int band_data_end)
+{
+    DCAXllBand *b = &c->bands[band];
+    int i, j, k;
+
+    // Start unpacking MSB portion of the segment
+    if (!(seg && get_bits1(&s->gb))) {
+        // Unpack segment type
+        // 0 - distinct coding parameters for each channel
+        // 1 - common coding parameters for all channels
+        c->seg_common = get_bits1(&s->gb);
+
+        // Determine number of coding parameters encoded in segment
+        k = c->seg_common ? 1 : c->nchannels;
+
+        // Unpack Rice coding parameters
+        for (i = 0; i < k; i++) {
+            // Unpack Rice coding flag
+            // 0 - linear code, 1 - Rice code
+            c->rice_code_flag[i] = get_bits1(&s->gb);
+            // Unpack Hybrid Rice coding flag
+            // 0 - Rice code, 1 - Hybrid Rice code
+            if (!c->seg_common && c->rice_code_flag[i] && get_bits1(&s->gb))
+                // Unpack binary code length for isolated samples
+                c->bitalloc_hybrid_linear[i] = get_bits(&s->gb, c->nabits) + 1;
             else
-                use_seg_state_code_param = 0;
+                // 0 indicates no Hybrid Rice coding
+                c->bitalloc_hybrid_linear[i] = 0;
+        }
 
-            if (!use_seg_state_code_param) {
-                int num_param_sets, i;
-                unsigned bits4ABIT;
+        // Unpack coding parameters
+        for (i = 0; i < k; i++) {
+            if (seg == 0) {
+                // Unpack coding parameter for part A of segment 0
+                c->bitalloc_part_a[i] = get_bits(&s->gb, c->nabits);
 
-                params->seg_type = get_bits(gb, 1);
-                num_param_sets   = params->seg_type ? 1 : chset->channels;
+                // Adjust for the linear code
+                if (!c->rice_code_flag[i] && c->bitalloc_part_a[i])
+                    c->bitalloc_part_a[i]++;
 
-                if (chset->bit_width > 16) {
-                    bits4ABIT = 5;
-                } else {
-                    if (chset->bit_width > 8)
-                        bits4ABIT = 4;
-                    else
-                        bits4ABIT = 3;
-                    if (s->xll_nch_sets > 1)
-                        bits4ABIT++;
+                if (!c->seg_common)
+                    c->nsamples_part_a[i] = b->adapt_pred_order[i];
+                else
+                    c->nsamples_part_a[i] = b->highest_pred_order;
+            } else {
+                c->bitalloc_part_a[i] = 0;
+                c->nsamples_part_a[i] = 0;
+            }
+
+            // Unpack coding parameter for part B of segment
+            c->bitalloc_part_b[i] = get_bits(&s->gb, c->nabits);
+
+            // Adjust for the linear code
+            if (!c->rice_code_flag[i] && c->bitalloc_part_b[i])
+                c->bitalloc_part_b[i]++;
+        }
+    }
+
+    // Unpack entropy codes
+    for (i = 0; i < c->nchannels; i++) {
+        int32_t *part_a, *part_b;
+        int nsamples_part_b;
+
+        // Select index of coding parameters
+        k = c->seg_common ? 0 : i;
+
+        // Slice the segment into parts A and B
+        part_a = b->msb_sample_buffer[i] + seg * s->nsegsamples;
+        part_b = part_a + c->nsamples_part_a[k];
+        nsamples_part_b = s->nsegsamples - c->nsamples_part_a[k];
+
+        if (get_bits_left(&s->gb) < 0)
+            return AVERROR_INVALIDDATA;
+
+        if (!c->rice_code_flag[k]) {
+            // Linear codes
+            // Unpack all residuals of part A of segment 0
+            get_linear_array(&s->gb, part_a, c->nsamples_part_a[k],
+                             c->bitalloc_part_a[k]);
+
+            // Unpack all residuals of part B of segment 0 and others
+            get_linear_array(&s->gb, part_b, nsamples_part_b,
+                             c->bitalloc_part_b[k]);
+        } else {
+            // Rice codes
+            // Unpack all residuals of part A of segment 0
+            get_rice_array(&s->gb, part_a, c->nsamples_part_a[k],
+                           c->bitalloc_part_a[k]);
+
+            if (c->bitalloc_hybrid_linear[k]) {
+                // Hybrid Rice codes
+                // Unpack the number of isolated samples
+                int nisosamples = get_bits(&s->gb, s->nsegsamples_log2);
+
+                // Set all locations to 0
+                memset(part_b, 0, sizeof(*part_b) * nsamples_part_b);
+
+                // Extract the locations of isolated samples and flag by -1
+                for (j = 0; j < nisosamples; j++) {
+                    int loc = get_bits(&s->gb, s->nsegsamples_log2);
+                    if (loc >= nsamples_part_b) {
+                        av_log(s->avctx, AV_LOG_ERROR, "Invalid isolated sample location\n");
+                        return AVERROR_INVALIDDATA;
+                    }
+                    part_b[loc] = -1;
                 }
 
-                for (i = 0; i < num_param_sets; i++) {
-                    params->rice_code_flag[i] = get_bits(gb, 1);
-                    if (!params->seg_type && params->rice_code_flag[i] && get_bits(gb, 1))
-                        params->pancAuxABIT[i] = get_bits(gb, bits4ABIT) + 1;
+                // Unpack all residuals of part B of segment 0 and others
+                for (j = 0; j < nsamples_part_b; j++) {
+                    if (part_b[j])
+                        part_b[j] = get_linear(&s->gb, c->bitalloc_hybrid_linear[k]);
                     else
-                        params->pancAuxABIT[i] = 0;
+                        part_b[j] = get_rice(&s->gb, c->bitalloc_part_b[k]);
                 }
+            } else {
+                // Rice codes
+                // Unpack all residuals of part B of segment 0 and others
+                get_rice_array(&s->gb, part_b, nsamples_part_b, c->bitalloc_part_b[k]);
+            }
+        }
+    }
 
-                for (i = 0; i < num_param_sets; i++) {
-                    if (!seg) {
-                        /* Parameters for part 1 */
-                        params->pancABIT0[i] = get_bits(gb, bits4ABIT);
-                        if (params->rice_code_flag[i] == 0 && params->pancABIT0[i] > 0)
-                            /* For linear code */
-                            params->pancABIT0[i]++;
-
-                        /* NOTE: In the spec, not indexed by band??? */
-                        if (params->seg_type == 0)
-                            params->nSamplPart0[i] = chset->adapt_order[0][i];
-                        else
-                            params->nSamplPart0[i] = chset->adapt_order_max[0];
-                    } else
-                        params->nSamplPart0[i] = 0;
-
-                    /* Parameters for part 2 */
-                    params->pancABIT[i] = get_bits(gb, bits4ABIT);
-                    if (params->rice_code_flag[i] == 0 && params->pancABIT[i] > 0)
-                        /* For linear code */
-                        params->pancABIT[i]++;
-                }
+    // Unpack decimator history for frequency band 1
+    if (seg == 0 && band == 1) {
+        int nbits = get_bits(&s->gb, 5) + 1;
+        for (i = 0; i < c->nchannels; i++)
+            for (j = 1; j < DCA_XLL_DECI_HISTORY_MAX; j++)
+                c->deci_history[i][j] = get_sbits_long(&s->gb, nbits);
+    }
+
+    // Start unpacking LSB portion of the segment
+    if (b->lsb_section_size) {
+        // Skip to the start of LSB portion
+        if (ff_dca_seek_bits(&s->gb, band_data_end - b->lsb_section_size * 8)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL band data\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Unpack all LSB parts of residuals of this segment
+        for (i = 0; i < c->nchannels; i++) {
+            if (b->nscalablelsbs[i]) {
+                get_array(&s->gb,
+                          b->lsb_sample_buffer[i] + seg * s->nsegsamples,
+                          s->nsegsamples, b->nscalablelsbs[i]);
             }
-            for (i = 0; i < chset->channels; i++) {
-                int param_index = params->seg_type ? 0 : i;
-                int part0       = params->nSamplPart0[param_index];
-                int bits        = part0 ? params->pancABIT0[param_index] : 0;
-                int *sample_buf = s->xll_sample_buf +
-                                  (in_channel + i) * s->xll_smpl_in_seg;
-
-                if (!params->rice_code_flag[param_index]) {
-                    /* Linear code */
-                    if (bits)
-                        for (j = 0; j < part0; j++)
-                            sample_buf[j] = get_bits_sm(gb, bits);
-                    else
-                        memset(sample_buf, 0, part0 * sizeof(sample_buf[0]));
+        }
+    }
 
-                    /* Second part */
-                    bits = params->pancABIT[param_index];
-                    if (bits)
-                        for (j = part0; j < s->xll_smpl_in_seg; j++)
-                            sample_buf[j] = get_bits_sm(gb, bits);
-                    else
-                        memset(sample_buf + part0, 0,
-                               (s->xll_smpl_in_seg - part0) * sizeof(sample_buf[0]));
-                } else {
-                    int aux_bits = params->pancAuxABIT[param_index];
-
-                    for (j = 0; j < part0; j++) {
-                        /* FIXME: Is this identical to Golomb code? */
-                        int t = get_unary(gb, 1, 33) << bits;
-                        /* FIXME: Could move this test outside of the loop, for efficiency. */
-                        if (bits)
-                            t |= get_bits(gb, bits);
-                        sample_buf[j] = (t & 1) ? -(t >> 1) - 1 : (t >> 1);
-                    }
+    // Skip to the end of band data
+    if (ff_dca_seek_bits(&s->gb, band_data_end)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL band data\n");
+        return AVERROR_INVALIDDATA;
+    }
 
-                    /* Second part */
-                    bits = params->pancABIT[param_index];
-
-                    /* Follow the spec's suggestion of using the
-                     * buffer also to store the hybrid-rice flags. */
-                    memset(sample_buf + part0, 0,
-                           (s->xll_smpl_in_seg - part0) * sizeof(sample_buf[0]));
-
-                    if (aux_bits > 0) {
-                        /* For hybrid rice encoding, some samples are linearly
-                         * coded. According to the spec, "nBits4SamplLoci" bits
-                         * are used for each index, but this value is not
-                         * defined. I guess we should use log2(xll_smpl_in_seg)
-                         * bits. */
-                        int count = get_bits(gb, s->xll_log_smpl_in_seg);
-                        av_log(s->avctx, AV_LOG_DEBUG, "aux count %d (bits %d)\n",
-                               count, s->xll_log_smpl_in_seg);
-
-                        for (j = 0; j < count; j++)
-                            sample_buf[get_bits(gb, s->xll_log_smpl_in_seg)] = 1;
-                    }
-                    for (j = part0; j < s->xll_smpl_in_seg; j++) {
-                        if (!sample_buf[j]) {
-                            int t = get_unary(gb, 1, 33);
-                            if (bits)
-                                t = (t << bits) | get_bits(gb, bits);
-                            sample_buf[j] = (t & 1) ? -(t >> 1) - 1 : (t >> 1);
-                        } else
-                            sample_buf[j] = get_bits_sm(gb, aux_bits);
-                    }
+    return 0;
+}
+
+static av_cold void chs_clear_band_data(DCAXllDecoder *s, DCAXllChSet *c, int band, int seg)
+{
+    DCAXllBand *b = &c->bands[band];
+    int i, offset, nsamples;
+
+    if (seg < 0) {
+        offset = 0;
+        nsamples = s->nframesamples;
+    } else {
+        offset = seg * s->nsegsamples;
+        nsamples = s->nsegsamples;
+    }
+
+    for (i = 0; i < c->nchannels; i++) {
+        memset(b->msb_sample_buffer[i] + offset, 0, nsamples * sizeof(int32_t));
+        if (b->lsb_section_size)
+            memset(b->lsb_sample_buffer[i] + offset, 0, nsamples * sizeof(int32_t));
+    }
+
+    if (seg <= 0 && band)
+        memset(c->deci_history, 0, sizeof(c->deci_history));
+
+    if (seg < 0) {
+        memset(b->nscalablelsbs, 0, sizeof(b->nscalablelsbs));
+        memset(b->bit_width_adjust, 0, sizeof(b->bit_width_adjust));
+    }
+}
+
+static void chs_filter_band_data(DCAXllDecoder *s, DCAXllChSet *c, int band)
+{
+    DCAXllBand *b = &c->bands[band];
+    int nsamples = s->nframesamples;
+    int i, j, k;
+
+    // Inverse adaptive or fixed prediction
+    for (i = 0; i < c->nchannels; i++) {
+        int32_t *buf = b->msb_sample_buffer[i];
+        int order = b->adapt_pred_order[i];
+        if (order > 0) {
+            int coeff[DCA_XLL_ADAPT_PRED_ORDER_MAX];
+            // Conversion from reflection coefficients to direct form coefficients
+            for (j = 0; j < order; j++) {
+                int rc = b->adapt_refl_coeff[i][j];
+                for (k = 0; k < (j + 1) / 2; k++) {
+                    int tmp1 = coeff[    k    ];
+                    int tmp2 = coeff[j - k - 1];
+                    coeff[    k    ] = tmp1 + mul16(rc, tmp2);
+                    coeff[j - k - 1] = tmp2 + mul16(rc, tmp1);
                 }
+                coeff[j] = rc;
+            }
+            // Inverse adaptive prediction
+            for (j = 0; j < nsamples - order; j++) {
+                int64_t err = 0;
+                for (k = 0; k < order; k++)
+                    err += (int64_t)buf[j + k] * coeff[order - k - 1];
+                buf[j + k] -= (SUINT)clip23(norm16(err));
             }
+        } else {
+            // Inverse fixed coefficient prediction
+            for (j = 0; j < b->fixed_pred_order[i]; j++)
+                for (k = 1; k < nsamples; k++)
+                    buf[k] += (unsigned)buf[k - 1];
+        }
+    }
 
-            for (i = 0; i < chset->channels; i++) {
-                unsigned adapt_order = chset->adapt_order[0][i];
-                int *sample_buf = s->xll_sample_buf +
-                                  (in_channel + i) * s->xll_smpl_in_seg;
-                int *prev = history + (in_channel + i) * DCA_XLL_AORDER_MAX;
-
-                if (!adapt_order) {
-                    unsigned order;
-                    for (order = chset->fixed_order[0][i]; order > 0; order--) {
-                        unsigned j;
-                        for (j = 1; j < s->xll_smpl_in_seg; j++)
-                            sample_buf[j] += sample_buf[j - 1];
-                    }
-                } else
-                    /* Inverse adaptive prediction, in place. */
-                    dca_xll_inv_adapt_pred(sample_buf, s->xll_smpl_in_seg,
-                                           adapt_order, seg ? prev : NULL,
-                                           chset->lpc_refl_coeffs_q_ind[0][i]);
-                memcpy(prev, sample_buf + s->xll_smpl_in_seg - DCA_XLL_AORDER_MAX,
-                       DCA_XLL_AORDER_MAX * sizeof(*prev));
+    // Inverse pairwise channel decorrellation
+    if (b->decor_enabled) {
+        int32_t *tmp[DCA_XLL_CHANNELS_MAX];
+
+        for (i = 0; i < c->nchannels / 2; i++) {
+            int coeff = b->decor_coeff[i];
+            if (coeff) {
+                s->dcadsp->decor(b->msb_sample_buffer[i * 2 + 1],
+                                 b->msb_sample_buffer[i * 2    ],
+                                 coeff, nsamples);
             }
-            for (i = 1; i < chset->channels; i += 2) {
-                int coeff = chset->pw_ch_pairs_coeffs[0][i / 2];
-                if (coeff != 0) {
-                    int *sample_buf = s->xll_sample_buf +
-                                      (in_channel + i) * s->xll_smpl_in_seg;
-                    int *prev = sample_buf - s->xll_smpl_in_seg;
-                    unsigned j;
-                    for (j = 0; j < s->xll_smpl_in_seg; j++)
-                        /* Shift is unspecified, but should apparently be 3. */
-                        sample_buf[j] += ((int64_t) coeff * prev[j] + 4) >> 3;
-                }
+        }
+
+        // Reorder channel pointers to the original order
+        for (i = 0; i < c->nchannels; i++)
+            tmp[i] = b->msb_sample_buffer[i];
+
+        for (i = 0; i < c->nchannels; i++)
+            b->msb_sample_buffer[b->orig_order[i]] = tmp[i];
+    }
+
+    // Map output channel pointers for frequency band 0
+    if (c->nfreqbands == 1)
+        for (i = 0; i < c->nchannels; i++)
+            s->output_samples[c->ch_remap[i]] = b->msb_sample_buffer[i];
+}
+
+static int chs_get_lsb_width(DCAXllDecoder *s, DCAXllChSet *c, int band, int ch)
+{
+    int adj = c->bands[band].bit_width_adjust[ch];
+    int shift = c->bands[band].nscalablelsbs[ch];
+
+    if (s->fixed_lsb_width)
+        shift = s->fixed_lsb_width;
+    else if (shift && adj)
+        shift += adj - 1;
+    else
+        shift += adj;
+
+    return shift;
+}
+
+static void chs_assemble_msbs_lsbs(DCAXllDecoder *s, DCAXllChSet *c, int band)
+{
+    DCAXllBand *b = &c->bands[band];
+    int n, ch, nsamples = s->nframesamples;
+
+    for (ch = 0; ch < c->nchannels; ch++) {
+        int shift = chs_get_lsb_width(s, c, band, ch);
+        if (shift) {
+            int32_t *msb = b->msb_sample_buffer[ch];
+            if (b->nscalablelsbs[ch]) {
+                int32_t *lsb = b->lsb_sample_buffer[ch];
+                int adj = b->bit_width_adjust[ch];
+                for (n = 0; n < nsamples; n++)
+                    msb[n] = msb[n] * (SUINT)(1 << shift) + (lsb[n] << adj);
+            } else {
+                for (n = 0; n < nsamples; n++)
+                    msb[n] = msb[n] * (SUINT)(1 << shift);
             }
+        }
+    }
+}
 
-            if (s->xll_scalable_lsb) {
-                int lsb_start = end_pos - 8 * chset->lsb_fsize[0] -
-                                8 * (s->xll_banddata_crc & 2);
-                int done;
-                i = get_bits_count(gb);
-                if (i > lsb_start) {
-                    av_log(s->avctx, AV_LOG_ERROR,
-                           "chset data lsb exceeds NAVI size, end_pos %d, lsb_start %d, pos %d\n",
-                           end_pos, lsb_start, i);
-                    return AVERROR_INVALIDDATA;
-                }
-                if (i < lsb_start)
-                    skip_bits_long(gb, lsb_start - i);
-
-                for (i = done = 0; i < chset->channels; i++) {
-                    int bits = chset->scalable_lsbs[0][i];
-                    if (bits > 0) {
-                        /* The channel reordering is conceptually done
-                         * before adding the lsb:s, so we need to do
-                         * the inverse permutation here. */
-                        unsigned pi = chset->orig_chan_order_inv[0][i];
-                        int *sample_buf = s->xll_sample_buf +
-                                          (in_channel + pi) * s->xll_smpl_in_seg;
-                        int adj = chset->bit_width_adj_per_ch[0][i];
-                        int msb_shift = bits;
-                        unsigned j;
-
-                        if (adj > 0)
-                            msb_shift += adj - 1;
-
-                        for (j = 0; j < s->xll_smpl_in_seg; j++)
-                            sample_buf[j] = (sample_buf[j] << msb_shift) +
-                                            (get_bits(gb, bits) << adj);
-
-                        done += bits * s->xll_smpl_in_seg;
+static int chs_assemble_freq_bands(DCAXllDecoder *s, DCAXllChSet *c)
+{
+    int ch, nsamples = s->nframesamples;
+    int32_t *ptr;
+
+    av_assert1(c->nfreqbands > 1);
+
+    // Reallocate frequency band assembly buffer
+    av_fast_malloc(&c->sample_buffer[2], &c->sample_size[2],
+                   2 * nsamples * c->nchannels * sizeof(int32_t));
+    if (!c->sample_buffer[2])
+        return AVERROR(ENOMEM);
+
+    // Assemble frequency bands 0 and 1
+    ptr = c->sample_buffer[2];
+    for (ch = 0; ch < c->nchannels; ch++) {
+        int32_t *band0 = c->bands[0].msb_sample_buffer[ch];
+        int32_t *band1 = c->bands[1].msb_sample_buffer[ch];
+
+        // Copy decimator history
+        memcpy(band0 - DCA_XLL_DECI_HISTORY_MAX,
+               c->deci_history[ch], sizeof(c->deci_history[0]));
+
+        // Filter
+        s->dcadsp->assemble_freq_bands(ptr, band0, band1,
+                                       ff_dca_xll_band_coeff,
+                                       nsamples);
+
+        // Remap output channel pointer to assembly buffer
+        s->output_samples[c->ch_remap[ch]] = ptr;
+        ptr += nsamples * 2;
+    }
+
+    return 0;
+}
+
+static int parse_common_header(DCAXllDecoder *s)
+{
+    int stream_ver, header_size, frame_size_nbits, nframesegs_log2;
+
+    // XLL extension sync word
+    if (get_bits_long(&s->gb, 32) != DCA_SYNCWORD_XLL) {
+        av_log(s->avctx, AV_LOG_VERBOSE, "Invalid XLL sync word\n");
+        return AVERROR(EAGAIN);
+    }
+
+    // Version number
+    stream_ver = get_bits(&s->gb, 4) + 1;
+    if (stream_ver > 1) {
+        avpriv_request_sample(s->avctx, "XLL stream version %d", stream_ver);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Lossless frame header length
+    header_size = get_bits(&s->gb, 8) + 1;
+
+    // Check CRC
+    if (ff_dca_check_crc(s->avctx, &s->gb, 32, header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL common header checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Number of bits used to read frame size
+    frame_size_nbits = get_bits(&s->gb, 5) + 1;
+
+    // Number of bytes in a lossless frame
+    s->frame_size = get_bits_long(&s->gb, frame_size_nbits);
+    if (s->frame_size < 0 || s->frame_size >= DCA_XLL_PBR_BUFFER_MAX) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL frame size (%d bytes)\n", s->frame_size);
+        return AVERROR_INVALIDDATA;
+    }
+    s->frame_size++;
+
+    // Number of channels sets per frame
+    s->nchsets = get_bits(&s->gb, 4) + 1;
+    if (s->nchsets > DCA_XLL_CHSETS_MAX) {
+        avpriv_request_sample(s->avctx, "%d XLL channel sets", s->nchsets);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Number of segments per frame
+    nframesegs_log2 = get_bits(&s->gb, 4);
+    s->nframesegs = 1 << nframesegs_log2;
+    if (s->nframesegs > 1024) {
+        av_log(s->avctx, AV_LOG_ERROR, "Too many segments per XLL frame\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Samples in segment per one frequency band for the first channel set
+    // Maximum value is 256 for sampling frequencies <= 48 kHz
+    // Maximum value is 512 for sampling frequencies > 48 kHz
+    s->nsegsamples_log2 = get_bits(&s->gb, 4);
+    if (!s->nsegsamples_log2) {
+        av_log(s->avctx, AV_LOG_ERROR, "Too few samples per XLL segment\n");
+        return AVERROR_INVALIDDATA;
+    }
+    s->nsegsamples = 1 << s->nsegsamples_log2;
+    if (s->nsegsamples > 512) {
+        av_log(s->avctx, AV_LOG_ERROR, "Too many samples per XLL segment\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Samples in frame per one frequency band for the first channel set
+    s->nframesamples_log2 = s->nsegsamples_log2 + nframesegs_log2;
+    s->nframesamples = 1 << s->nframesamples_log2;
+    if (s->nframesamples > 65536) {
+        av_log(s->avctx, AV_LOG_ERROR, "Too many samples per XLL frame\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Number of bits used to read segment size
+    s->seg_size_nbits = get_bits(&s->gb, 5) + 1;
+
+    // Presence of CRC16 within each frequency band
+    // 0 - No CRC16 within band
+    // 1 - CRC16 placed at the end of MSB0
+    // 2 - CRC16 placed at the end of MSB0 and LSB0
+    // 3 - CRC16 placed at the end of MSB0 and LSB0 and other frequency bands
+    s->band_crc_present = get_bits(&s->gb, 2);
+
+    // MSB/LSB split flag
+    s->scalable_lsbs = get_bits1(&s->gb);
+
+    // Channel position mask
+    s->ch_mask_nbits = get_bits(&s->gb, 5) + 1;
+
+    // Fixed LSB width
+    if (s->scalable_lsbs)
+        s->fixed_lsb_width = get_bits(&s->gb, 4);
+    else
+        s->fixed_lsb_width = 0;
+
+    // Reserved
+    // Byte align
+    // Header CRC16 protection
+    if (ff_dca_seek_bits(&s->gb, header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL common header\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static int is_hier_dmix_chset(DCAXllChSet *c)
+{
+    return !c->primary_chset && c->dmix_embedded && c->hier_chset;
+}
+
+static DCAXllChSet *find_next_hier_dmix_chset(DCAXllDecoder *s, DCAXllChSet *c)
+{
+    if (c->hier_chset)
+        while (++c < &s->chset[s->nchsets])
+            if (is_hier_dmix_chset(c))
+                return c;
+
+    return NULL;
+}
+
+static void prescale_down_mix(DCAXllChSet *c, DCAXllChSet *o)
+{
+    int i, j, *coeff_ptr = c->dmix_coeff;
+
+    for (i = 0; i < c->hier_ofs; i++) {
+        int scale = o->dmix_scale[i];
+        int scale_inv = o->dmix_scale_inv[i];
+        c->dmix_scale[i] = mul15(c->dmix_scale[i], scale);
+        c->dmix_scale_inv[i] = mul16(c->dmix_scale_inv[i], scale_inv);
+        for (j = 0; j < c->nchannels; j++) {
+            int coeff = mul16(*coeff_ptr, scale_inv);
+            *coeff_ptr++ = mul15(coeff, o->dmix_scale[c->hier_ofs + j]);
+        }
+    }
+}
+
+static int parse_sub_headers(DCAXllDecoder *s, DCAExssAsset *asset)
+{
+    DCAContext *dca = s->avctx->priv_data;
+    DCAXllChSet *c;
+    int i, ret;
+
+    // Parse channel set headers
+    s->nfreqbands = 0;
+    s->nchannels = 0;
+    s->nreschsets = 0;
+    for (i = 0, c = s->chset; i < s->nchsets; i++, c++) {
+        c->hier_ofs = s->nchannels;
+        if ((ret = chs_parse_header(s, c, asset)) < 0)
+            return ret;
+        if (c->nfreqbands > s->nfreqbands)
+            s->nfreqbands = c->nfreqbands;
+        if (c->hier_chset)
+            s->nchannels += c->nchannels;
+        if (c->residual_encode != (1 << c->nchannels) - 1)
+            s->nreschsets++;
+    }
+
+    // Pre-scale downmixing coefficients for all non-primary channel sets
+    for (i = s->nchsets - 1, c = &s->chset[i]; i > 0; i--, c--) {
+        if (is_hier_dmix_chset(c)) {
+            DCAXllChSet *o = find_next_hier_dmix_chset(s, c);
+            if (o)
+                prescale_down_mix(c, o);
+        }
+    }
+
+    // Determine number of active channel sets to decode
+    switch (dca->request_channel_layout) {
+    case DCA_SPEAKER_LAYOUT_STEREO:
+        s->nactivechsets = 1;
+        break;
+    case DCA_SPEAKER_LAYOUT_5POINT0:
+    case DCA_SPEAKER_LAYOUT_5POINT1:
+        s->nactivechsets = (s->chset[0].nchannels < 5 && s->nchsets > 1) ? 2 : 1;
+        break;
+    default:
+        s->nactivechsets = s->nchsets;
+        break;
+    }
+
+    return 0;
+}
+
+static int parse_navi_table(DCAXllDecoder *s)
+{
+    int chs, seg, band, navi_nb, navi_pos, *navi_ptr;
+    DCAXllChSet *c;
+
+    // Determine size of NAVI table
+    navi_nb = s->nfreqbands * s->nframesegs * s->nchsets;
+    if (navi_nb > 1024) {
+        av_log(s->avctx, AV_LOG_ERROR, "Too many NAVI entries (%d)\n", navi_nb);
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Reallocate NAVI table
+    av_fast_malloc(&s->navi, &s->navi_size, navi_nb * sizeof(*s->navi));
+    if (!s->navi)
+        return AVERROR(ENOMEM);
+
+    // Parse NAVI
+    navi_pos = get_bits_count(&s->gb);
+    navi_ptr = s->navi;
+    for (band = 0; band < s->nfreqbands; band++) {
+        for (seg = 0; seg < s->nframesegs; seg++) {
+            for (chs = 0, c = s->chset; chs < s->nchsets; chs++, c++) {
+                int size = 0;
+                if (c->nfreqbands > band) {
+                    size = get_bits_long(&s->gb, s->seg_size_nbits);
+                    if (size < 0 || size >= s->frame_size) {
+                        av_log(s->avctx, AV_LOG_ERROR, "Invalid NAVI segment size (%d bytes)\n", size);
+                        return AVERROR_INVALIDDATA;
                     }
+                    size++;
                 }
-                if (done > 8 * chset->lsb_fsize[0]) {
-                    av_log(s->avctx, AV_LOG_ERROR,
-                           "chset lsb exceeds lsb_size\n");
-                    return AVERROR_INVALIDDATA;
-                }
+                *navi_ptr++ = size;
             }
+        }
+    }
+
+    // Byte align
+    // CRC16
+    skip_bits(&s->gb, -get_bits_count(&s->gb) & 7);
+    skip_bits(&s->gb, 16);
+
+    // Check CRC
+    if (ff_dca_check_crc(s->avctx, &s->gb, navi_pos, get_bits_count(&s->gb))) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid NAVI checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static int parse_band_data(DCAXllDecoder *s)
+{
+    int ret, chs, seg, band, navi_pos, *navi_ptr;
+    DCAXllChSet *c;
+
+    for (chs = 0, c = s->chset; chs < s->nactivechsets; chs++, c++) {
+        if ((ret = chs_alloc_msb_band_data(s, c)) < 0)
+            return ret;
+        if ((ret = chs_alloc_lsb_band_data(s, c)) < 0)
+            return ret;
+    }
 
-            /* Store output. */
-            for (i = 0; i < chset->channels; i++) {
-                int *sample_buf = s->xll_sample_buf +
-                                  (in_channel + i) * s->xll_smpl_in_seg;
-                int shift = 1 - chset->bit_resolution;
-                int out_channel = chset->orig_chan_order[0][i];
-                float *out;
-
-                /* XLL uses the channel order C, L, R, and we want L,
-                 * R, C. FIXME: Generalize. */
-                if (chset->ch_mask_enabled &&
-                    (chset->ch_mask & 7) == 7 && out_channel < 3)
-                    out_channel = out_channel ? out_channel - 1 : 2;
-
-                out_channel += in_channel;
-                if (out_channel >= s->avctx->channels)
-                    continue;
-
-                out  = (float *) frame->extended_data[out_channel];
-                out += seg * s->xll_smpl_in_seg;
-
-                /* NOTE: A one bit means residual encoding is *not* used. */
-                if ((chset->residual_encode >> i) & 1) {
-                    /* Replace channel samples.
-                     * FIXME: Most likely not the right thing to do. */
-                    for (j = 0; j < s->xll_smpl_in_seg; j++)
-                        out[j] = ldexpf(sample_buf[j], shift);
-                } else {
-                    /* Add residual signal to core channel */
-                    for (j = 0; j < s->xll_smpl_in_seg; j++)
-                        out[j] += ldexpf(sample_buf[j], shift);
+    navi_pos = get_bits_count(&s->gb);
+    navi_ptr = s->navi;
+    for (band = 0; band < s->nfreqbands; band++) {
+        for (seg = 0; seg < s->nframesegs; seg++) {
+            for (chs = 0, c = s->chset; chs < s->nchsets; chs++, c++) {
+                if (c->nfreqbands > band) {
+                    navi_pos += *navi_ptr * 8;
+                    if (navi_pos > s->gb.size_in_bits) {
+                        av_log(s->avctx, AV_LOG_ERROR, "Invalid NAVI position\n");
+                        return AVERROR_INVALIDDATA;
+                    }
+                    if (chs < s->nactivechsets &&
+                        (ret = chs_parse_band_data(s, c, band, seg, navi_pos)) < 0) {
+                        if (s->avctx->err_recognition & AV_EF_EXPLODE)
+                            return ret;
+                        chs_clear_band_data(s, c, band, seg);
+                    }
+                    skip_bits_long(&s->gb, navi_pos - get_bits_count(&s->gb));
                 }
+                navi_ptr++;
             }
+        }
+    }
 
-            if (chset->downmix_coeff_code_embedded &&
-                !chset->primary_ch_set && chset->hier_chset) {
-                /* Undo hierarchical downmix of earlier channels. */
-                unsigned mix_channel;
-                for (mix_channel = 0; mix_channel < in_channel; mix_channel++) {
-                    float *mix_buf;
-                    const int *col;
-                    float coeff;
-                    unsigned row;
-                    /* Similar channel reorder C, L, R vs L, R, C reorder. */
-                    if (chset->ch_mask_enabled &&
-                        (chset->ch_mask & 7) == 7 && mix_channel < 3)
-                        mix_buf = (float *) frame->extended_data[mix_channel ? mix_channel - 1 : 2];
-                    else
-                        mix_buf = (float *) frame->extended_data[mix_channel];
-
-                    mix_buf += seg * s->xll_smpl_in_seg;
-                    col = &chset->downmix_coeffs[mix_channel * (chset->channels + 1)];
-
-                    /* Scale */
-                    coeff = ldexpf(col[0], -16);
-                    for (j = 0; j < s->xll_smpl_in_seg; j++)
-                        mix_buf[j] *= coeff;
-
-                    for (row = 0;
-                         row < chset->channels && in_channel + row < s->avctx->channels;
-                         row++)
-                        if (col[row + 1]) {
-                            const float *new_channel =
-                                (const float *) frame->extended_data[in_channel + row];
-                            new_channel += seg * s->xll_smpl_in_seg;
-                            coeff        = ldexpf(col[row + 1], -15);
-                            for (j = 0; j < s->xll_smpl_in_seg; j++)
-                                mix_buf[j] -= coeff * new_channel[j];
-                        }
+    return 0;
+}
+
+static int parse_frame(DCAXllDecoder *s, uint8_t *data, int size, DCAExssAsset *asset)
+{
+    int ret;
+
+    if ((ret = init_get_bits8(&s->gb, data, size)) < 0)
+        return ret;
+    if ((ret = parse_common_header(s)) < 0)
+        return ret;
+    if ((ret = parse_sub_headers(s, asset)) < 0)
+        return ret;
+    if ((ret = parse_navi_table(s)) < 0)
+        return ret;
+    if ((ret = parse_band_data(s)) < 0)
+        return ret;
+    if (ff_dca_seek_bits(&s->gb, s->frame_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL frame\n");
+        return AVERROR_INVALIDDATA;
+    }
+    return ret;
+}
+
+static void clear_pbr(DCAXllDecoder *s)
+{
+    s->pbr_length = 0;
+    s->pbr_delay = 0;
+}
+
+static int copy_to_pbr(DCAXllDecoder *s, uint8_t *data, int size, int delay)
+{
+    if (size > DCA_XLL_PBR_BUFFER_MAX)
+        return AVERROR(ENOSPC);
+
+    if (!s->pbr_buffer && !(s->pbr_buffer = av_malloc(DCA_XLL_PBR_BUFFER_MAX + AV_INPUT_BUFFER_PADDING_SIZE)))
+        return AVERROR(ENOMEM);
+
+    memcpy(s->pbr_buffer, data, size);
+    s->pbr_length = size;
+    s->pbr_delay = delay;
+    return 0;
+}
+
+static int parse_frame_no_pbr(DCAXllDecoder *s, uint8_t *data, int size, DCAExssAsset *asset)
+{
+    int ret = parse_frame(s, data, size, asset);
+
+    // If XLL packet data didn't start with a sync word, we must have jumped
+    // right into the middle of PBR smoothing period
+    if (ret == AVERROR(EAGAIN) && asset->xll_sync_present && asset->xll_sync_offset < size) {
+        // Skip to the next sync word in this packet
+        data += asset->xll_sync_offset;
+        size -= asset->xll_sync_offset;
+
+        // If decoding delay is set, put the frame into PBR buffer and return
+        // failure code. Higher level decoder is expected to switch to lossy
+        // core decoding or mute its output until decoding delay expires.
+        if (asset->xll_delay_nframes > 0) {
+            if ((ret = copy_to_pbr(s, data, size, asset->xll_delay_nframes)) < 0)
+                return ret;
+            return AVERROR(EAGAIN);
+        }
+
+        // No decoding delay, just parse the frame in place
+        ret = parse_frame(s, data, size, asset);
+    }
+
+    if (ret < 0)
+        return ret;
+
+    if (s->frame_size > size)
+        return AVERROR(EINVAL);
+
+    // If the XLL decoder didn't consume full packet, start PBR smoothing period
+    if (s->frame_size < size)
+        if ((ret = copy_to_pbr(s, data + s->frame_size, size - s->frame_size, 0)) < 0)
+            return ret;
+
+    return 0;
+}
+
+static int parse_frame_pbr(DCAXllDecoder *s, uint8_t *data, int size, DCAExssAsset *asset)
+{
+    int ret;
+
+    if (size > DCA_XLL_PBR_BUFFER_MAX - s->pbr_length) {
+        ret = AVERROR(ENOSPC);
+        goto fail;
+    }
+
+    memcpy(s->pbr_buffer + s->pbr_length, data, size);
+    s->pbr_length += size;
+
+    // Respect decoding delay after synchronization error
+    if (s->pbr_delay > 0 && --s->pbr_delay)
+        return AVERROR(EAGAIN);
+
+    if ((ret = parse_frame(s, s->pbr_buffer, s->pbr_length, asset)) < 0)
+        goto fail;
+
+    if (s->frame_size > s->pbr_length) {
+        ret = AVERROR(EINVAL);
+        goto fail;
+    }
+
+    if (s->frame_size == s->pbr_length) {
+        // End of PBR smoothing period
+        clear_pbr(s);
+    } else {
+        s->pbr_length -= s->frame_size;
+        memmove(s->pbr_buffer, s->pbr_buffer + s->frame_size, s->pbr_length);
+    }
+
+    return 0;
+
+fail:
+    // For now, throw out all PBR state on failure.
+    // Perhaps we can be smarter and try to resync somehow.
+    clear_pbr(s);
+    return ret;
+}
+
+int ff_dca_xll_parse(DCAXllDecoder *s, uint8_t *data, DCAExssAsset *asset)
+{
+    int ret;
+
+    if (s->hd_stream_id != asset->hd_stream_id) {
+        clear_pbr(s);
+        s->hd_stream_id = asset->hd_stream_id;
+    }
+
+    if (s->pbr_length)
+        ret = parse_frame_pbr(s, data + asset->xll_offset, asset->xll_size, asset);
+    else
+        ret = parse_frame_no_pbr(s, data + asset->xll_offset, asset->xll_size, asset);
+
+    return ret;
+}
+
+static void undo_down_mix(DCAXllDecoder *s, DCAXllChSet *o, int band)
+{
+    int i, j, k, nchannels = 0, *coeff_ptr = o->dmix_coeff;
+    DCAXllChSet *c;
+
+    for (i = 0, c = s->chset; i < s->nactivechsets; i++, c++) {
+        if (!c->hier_chset)
+            continue;
+
+        av_assert1(band < c->nfreqbands);
+        for (j = 0; j < c->nchannels; j++) {
+            for (k = 0; k < o->nchannels; k++) {
+                int coeff = *coeff_ptr++;
+                if (coeff) {
+                    s->dcadsp->dmix_sub(c->bands[band].msb_sample_buffer[j],
+                                        o->bands[band].msb_sample_buffer[k],
+                                        coeff, s->nframesamples);
+                    if (band)
+                        s->dcadsp->dmix_sub(c->deci_history[j],
+                                            o->deci_history[k],
+                                            coeff, DCA_XLL_DECI_HISTORY_MAX);
                 }
             }
+        }
 
-next_chset:
-            in_channel += chset->channels;
-            /* Skip to next channel set using the NAVI info. */
-            i = get_bits_count(gb);
-            if (i > end_pos) {
-                av_log(s->avctx, AV_LOG_ERROR,
-                       "chset data exceeds NAVI size\n");
-                return AVERROR_INVALIDDATA;
+        nchannels += c->nchannels;
+        if (nchannels >= o->hier_ofs)
+            break;
+    }
+}
+
+static void scale_down_mix(DCAXllDecoder *s, DCAXllChSet *o, int band)
+{
+    int i, j, nchannels = 0;
+    DCAXllChSet *c;
+
+    for (i = 0, c = s->chset; i < s->nactivechsets; i++, c++) {
+        if (!c->hier_chset)
+            continue;
+
+        av_assert1(band < c->nfreqbands);
+        for (j = 0; j < c->nchannels; j++) {
+            int scale = o->dmix_scale[nchannels++];
+            if (scale != (1 << 15)) {
+                s->dcadsp->dmix_scale(c->bands[band].msb_sample_buffer[j],
+                                      scale, s->nframesamples);
+                if (band)
+                    s->dcadsp->dmix_scale(c->deci_history[j],
+                                          scale, DCA_XLL_DECI_HISTORY_MAX);
             }
-            if (i < end_pos)
-                skip_bits_long(gb, end_pos - i);
         }
+
+        if (nchannels >= o->hier_ofs)
+            break;
+    }
+}
+
+// Clear all band data and replace non-residual encoded channels with lossy
+// counterparts
+static av_cold void force_lossy_output(DCAXllDecoder *s, DCAXllChSet *c)
+{
+    DCAContext *dca = s->avctx->priv_data;
+    int band, ch;
+
+    for (band = 0; band < c->nfreqbands; band++)
+        chs_clear_band_data(s, c, band, -1);
+
+    for (ch = 0; ch < c->nchannels; ch++) {
+        if (!(c->residual_encode & (1 << ch)))
+            continue;
+        if (ff_dca_core_map_spkr(&dca->core, c->ch_remap[ch]) < 0)
+            continue;
+        c->residual_encode &= ~(1 << ch);
+    }
+}
+
+static int combine_residual_frame(DCAXllDecoder *s, DCAXllChSet *c)
+{
+    DCAContext *dca = s->avctx->priv_data;
+    int ch, nsamples = s->nframesamples;
+    DCAXllChSet *o;
+
+    // Verify that core is compatible
+    if (!(dca->packet & DCA_PACKET_CORE)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Residual encoded channels are present without core\n");
+        return AVERROR(EINVAL);
     }
+
+    if (c->freq != dca->core.output_rate) {
+        av_log(s->avctx, AV_LOG_WARNING, "Sample rate mismatch between core (%d Hz) and XLL (%d Hz)\n", dca->core.output_rate, c->freq);
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (nsamples != dca->core.npcmsamples) {
+        av_log(s->avctx, AV_LOG_WARNING, "Number of samples per frame mismatch between core (%d) and XLL (%d)\n", dca->core.npcmsamples, nsamples);
+        return AVERROR_INVALIDDATA;
+    }
+
+    // See if this channel set is downmixed and find the next channel set in
+    // hierarchy. If downmixed, undo core pre-scaling before combining with
+    // residual (residual is not scaled).
+    o = find_next_hier_dmix_chset(s, c);
+
+    // Reduce core bit width and combine with residual
+    for (ch = 0; ch < c->nchannels; ch++) {
+        int n, spkr, shift, round;
+        int32_t *src, *dst;
+
+        if (c->residual_encode & (1 << ch))
+            continue;
+
+        // Map this channel to core speaker
+        spkr = ff_dca_core_map_spkr(&dca->core, c->ch_remap[ch]);
+        if (spkr < 0) {
+            av_log(s->avctx, AV_LOG_WARNING, "Residual encoded channel (%d) references unavailable core channel\n", c->ch_remap[ch]);
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Account for LSB width
+        shift = 24 - c->pcm_bit_res + chs_get_lsb_width(s, c, 0, ch);
+        if (shift > 24) {
+            av_log(s->avctx, AV_LOG_WARNING, "Invalid core shift (%d bits)\n", shift);
+            return AVERROR_INVALIDDATA;
+        }
+
+        round = shift > 0 ? 1 << (shift - 1) : 0;
+
+        src = dca->core.output_samples[spkr];
+        dst = c->bands[0].msb_sample_buffer[ch];
+        if (o) {
+            // Undo embedded core downmix pre-scaling
+            int scale_inv = o->dmix_scale_inv[c->hier_ofs + ch];
+            for (n = 0; n < nsamples; n++)
+                dst[n] += (SUINT)clip23((mul16(src[n], scale_inv) + round) >> shift);
+        } else {
+            // No downmix scaling
+            for (n = 0; n < nsamples; n++)
+                dst[n] += (unsigned)((src[n] + round) >> shift);
+        }
+    }
+
+    return 0;
+}
+
+int ff_dca_xll_filter_frame(DCAXllDecoder *s, AVFrame *frame)
+{
+    AVCodecContext *avctx = s->avctx;
+    DCAContext *dca = avctx->priv_data;
+    DCAExssAsset *asset = &dca->exss.assets[0];
+    DCAXllChSet *p = &s->chset[0], *c;
+    enum AVMatrixEncoding matrix_encoding = AV_MATRIX_ENCODING_NONE;
+    int i, j, k, ret, shift, nsamples, request_mask;
+    int ch_remap[DCA_SPEAKER_COUNT];
+
+    // Force lossy downmixed output during recovery
+    if (dca->packet & DCA_PACKET_RECOVERY) {
+        for (i = 0, c = s->chset; i < s->nchsets; i++, c++) {
+            if (i < s->nactivechsets)
+                force_lossy_output(s, c);
+
+            if (!c->primary_chset)
+                c->dmix_embedded = 0;
+        }
+
+        s->scalable_lsbs = 0;
+        s->fixed_lsb_width = 0;
+    }
+
+    // Filter frequency bands for active channel sets
+    s->output_mask = 0;
+    for (i = 0, c = s->chset; i < s->nactivechsets; i++, c++) {
+        chs_filter_band_data(s, c, 0);
+
+        if (c->residual_encode != (1 << c->nchannels) - 1
+            && (ret = combine_residual_frame(s, c)) < 0)
+            return ret;
+
+        if (s->scalable_lsbs)
+            chs_assemble_msbs_lsbs(s, c, 0);
+
+        if (c->nfreqbands > 1) {
+            chs_filter_band_data(s, c, 1);
+            chs_assemble_msbs_lsbs(s, c, 1);
+        }
+
+        s->output_mask |= c->ch_mask;
+    }
+
+    // Undo hierarchial downmix and/or apply scaling
+    for (i = 1, c = &s->chset[1]; i < s->nchsets; i++, c++) {
+        if (!is_hier_dmix_chset(c))
+            continue;
+
+        if (i >= s->nactivechsets) {
+            for (j = 0; j < c->nfreqbands; j++)
+                if (c->bands[j].dmix_embedded)
+                    scale_down_mix(s, c, j);
+            break;
+        }
+
+        for (j = 0; j < c->nfreqbands; j++)
+            if (c->bands[j].dmix_embedded)
+                undo_down_mix(s, c, j);
+    }
+
+    // Assemble frequency bands for active channel sets
+    if (s->nfreqbands > 1) {
+        for (i = 0; i < s->nactivechsets; i++)
+            if ((ret = chs_assemble_freq_bands(s, &s->chset[i])) < 0)
+                return ret;
+    }
+
+    // Normalize to regular 5.1 layout if downmixing
+    if (dca->request_channel_layout) {
+        if (s->output_mask & DCA_SPEAKER_MASK_Lss) {
+            s->output_samples[DCA_SPEAKER_Ls] = s->output_samples[DCA_SPEAKER_Lss];
+            s->output_mask = (s->output_mask & ~DCA_SPEAKER_MASK_Lss) | DCA_SPEAKER_MASK_Ls;
+        }
+        if (s->output_mask & DCA_SPEAKER_MASK_Rss) {
+            s->output_samples[DCA_SPEAKER_Rs] = s->output_samples[DCA_SPEAKER_Rss];
+            s->output_mask = (s->output_mask & ~DCA_SPEAKER_MASK_Rss) | DCA_SPEAKER_MASK_Rs;
+        }
+    }
+
+    // Handle downmixing to stereo request
+    if (dca->request_channel_layout == DCA_SPEAKER_LAYOUT_STEREO
+        && DCA_HAS_STEREO(s->output_mask) && p->dmix_embedded
+        && (p->dmix_type == DCA_DMIX_TYPE_LoRo ||
+            p->dmix_type == DCA_DMIX_TYPE_LtRt))
+        request_mask = DCA_SPEAKER_LAYOUT_STEREO;
+    else
+        request_mask = s->output_mask;
+    if (!ff_dca_set_channel_layout(avctx, ch_remap, request_mask))
+        return AVERROR(EINVAL);
+
+    avctx->sample_rate = p->freq << (s->nfreqbands - 1);
+
+    switch (p->storage_bit_res) {
+    case 16:
+        avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
+        shift = 16 - p->pcm_bit_res;
+        break;
+    case 20:
+    case 24:
+        avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
+        shift = 24 - p->pcm_bit_res;
+        break;
+    default:
+        return AVERROR(EINVAL);
+    }
+
+    avctx->bits_per_raw_sample = p->storage_bit_res;
+    avctx->profile = FF_PROFILE_DTS_HD_MA;
+    avctx->bit_rate = 0;
+
+    frame->nb_samples = nsamples = s->nframesamples << (s->nfreqbands - 1);
+    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
+        return ret;
+
+    // Downmix primary channel set to stereo
+    if (request_mask != s->output_mask) {
+        ff_dca_downmix_to_stereo_fixed(s->dcadsp, s->output_samples,
+                                       p->dmix_coeff, nsamples,
+                                       s->output_mask);
+    }
+
+    for (i = 0; i < avctx->channels; i++) {
+        int32_t *samples = s->output_samples[ch_remap[i]];
+        if (frame->format == AV_SAMPLE_FMT_S16P) {
+            int16_t *plane = (int16_t *)frame->extended_data[i];
+            for (k = 0; k < nsamples; k++)
+                plane[k] = av_clip_int16(samples[k] * (SUINT)(1 << shift));
+        } else {
+            int32_t *plane = (int32_t *)frame->extended_data[i];
+            for (k = 0; k < nsamples; k++)
+                plane[k] = clip23(samples[k] * (SUINT)(1 << shift)) * (1 << 8);
+        }
+    }
+
+    if (!asset->one_to_one_map_ch_to_spkr) {
+        if (asset->representation_type == DCA_REPR_TYPE_LtRt)
+            matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
+        else if (asset->representation_type == DCA_REPR_TYPE_LhRh)
+            matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
+    } else if (request_mask != s->output_mask && p->dmix_type == DCA_DMIX_TYPE_LtRt) {
+        matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
+    }
+    if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
+        return ret;
+
     return 0;
 }
+
+av_cold void ff_dca_xll_flush(DCAXllDecoder *s)
+{
+    clear_pbr(s);
+}
+
+av_cold void ff_dca_xll_close(DCAXllDecoder *s)
+{
+    DCAXllChSet *c;
+    int i, j;
+
+    for (i = 0, c = s->chset; i < DCA_XLL_CHSETS_MAX; i++, c++) {
+        for (j = 0; j < DCA_XLL_SAMPLE_BUFFERS_MAX; j++) {
+            av_freep(&c->sample_buffer[j]);
+            c->sample_size[j] = 0;
+        }
+    }
+
+    av_freep(&s->navi);
+    s->navi_size = 0;
+
+    av_freep(&s->pbr_buffer);
+    clear_pbr(s);
+}