avcodec/dca: add new decoder based on libdcadec

Changelog

@@ -61,6 +61,7 @@ version <next>:
 - support for dvaudio in wav and avi
 - libaacplus and libvo-aacenc support removed
 - Cineform HD decoder
+- new DCA decoder with full support for DTS-HD extensions
 
 
 version 2.8:

configure

@@ -2271,6 +2271,7 @@ comfortnoise_encoder_select="lpc"
 cook_decoder_select="audiodsp mdct sinewin"
 cscd_decoder_select="lzo"
 cscd_decoder_suggest="zlib"
+dca_decoder_select="mdct"
 dds_decoder_select="texturedsp"
 dirac_decoder_select="dirac_parse dwt golomb videodsp mpegvideoenc"
 dnxhd_decoder_select="blockdsp idctdsp"

libavcodec/Makefile

@@ -222,6 +222,9 @@ OBJS-$(CONFIG_COMFORTNOISE_ENCODER)    += cngenc.o
 OBJS-$(CONFIG_CPIA_DECODER)            += cpia.o
 OBJS-$(CONFIG_CSCD_DECODER)            += cscd.o
 OBJS-$(CONFIG_CYUV_DECODER)            += cyuv.o
+OBJS-$(CONFIG_DCA_DECODER)             += dcadec.o dca.o dcadata.o        \
+                                          dca_core.o dca_exss.o dca_xll.o \
+                                          dcadsp.o dcadct.o synth_filter.o
 OBJS-$(CONFIG_DCA_ENCODER)             += dcaenc.o dca.o dcadata.o
 OBJS-$(CONFIG_DDS_DECODER)             += dds.o
 OBJS-$(CONFIG_DIRAC_DECODER)           += diracdec.o dirac.o diracdsp.o \

libavcodec/aarch64/Makefile

-#OBJS-$(CONFIG_DCA_DECODER)              += aarch64/synth_filter_init.o
+OBJS-$(CONFIG_DCA_DECODER)              += aarch64/synth_filter_init.o
 OBJS-$(CONFIG_FFT)                      += aarch64/fft_init_aarch64.o
 OBJS-$(CONFIG_FMTCONVERT)               += aarch64/fmtconvert_init.o
 OBJS-$(CONFIG_H264CHROMA)               += aarch64/h264chroma_init_aarch64.o
@@ -17,7 +17,7 @@ OBJS-$(CONFIG_VORBIS_DECODER)           += aarch64/vorbisdsp_init.o
 
 ARMV8-OBJS-$(CONFIG_VIDEODSP)           += aarch64/videodsp.o
 
-#NEON-OBJS-$(CONFIG_DCA_DECODER)         += aarch64/synth_filter_neon.o
+NEON-OBJS-$(CONFIG_DCA_DECODER)         += aarch64/synth_filter_neon.o
 NEON-OBJS-$(CONFIG_FFT)                 += aarch64/fft_neon.o
 NEON-OBJS-$(CONFIG_FMTCONVERT)          += aarch64/fmtconvert_neon.o
 NEON-OBJS-$(CONFIG_H264CHROMA)          += aarch64/h264cmc_neon.o

libavcodec/allcodecs.c

@@ -391,7 +391,7 @@ void avcodec_register_all(void)
     REGISTER_DECODER(BINKAUDIO_RDFT,    binkaudio_rdft);
     REGISTER_DECODER(BMV_AUDIO,         bmv_audio);
     REGISTER_DECODER(COOK,              cook);
-    REGISTER_ENCODER(DCA,               dca);
+    REGISTER_ENCDEC (DCA,               dca);
     REGISTER_DECODER(DSD_LSBF,          dsd_lsbf);
     REGISTER_DECODER(DSD_MSBF,          dsd_msbf);
     REGISTER_DECODER(DSD_LSBF_PLANAR,   dsd_lsbf_planar);

libavcodec/arm/Makefile

@@ -36,7 +36,7 @@ OBJS-$(CONFIG_VP8DSP)                  += arm/vp8dsp_init_arm.o
 # decoders/encoders
 OBJS-$(CONFIG_AAC_DECODER)             += arm/aacpsdsp_init_arm.o       \
                                           arm/sbrdsp_init_arm.o
-#OBJS-$(CONFIG_DCA_DECODER)             += arm/synth_filter_init_arm.o
+OBJS-$(CONFIG_DCA_DECODER)             += arm/synth_filter_init_arm.o
 OBJS-$(CONFIG_HEVC_DECODER)            += arm/hevcdsp_init_arm.o
 OBJS-$(CONFIG_MLP_DECODER)             += arm/mlpdsp_init_arm.o
 OBJS-$(CONFIG_RV40_DECODER)            += arm/rv40dsp_init_arm.o
@@ -87,7 +87,7 @@ VFP-OBJS-$(CONFIG_FMTCONVERT)          += arm/fmtconvert_vfp.o
 VFP-OBJS-$(CONFIG_MDCT)                += arm/mdct_vfp.o
 
 # decoders/encoders
-#VFP-OBJS-$(CONFIG_DCA_DECODER)         += arm/synth_filter_vfp.o
+VFP-OBJS-$(CONFIG_DCA_DECODER)         += arm/synth_filter_vfp.o
 
 
 # NEON optimizations
@@ -126,7 +126,7 @@ NEON-OBJS-$(CONFIG_VP8DSP)             += arm/vp8dsp_init_neon.o        \
 NEON-OBJS-$(CONFIG_AAC_DECODER)        += arm/aacpsdsp_neon.o           \
                                           arm/sbrdsp_neon.o
 NEON-OBJS-$(CONFIG_LLAUDDSP)           += arm/lossless_audiodsp_neon.o
-#NEON-OBJS-$(CONFIG_DCA_DECODER)        += arm/synth_filter_neon.o
+NEON-OBJS-$(CONFIG_DCA_DECODER)        += arm/synth_filter_neon.o
 NEON-OBJS-$(CONFIG_HEVC_DECODER)       += arm/hevcdsp_init_neon.o       \
                                           arm/hevcdsp_deblock_neon.o    \
                                           arm/hevcdsp_idct_neon.o       \

libavcodec/dca_core.c

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "dcadec.h"
+#include "dcadata.h"
+#include "dcahuff.h"
+#include "dcamath.h"
+#include "dca_syncwords.h"
+
+#if ARCH_ARM
+#include "arm/dca.h"
+#endif
+
+enum HeaderType {
+    HEADER_CORE,
+    HEADER_XCH,
+    HEADER_XXCH
+};
+
+enum AudioMode {
+    AMODE_MONO,             // Mode 0: A (mono)
+    AMODE_MONO_DUAL,        // Mode 1: A + B (dual mono)
+    AMODE_STEREO,           // Mode 2: L + R (stereo)
+    AMODE_STEREO_SUMDIFF,   // Mode 3: (L+R) + (L-R) (sum-diff)
+    AMODE_STEREO_TOTAL,     // Mode 4: LT + RT (left and right total)
+    AMODE_3F,               // Mode 5: C + L + R
+    AMODE_2F1R,             // Mode 6: L + R + S
+    AMODE_3F1R,             // Mode 7: C + L + R + S
+    AMODE_2F2R,             // Mode 8: L + R + SL + SR
+    AMODE_3F2R,             // Mode 9: C + L + R + SL + SR
+
+    AMODE_COUNT
+};
+
+enum ExtAudioType {
+    EXT_AUDIO_XCH   = 0,
+    EXT_AUDIO_X96   = 2,
+    EXT_AUDIO_XXCH  = 6
+};
+
+enum LFEFlag {
+    LFE_FLAG_NONE,
+    LFE_FLAG_128,
+    LFE_FLAG_64,
+    LFE_FLAG_INVALID
+};
+
+static const int8_t prm_ch_to_spkr_map[AMODE_COUNT][5] = {
+    { DCA_SPEAKER_C,            -1,             -1,             -1,             -1 },
+    { DCA_SPEAKER_L, DCA_SPEAKER_R,             -1,             -1,             -1 },
+    { DCA_SPEAKER_L, DCA_SPEAKER_R,             -1,             -1,             -1 },
+    { DCA_SPEAKER_L, DCA_SPEAKER_R,             -1,             -1,             -1 },
+    { DCA_SPEAKER_L, DCA_SPEAKER_R,             -1,             -1,             -1 },
+    { DCA_SPEAKER_C, DCA_SPEAKER_L, DCA_SPEAKER_R ,             -1,             -1 },
+    { DCA_SPEAKER_L, DCA_SPEAKER_R, DCA_SPEAKER_Cs,             -1,             -1 },
+    { DCA_SPEAKER_C, DCA_SPEAKER_L, DCA_SPEAKER_R , DCA_SPEAKER_Cs,             -1 },
+    { DCA_SPEAKER_L, DCA_SPEAKER_R, DCA_SPEAKER_Ls, DCA_SPEAKER_Rs,             -1 },
+    { DCA_SPEAKER_C, DCA_SPEAKER_L, DCA_SPEAKER_R,  DCA_SPEAKER_Ls, DCA_SPEAKER_Rs }
+};
+
+static const uint8_t audio_mode_ch_mask[AMODE_COUNT] = {
+    DCA_SPEAKER_LAYOUT_MONO,
+    DCA_SPEAKER_LAYOUT_STEREO,
+    DCA_SPEAKER_LAYOUT_STEREO,
+    DCA_SPEAKER_LAYOUT_STEREO,
+    DCA_SPEAKER_LAYOUT_STEREO,
+    DCA_SPEAKER_LAYOUT_3_0,
+    DCA_SPEAKER_LAYOUT_2_1,
+    DCA_SPEAKER_LAYOUT_3_1,
+    DCA_SPEAKER_LAYOUT_2_2,
+    DCA_SPEAKER_LAYOUT_5POINT0
+};
+
+static const uint8_t block_code_nbits[7] = {
+    7, 10, 12, 13, 15, 17, 19
+};
+
+static const uint8_t quant_index_sel_nbits[DCA_CODE_BOOKS] = {
+    1, 2, 2, 2, 2, 3, 3, 3, 3, 3
+};
+
+static const uint8_t quant_index_group_size[DCA_CODE_BOOKS] = {
+    1, 3, 3, 3, 3, 7, 7, 7, 7, 7
+};
+
+typedef struct DCAVLC {
+    int offset;         ///< Code values offset
+    int max_depth;      ///< Parameter for get_vlc2()
+    VLC vlc[7];         ///< Actual codes
+} DCAVLC;
+
+static DCAVLC   vlc_bit_allocation;
+static DCAVLC   vlc_transition_mode;
+static DCAVLC   vlc_scale_factor;
+static DCAVLC   vlc_quant_index[DCA_CODE_BOOKS];
+
+static av_cold void dca_init_vlcs(void)
+{
+    static VLC_TYPE dca_table[23622][2];
+    static int vlcs_initialized = 0;
+    int i, j, k;
+
+    if (vlcs_initialized)
+        return;
+
+#define DCA_INIT_VLC(vlc, a, b, c, d)                                      \
+    do {                                                                   \
+        vlc.table           = &dca_table[ff_dca_vlc_offs[k]];              \
+        vlc.table_allocated = ff_dca_vlc_offs[k + 1] - ff_dca_vlc_offs[k]; \
+        init_vlc(&vlc, a, b, c, 1, 1, d, 2, 2, INIT_VLC_USE_NEW_STATIC);   \
+    } while (0)
+
+    vlc_bit_allocation.offset    = 1;
+    vlc_bit_allocation.max_depth = 2;
+    for (i = 0, k = 0; i < 5; i++, k++)
+        DCA_INIT_VLC(vlc_bit_allocation.vlc[i], bitalloc_12_vlc_bits[i], 12,
+                     bitalloc_12_bits[i], bitalloc_12_codes[i]);
+
+    vlc_scale_factor.offset    = -64;
+    vlc_scale_factor.max_depth = 2;
+    for (i = 0; i < 5; i++, k++)
+        DCA_INIT_VLC(vlc_scale_factor.vlc[i], SCALES_VLC_BITS, 129,
+                     scales_bits[i], scales_codes[i]);
+
+    vlc_transition_mode.offset    = 0;
+    vlc_transition_mode.max_depth = 1;
+    for (i = 0; i < 4; i++, k++)
+        DCA_INIT_VLC(vlc_transition_mode.vlc[i], tmode_vlc_bits[i], 4,
+                     tmode_bits[i], tmode_codes[i]);
+
+    for (i = 0; i < DCA_CODE_BOOKS; i++) {
+        vlc_quant_index[i].offset    = bitalloc_offsets[i];
+        vlc_quant_index[i].max_depth = 1 + (i > 4);
+        for (j = 0; j < quant_index_group_size[i]; j++, k++)
+            DCA_INIT_VLC(vlc_quant_index[i].vlc[j], bitalloc_maxbits[i][j],
+                         bitalloc_sizes[i], bitalloc_bits[i][j], bitalloc_codes[i][j]);
+    }
+
+    vlcs_initialized = 1;
+}
+
+static int get_vlc(GetBitContext *s, DCAVLC *v, int i)
+{
+    return get_vlc2(s, v->vlc[i].table, v->vlc[i].bits, v->max_depth) + v->offset;
+}
+
+static void get_array(GetBitContext *s, int32_t *array, int size, int n)
+{
+    int i;
+
+    for (i = 0; i < size; i++)
+        array[i] = get_sbits(s, n);
+}
+
+// 5.3.1 - Bit stream header
+static int parse_frame_header(DCACoreDecoder *s)
+{
+    int normal_frame, pcmr_index;
+
+    // Frame type
+    normal_frame = get_bits1(&s->gb);
+
+    // Deficit sample count
+    if (get_bits(&s->gb, 5) != DCA_PCMBLOCK_SAMPLES - 1) {
+        av_log(s->avctx, AV_LOG_ERROR, "Deficit samples are not supported\n");
+        return normal_frame ? AVERROR_INVALIDDATA : AVERROR_PATCHWELCOME;
+    }
+
+    // CRC present flag
+    s->crc_present = get_bits1(&s->gb);
+
+    // Number of PCM sample blocks
+    s->npcmblocks = get_bits(&s->gb, 7) + 1;
+    if (s->npcmblocks & (DCA_SUBBAND_SAMPLES - 1)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Unsupported number of PCM sample blocks (%d)\n", s->npcmblocks);
+        return (s->npcmblocks < 6 || normal_frame) ? AVERROR_INVALIDDATA : AVERROR_PATCHWELCOME;
+    }
+
+    // Primary frame byte size
+    s->frame_size = get_bits(&s->gb, 14) + 1;
+    if (s->frame_size < 96) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid core frame size (%d bytes)\n", s->frame_size);
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Audio channel arrangement
+    s->audio_mode = get_bits(&s->gb, 6);
+    if (s->audio_mode >= AMODE_COUNT) {
+        av_log(s->avctx, AV_LOG_ERROR, "Unsupported audio channel arrangement (%d)\n", s->audio_mode);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Core audio sampling frequency
+    s->sample_rate = avpriv_dca_sample_rates[get_bits(&s->gb, 4)];
+    if (!s->sample_rate) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid core audio sampling frequency\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Transmission bit rate
+    s->bit_rate = ff_dca_bit_rates[get_bits(&s->gb, 5)];
+
+    // Reserved field
+    skip_bits1(&s->gb);
+
+    // Embedded dynamic range flag
+    s->drc_present = get_bits1(&s->gb);
+
+    // Embedded time stamp flag
+    s->ts_present = get_bits1(&s->gb);
+
+    // Auxiliary data flag
+    s->aux_present = get_bits1(&s->gb);
+
+    // HDCD mastering flag
+    skip_bits1(&s->gb);
+
+    // Extension audio descriptor flag
+    s->ext_audio_type = get_bits(&s->gb, 3);
+
+    // Extended coding flag
+    s->ext_audio_present = get_bits1(&s->gb);
+
+    // Audio sync word insertion flag
+    s->sync_ssf = get_bits1(&s->gb);
+
+    // Low frequency effects flag
+    s->lfe_present = get_bits(&s->gb, 2);
+    if (s->lfe_present == LFE_FLAG_INVALID) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid low frequency effects flag\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Predictor history flag switch
+    s->predictor_history = get_bits1(&s->gb);
+
+    // Header CRC check bytes
+    if (s->crc_present)
+        skip_bits(&s->gb, 16);
+
+    // Multirate interpolator switch
+    s->filter_perfect = get_bits1(&s->gb);
+
+    // Encoder software revision
+    skip_bits(&s->gb, 4);
+
+    // Copy history
+    skip_bits(&s->gb, 2);
+
+    // Source PCM resolution
+    s->source_pcm_res = ff_dca_bits_per_sample[pcmr_index = get_bits(&s->gb, 3)];
+    if (!s->source_pcm_res) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid source PCM resolution\n");
+        return AVERROR_INVALIDDATA;
+    }
+    s->es_format = pcmr_index & 1;
+
+    // Front sum/difference flag
+    s->sumdiff_front = get_bits1(&s->gb);
+
+    // Surround sum/difference flag
+    s->sumdiff_surround = get_bits1(&s->gb);
+
+    // Dialog normalization / unspecified
+    skip_bits(&s->gb, 4);
+
+    return 0;
+}
+
+// 5.3.2 - Primary audio coding header
+static int parse_coding_header(DCACoreDecoder *s, enum HeaderType header, int xch_base)
+{
+    int n, ch, nchannels, header_size = 0, header_pos = get_bits_count(&s->gb);
+    unsigned int mask, index;
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    switch (header) {
+    case HEADER_CORE:
+        // Number of subframes
+        s->nsubframes = get_bits(&s->gb, 4) + 1;
+
+        // Number of primary audio channels
+        s->nchannels = get_bits(&s->gb, 3) + 1;
+        if (s->nchannels != ff_dca_channels[s->audio_mode]) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid number of primary audio channels (%d) for audio channel arrangement (%d)\n", s->nchannels, s->audio_mode);
+            return AVERROR_INVALIDDATA;
+        }
+        av_assert1(s->nchannels <= DCA_CHANNELS - 2);
+
+        s->ch_mask = audio_mode_ch_mask[s->audio_mode];
+
+        // Add LFE channel if present
+        if (s->lfe_present)
+            s->ch_mask |= DCA_SPEAKER_MASK_LFE1;
+        break;
+
+    case HEADER_XCH:
+        s->nchannels = ff_dca_channels[s->audio_mode] + 1;
+        av_assert1(s->nchannels <= DCA_CHANNELS - 1);
+        s->ch_mask |= DCA_SPEAKER_MASK_Cs;
+        break;
+
+    case HEADER_XXCH:
+        // Channel set header length
+        header_size = get_bits(&s->gb, 7) + 1;
+
+        // Check CRC
+        if (s->xxch_crc_present
+            && (s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+            && ff_dca_check_crc(&s->gb, header_pos, header_pos + header_size * 8)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid XXCH channel set header checksum\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Number of channels in a channel set
+        nchannels = get_bits(&s->gb, 3) + 1;
+        if (nchannels > DCA_XXCH_CHANNELS_MAX) {
+            avpriv_request_sample(s->avctx, "%d XXCH channels", nchannels);
+            return AVERROR_PATCHWELCOME;
+        }
+        s->nchannels = ff_dca_channels[s->audio_mode] + nchannels;
+        av_assert1(s->nchannels <= DCA_CHANNELS);
+
+        // Loudspeaker layout mask
+        mask = get_bits_long(&s->gb, s->xxch_mask_nbits - DCA_SPEAKER_Cs);
+        s->xxch_spkr_mask = mask << DCA_SPEAKER_Cs;
+
+        if (av_popcount(s->xxch_spkr_mask) != nchannels) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid XXCH speaker layout mask (%#x)\n", s->xxch_spkr_mask);
+            return AVERROR_INVALIDDATA;
+        }
+
+        if (s->xxch_core_mask & s->xxch_spkr_mask) {
+            av_log(s->avctx, AV_LOG_ERROR, "XXCH speaker layout mask (%#x) overlaps with core (%#x)\n", s->xxch_spkr_mask, s->xxch_core_mask);
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Combine core and XXCH masks together
+        s->ch_mask = s->xxch_core_mask | s->xxch_spkr_mask;
+
+        // Downmix coefficients present in stream
+        if (get_bits1(&s->gb)) {
+            int *coeff_ptr = s->xxch_dmix_coeff;
+
+            // Downmix already performed by encoder
+            s->xxch_dmix_embedded = get_bits1(&s->gb);
+
+            // Downmix scale factor
+            index = get_bits(&s->gb, 6) * 4 - FF_DCA_DMIXTABLE_OFFSET - 3;
+            if (index >= FF_DCA_INV_DMIXTABLE_SIZE) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid XXCH downmix scale index (%d)\n", index);
+                return AVERROR_INVALIDDATA;
+            }
+            s->xxch_dmix_scale_inv = ff_dca_inv_dmixtable[index];
+
+            // Downmix channel mapping mask
+            for (ch = 0; ch < nchannels; ch++) {
+                mask = get_bits_long(&s->gb, s->xxch_mask_nbits);
+                if ((mask & s->xxch_core_mask) != mask) {
+                    av_log(s->avctx, AV_LOG_ERROR, "Invalid XXCH downmix channel mapping mask (%#x)\n", mask);
+                    return AVERROR_INVALIDDATA;
+                }
+                s->xxch_dmix_mask[ch] = mask;
+            }
+
+            // Downmix coefficients
+            for (ch = 0; ch < nchannels; ch++) {
+                for (n = 0; n < s->xxch_mask_nbits; n++) {
+                    if (s->xxch_dmix_mask[ch] & (1U << n)) {
+                        int code = get_bits(&s->gb, 7);
+                        int sign = (code >> 6) - 1;
+                        if (code &= 63) {
+                            index = code * 4 - 3;
+                            if (index >= FF_DCA_DMIXTABLE_SIZE) {
+                                av_log(s->avctx, AV_LOG_ERROR, "Invalid XXCH downmix coefficient index (%d)\n", index);
+                                return AVERROR_INVALIDDATA;
+                            }
+                            *coeff_ptr++ = (ff_dca_dmixtable[index] ^ sign) - sign;
+                        } else {
+                            *coeff_ptr++ = 0;
+                        }
+                    }
+                }
+            }
+        } else {
+            s->xxch_dmix_embedded = 0;
+        }
+
+        break;
+    }
+
+    // Subband activity count
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        s->nsubbands[ch] = get_bits(&s->gb, 5) + 2;
+        if (s->nsubbands[ch] > DCA_SUBBANDS) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid subband activity count\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // High frequency VQ start subband
+    for (ch = xch_base; ch < s->nchannels; ch++)
+        s->subband_vq_start[ch] = get_bits(&s->gb, 5) + 1;
+
+    // Joint intensity coding index
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        if ((n = get_bits(&s->gb, 3)) && header == HEADER_XXCH)
+            n += xch_base - 1;
+        if (n > s->nchannels) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid joint intensity coding index\n");
+            return AVERROR_INVALIDDATA;
+        }
+        s->joint_intensity_index[ch] = n;
+    }
+
+    // Transient mode code book
+    for (ch = xch_base; ch < s->nchannels; ch++)
+        s->transition_mode_sel[ch] = get_bits(&s->gb, 2);
+
+    // Scale factor code book
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        s->scale_factor_sel[ch] = get_bits(&s->gb, 3);
+        if (s->scale_factor_sel[ch] == 7) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid scale factor code book\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // Bit allocation quantizer select
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        s->bit_allocation_sel[ch] = get_bits(&s->gb, 3);
+        if (s->bit_allocation_sel[ch] == 7) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid bit allocation quantizer select\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // Quantization index codebook select
+    for (n = 0; n < DCA_CODE_BOOKS; n++)
+        for (ch = xch_base; ch < s->nchannels; ch++)
+            s->quant_index_sel[ch][n] = get_bits(&s->gb, quant_index_sel_nbits[n]);
+
+    // Scale factor adjustment index
+    for (n = 0; n < DCA_CODE_BOOKS; n++)
+        for (ch = xch_base; ch < s->nchannels; ch++)
+            if (s->quant_index_sel[ch][n] < quant_index_group_size[n])
+                s->scale_factor_adj[ch][n] = ff_dca_scale_factor_adj[get_bits(&s->gb, 2)];
+
+    if (header == HEADER_XXCH) {
+        // Reserved
+        // Byte align
+        // CRC16 of channel set header
+        if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Read past end of XXCH channel set header\n");
+            return AVERROR_INVALIDDATA;
+        }
+    } else {
+        // Audio header CRC check word
+        if (s->crc_present)
+            skip_bits(&s->gb, 16);
+    }
+
+    return 0;
+}
+
+static inline int parse_scale(DCACoreDecoder *s, int *scale_index, int sel)
+{
+    const uint32_t *scale_table;
+    unsigned int scale_size;
+
+    // Select the root square table
+    if (sel > 5) {
+        scale_table = ff_dca_scale_factor_quant7;
+        scale_size = FF_ARRAY_ELEMS(ff_dca_scale_factor_quant7);
+    } else {
+        scale_table = ff_dca_scale_factor_quant6;
+        scale_size = FF_ARRAY_ELEMS(ff_dca_scale_factor_quant6);
+    }
+
+    // If Huffman code was used, the difference of scales was encoded
+    if (sel < 5)
+        *scale_index += get_vlc(&s->gb, &vlc_scale_factor, sel);
+    else
+        *scale_index = get_bits(&s->gb, sel + 1);
+
+    // Look up scale factor from the root square table
+    if ((unsigned int)*scale_index >= scale_size) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid scale factor index\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return scale_table[*scale_index];
+}
+
+static inline int parse_joint_scale(DCACoreDecoder *s, int sel)
+{
+    int scale_index;
+
+    // Absolute value was encoded even when Huffman code was used
+    if (sel < 5)
+        scale_index = get_vlc(&s->gb, &vlc_scale_factor, sel);
+    else
+        scale_index = get_bits(&s->gb, sel + 1);
+
+    // Bias by 64
+    scale_index += 64;
+
+    // Look up joint scale factor
+    if ((unsigned int)scale_index >= FF_ARRAY_ELEMS(ff_dca_joint_scale_factors)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid joint scale factor index\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return ff_dca_joint_scale_factors[scale_index];
+}
+
+// 5.4.1 - Primary audio coding side information
+static int parse_subframe_header(DCACoreDecoder *s, int sf,
+                                 enum HeaderType header, int xch_base)
+{
+    int ch, band, ret;
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    if (header == HEADER_CORE) {
+        // Subsubframe count
+        s->nsubsubframes[sf] = get_bits(&s->gb, 2) + 1;
+
+        // Partial subsubframe sample count
+        skip_bits(&s->gb, 3);
+    }
+
+    // Prediction mode
+    for (ch = xch_base; ch < s->nchannels; ch++)
+        for (band = 0; band < s->nsubbands[ch]; band++)
+            s->prediction_mode[ch][band] = get_bits1(&s->gb);
+
+    // Prediction coefficients VQ address
+    for (ch = xch_base; ch < s->nchannels; ch++)
+        for (band = 0; band < s->nsubbands[ch]; band++)
+            if (s->prediction_mode[ch][band])
+                s->prediction_vq_index[ch][band] = get_bits(&s->gb, 12);
+
+    // Bit allocation index
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        int sel = s->bit_allocation_sel[ch];
+
+        for (band = 0; band < s->subband_vq_start[ch]; band++) {
+            int abits;
+
+            if (sel < 5)
+                abits = get_vlc(&s->gb, &vlc_bit_allocation, sel);
+            else
+                abits = get_bits(&s->gb, sel - 1);
+
+            if (abits > DCA_ABITS_MAX) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid bit allocation index\n");
+                return AVERROR_INVALIDDATA;
+            }
+
+            s->bit_allocation[ch][band] = abits;
+        }
+    }
+
+    // Transition mode
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        // Clear transition mode for all subbands
+        memset(s->transition_mode[sf][ch], 0, sizeof(s->transition_mode[0][0]));
+
+        // Transient possible only if more than one subsubframe
+        if (s->nsubsubframes[sf] > 1) {
+            int sel = s->transition_mode_sel[ch];
+            for (band = 0; band < s->subband_vq_start[ch]; band++)
+                if (s->bit_allocation[ch][band])
+                    s->transition_mode[sf][ch][band] = get_vlc(&s->gb, &vlc_transition_mode, sel);
+        }
+    }
+
+    // Scale factors
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        int sel = s->scale_factor_sel[ch];
+        int scale_index = 0;
+
+        // Extract scales for subbands up to VQ
+        for (band = 0; band < s->subband_vq_start[ch]; band++) {
+            if (s->bit_allocation[ch][band]) {
+                if ((ret = parse_scale(s, &scale_index, sel)) < 0)
+                    return ret;
+                s->scale_factors[ch][band][0] = ret;
+                if (s->transition_mode[sf][ch][band]) {
+                    if ((ret = parse_scale(s, &scale_index, sel)) < 0)
+                        return ret;
+                    s->scale_factors[ch][band][1] = ret;
+                }
+            } else {
+                s->scale_factors[ch][band][0] = 0;
+            }
+        }
+
+        // High frequency VQ subbands
+        for (band = s->subband_vq_start[ch]; band < s->nsubbands[ch]; band++) {
+            if ((ret = parse_scale(s, &scale_index, sel)) < 0)
+                return ret;
+            s->scale_factors[ch][band][0] = ret;
+        }
+    }
+
+    // Joint subband codebook select
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        if (s->joint_intensity_index[ch]) {
+            s->joint_scale_sel[ch] = get_bits(&s->gb, 3);
+            if (s->joint_scale_sel[ch] == 7) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid joint scale factor code book\n");
+                return AVERROR_INVALIDDATA;
+            }
+        }
+    }
+
+    // Scale factors for joint subband coding
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        int src_ch = s->joint_intensity_index[ch] - 1;
+        if (src_ch >= 0) {
+            int sel = s->joint_scale_sel[ch];
+            for (band = s->nsubbands[ch]; band < s->nsubbands[src_ch]; band++) {
+                if ((ret = parse_joint_scale(s, sel)) < 0)
+                    return ret;
+                s->joint_scale_factors[ch][band] = ret;
+            }
+        }
+    }
+
+    // Dynamic range coefficient
+    if (s->drc_present && header == HEADER_CORE)
+        skip_bits(&s->gb, 8);
+
+    // Side information CRC check word
+    if (s->crc_present)
+        skip_bits(&s->gb, 16);
+
+    return 0;
+}
+
+#ifndef decode_blockcodes
+static inline int decode_blockcodes(int code1, int code2, int levels, int32_t *audio)
+{
+    int offset = (levels - 1) / 2;
+    int n, div;
+
+    for (n = 0; n < DCA_SUBBAND_SAMPLES / 2; n++) {
+        div = FASTDIV(code1, levels);
+        audio[n] = code1 - div * levels - offset;
+        code1 = div;
+    }
+    for (; n < DCA_SUBBAND_SAMPLES; n++) {
+        div = FASTDIV(code2, levels);
+        audio[n] = code2 - div * levels - offset;
+        code2 = div;
+    }
+
+    return code1 | code2;
+}
+#endif
+
+static inline int parse_block_codes(DCACoreDecoder *s, int32_t *audio, int abits)
+{
+    // Extract block code indices from the bit stream
+    int code1 = get_bits(&s->gb, block_code_nbits[abits - 1]);
+    int code2 = get_bits(&s->gb, block_code_nbits[abits - 1]);
+    int levels = ff_dca_quant_levels[abits];
+
+    // Look up samples from the block code book
+    if (decode_blockcodes(code1, code2, levels, audio)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Failed to decode block code(s)\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static inline int parse_huffman_codes(DCACoreDecoder *s, int32_t *audio, int abits, int sel)
+{
+    int i;
+
+    // Extract Huffman codes from the bit stream
+    for (i = 0; i < DCA_SUBBAND_SAMPLES; i++)
+        audio[i] = get_vlc(&s->gb, &vlc_quant_index[abits - 1], sel);
+
+    return 1;
+}
+
+static inline int extract_audio(DCACoreDecoder *s, int32_t *audio, int abits, int ch)
+{
+    av_assert1(abits >= 0 && abits <= DCA_ABITS_MAX);
+
+    if (abits == 0) {
+        // No bits allocated
+        memset(audio, 0, DCA_SUBBAND_SAMPLES * sizeof(*audio));
+        return 0;
+    }
+
+    if (abits <= DCA_CODE_BOOKS) {
+        int sel = s->quant_index_sel[ch][abits - 1];
+        if (sel < quant_index_group_size[abits - 1]) {
+            // Huffman codes
+            return parse_huffman_codes(s, audio, abits, sel);
+        }
+        if (abits <= 7) {
+            // Block codes
+            return parse_block_codes(s, audio, abits);
+        }
+    }
+
+    // No further encoding
+    get_array(&s->gb, audio, DCA_SUBBAND_SAMPLES, abits - 3);
+    return 0;
+}
+
+static inline void dequantize(int32_t *output, const int32_t *input,
+                              int32_t step_size, int32_t scale, int residual)
+{
+    // Account for quantizer step size
+    int64_t step_scale = (int64_t)step_size * scale;
+    int n, shift = 0;
+
+    // Limit scale factor resolution to 22 bits
+    if (step_scale > (1 << 23)) {
+        shift = av_log2(step_scale >> 23) + 1;
+        step_scale >>= shift;
+    }
+
+    // Scale the samples
+    if (residual) {
+        for (n = 0; n < DCA_SUBBAND_SAMPLES; n++)
+            output[n] += clip23(norm__(input[n] * step_scale, 22 - shift));
+    } else {
+        for (n = 0; n < DCA_SUBBAND_SAMPLES; n++)
+            output[n]  = clip23(norm__(input[n] * step_scale, 22 - shift));
+    }
+}
+
+static inline void inverse_adpcm(int32_t **subband_samples,
+                                 const int16_t *vq_index,
+                                 const int8_t *prediction_mode,
+                                 int sb_start, int sb_end,
+                                 int ofs, int len)
+{
+    int i, j, k;
+
+    for (i = sb_start; i < sb_end; i++) {
+        if (prediction_mode[i]) {
+            const int16_t *coeff = ff_dca_adpcm_vb[vq_index[i]];
+            int32_t *ptr = subband_samples[i] + ofs;
+            for (j = 0; j < len; j++) {
+                int64_t err = 0;
+                for (k = 0; k < DCA_ADPCM_COEFFS; k++)
+                    err += (int64_t)ptr[j - k - 1] * coeff[k];
+                ptr[j] = clip23(ptr[j] + clip23(norm13(err)));
+            }
+        }
+    }
+}
+
+// 5.5 - Primary audio data arrays
+static int parse_subframe_audio(DCACoreDecoder *s, int sf, enum HeaderType header,
+                                int xch_base, int *sub_pos, int *lfe_pos)
+{
+    int32_t audio[16], scale;
+    int n, ssf, ofs, ch, band;
+
+    // Check number of subband samples in this subframe
+    int nsamples = s->nsubsubframes[sf] * DCA_SUBBAND_SAMPLES;
+    if (*sub_pos + nsamples > s->npcmblocks) {
+        av_log(s->avctx, AV_LOG_ERROR, "Subband sample buffer overflow\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    // VQ encoded subbands
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        int32_t vq_index[DCA_SUBBANDS];
+
+        for (band = s->subband_vq_start[ch]; band < s->nsubbands[ch]; band++)
+            // Extract the VQ address from the bit stream
+            vq_index[band] = get_bits(&s->gb, 10);
+
+        if (s->subband_vq_start[ch] < s->nsubbands[ch]) {
+            s->dcadsp->decode_hf(s->subband_samples[ch], vq_index,
+                                 ff_dca_high_freq_vq, s->scale_factors[ch],
+                                 s->subband_vq_start[ch], s->nsubbands[ch],
+                                 *sub_pos, nsamples);
+        }
+    }
+
+    // Low frequency effect data
+    if (s->lfe_present && header == HEADER_CORE) {
+        unsigned int index;
+
+        // Determine number of LFE samples in this subframe
+        int nlfesamples = 2 * s->lfe_present * s->nsubsubframes[sf];
+        av_assert1((unsigned int)nlfesamples <= FF_ARRAY_ELEMS(audio));
+
+        // Extract LFE samples from the bit stream
+        get_array(&s->gb, audio, nlfesamples, 8);
+
+        // Extract scale factor index from the bit stream
+        index = get_bits(&s->gb, 8);
+        if (index >= FF_ARRAY_ELEMS(ff_dca_scale_factor_quant7)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid LFE scale factor index\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Look up the 7-bit root square quantization table
+        scale = ff_dca_scale_factor_quant7[index];
+
+        // Account for quantizer step size which is 0.035
+        scale = mul23(4697620 /* 0.035 * (1 << 27) */, scale);
+
+        // Scale and take the LFE samples
+        for (n = 0, ofs = *lfe_pos; n < nlfesamples; n++, ofs++)
+            s->lfe_samples[ofs] = clip23(audio[n] * scale >> 4);
+
+        // Advance LFE sample pointer for the next subframe
+        *lfe_pos = ofs;
+    }
+
+    // Audio data
+    for (ssf = 0, ofs = *sub_pos; ssf < s->nsubsubframes[sf]; ssf++) {
+        for (ch = xch_base; ch < s->nchannels; ch++) {
+            if (get_bits_left(&s->gb) < 0)
+                return AVERROR_INVALIDDATA;
+
+            // Not high frequency VQ subbands
+            for (band = 0; band < s->subband_vq_start[ch]; band++) {
+                int ret, trans_ssf, abits = s->bit_allocation[ch][band];
+                int32_t step_size;
+
+                // Extract bits from the bit stream
+                if ((ret = extract_audio(s, audio, abits, ch)) < 0)
+                    return ret;
+
+                // Select quantization step size table and look up
+                // quantization step size
+                if (s->bit_rate == 3)
+                    step_size = ff_dca_lossless_quant[abits];
+                else
+                    step_size = ff_dca_lossy_quant[abits];
+
+                // Identify transient location
+                trans_ssf = s->transition_mode[sf][ch][band];
+
+                // Determine proper scale factor
+                if (trans_ssf == 0 || ssf < trans_ssf)
+                    scale = s->scale_factors[ch][band][0];
+                else
+                    scale = s->scale_factors[ch][band][1];
+
+                // Adjust scale factor when SEL indicates Huffman code
+                if (ret > 0) {
+                    int64_t adj = s->scale_factor_adj[ch][abits - 1];
+                    scale = clip23(adj * scale >> 22);
+                }
+
+                dequantize(s->subband_samples[ch][band] + ofs,
+                           audio, step_size, scale, 0);
+            }
+        }
+
+        // DSYNC
+        if ((ssf == s->nsubsubframes[sf] - 1 || s->sync_ssf) && get_bits(&s->gb, 16) != 0xffff) {
+            av_log(s->avctx, AV_LOG_ERROR, "DSYNC check failed\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        ofs += DCA_SUBBAND_SAMPLES;
+    }
+
+    // Inverse ADPCM
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        inverse_adpcm(s->subband_samples[ch], s->prediction_vq_index[ch],
+                      s->prediction_mode[ch], 0, s->nsubbands[ch],
+                      *sub_pos, nsamples);
+    }
+
+    // Joint subband coding
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        int src_ch = s->joint_intensity_index[ch] - 1;
+        if (src_ch >= 0) {
+            s->dcadsp->decode_joint(s->subband_samples[ch], s->subband_samples[src_ch],
+                                    s->joint_scale_factors[ch], s->nsubbands[ch],
+                                    s->nsubbands[src_ch], *sub_pos, nsamples);
+        }
+    }
+
+    // Advance subband sample pointer for the next subframe
+    *sub_pos = ofs;
+    return 0;
+}
+
+static void erase_adpcm_history(DCACoreDecoder *s)
+{
+    int ch, band;
+
+    // Erase ADPCM history from previous frame if
+    // predictor history switch was disabled
+    for (ch = 0; ch < DCA_CHANNELS; ch++)
+        for (band = 0; band < DCA_SUBBANDS; band++)
+            AV_ZERO128(s->subband_samples[ch][band] - DCA_ADPCM_COEFFS);
+}
+
+static int alloc_sample_buffer(DCACoreDecoder *s)
+{
+    int nchsamples = DCA_ADPCM_COEFFS + s->npcmblocks;
+    int nframesamples = nchsamples * DCA_CHANNELS * DCA_SUBBANDS;
+    int nlfesamples = DCA_LFE_HISTORY + s->npcmblocks / 2;
+    unsigned int size = s->subband_size;
+    int ch, band;
+
+    // Reallocate subband sample buffer
+    av_fast_mallocz(&s->subband_buffer, &s->subband_size,
+                    (nframesamples + nlfesamples) * sizeof(int32_t));
+    if (!s->subband_buffer)
+        return AVERROR(ENOMEM);
+
+    if (size != s->subband_size) {
+        for (ch = 0; ch < DCA_CHANNELS; ch++)
+            for (band = 0; band < DCA_SUBBANDS; band++)
+                s->subband_samples[ch][band] = s->subband_buffer +
+                    (ch * DCA_SUBBANDS + band) * nchsamples + DCA_ADPCM_COEFFS;
+        s->lfe_samples = s->subband_buffer + nframesamples;
+    }
+
+    if (!s->predictor_history)
+        erase_adpcm_history(s);
+
+    return 0;
+}
+
+static int parse_frame_data(DCACoreDecoder *s, enum HeaderType header, int xch_base)
+{
+    int sf, ch, ret, band, sub_pos, lfe_pos;
+
+    if ((ret = parse_coding_header(s, header, xch_base)) < 0)
+        return ret;
+
+    for (sf = 0, sub_pos = 0, lfe_pos = DCA_LFE_HISTORY; sf < s->nsubframes; sf++) {
+        if ((ret = parse_subframe_header(s, sf, header, xch_base)) < 0)
+            return ret;
+        if ((ret = parse_subframe_audio(s, sf, header, xch_base, &sub_pos, &lfe_pos)) < 0)
+            return ret;
+    }
+
+    for (ch = xch_base; ch < s->nchannels; ch++) {
+        // Determine number of active subbands for this channel
+        int nsubbands = s->nsubbands[ch];
+        if (s->joint_intensity_index[ch])
+            nsubbands = FFMAX(nsubbands, s->nsubbands[s->joint_intensity_index[ch] - 1]);
+
+        // Update history for ADPCM
+        for (band = 0; band < nsubbands; band++) {
+            int32_t *samples = s->subband_samples[ch][band] - DCA_ADPCM_COEFFS;
+            AV_COPY128(samples, samples + s->npcmblocks);
+        }
+
+        // Clear inactive subbands
+        for (; band < DCA_SUBBANDS; band++) {
+            int32_t *samples = s->subband_samples[ch][band] - DCA_ADPCM_COEFFS;
+            memset(samples, 0, (DCA_ADPCM_COEFFS + s->npcmblocks) * sizeof(int32_t));
+        }
+    }
+
+    return 0;
+}
+
+static int parse_xch_frame(DCACoreDecoder *s)
+{
+    int ret;
+
+    if (s->ch_mask & DCA_SPEAKER_MASK_Cs) {
+        av_log(s->avctx, AV_LOG_ERROR, "XCH with Cs speaker already present\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    if ((ret = parse_frame_data(s, HEADER_XCH, s->nchannels)) < 0)
+        return ret;
+
+    // Seek to the end of core frame, don't trust XCH frame size
+    if (ff_dca_seek_bits(&s->gb, s->frame_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XCH frame\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static int parse_xxch_frame(DCACoreDecoder *s)
+{
+    int xxch_nchsets, xxch_frame_size;
+    int ret, mask, header_size, header_pos = get_bits_count(&s->gb);
+
+    // XXCH sync word
+    if (get_bits_long(&s->gb, 32) != DCA_SYNCWORD_XXCH) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid XXCH sync word\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // XXCH frame header length
+    header_size = get_bits(&s->gb, 6) + 1;
+
+    // Check XXCH frame header CRC
+    if ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&s->gb, header_pos + 32, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid XXCH frame header checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // CRC presence flag for channel set header
+    s->xxch_crc_present = get_bits1(&s->gb);
+
+    // Number of bits for loudspeaker mask
+    s->xxch_mask_nbits = get_bits(&s->gb, 5) + 1;
+    if (s->xxch_mask_nbits <= DCA_SPEAKER_Cs) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid number of bits for XXCH speaker mask (%d)\n", s->xxch_mask_nbits);
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Number of channel sets
+    xxch_nchsets = get_bits(&s->gb, 2) + 1;
+    if (xxch_nchsets > 1) {
+        avpriv_request_sample(s->avctx, "%d XXCH channel sets", xxch_nchsets);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    // Channel set 0 data byte size
+    xxch_frame_size = get_bits(&s->gb, 14) + 1;
+
+    // Core loudspeaker activity mask
+    s->xxch_core_mask = get_bits_long(&s->gb, s->xxch_mask_nbits);
+
+    // Validate the core mask
+    mask = s->ch_mask;
+
+    if ((mask & DCA_SPEAKER_MASK_Ls) && (s->xxch_core_mask & DCA_SPEAKER_MASK_Lss))
+        mask = (mask & ~DCA_SPEAKER_MASK_Ls) | DCA_SPEAKER_MASK_Lss;
+
+    if ((mask & DCA_SPEAKER_MASK_Rs) && (s->xxch_core_mask & DCA_SPEAKER_MASK_Rss))
+        mask = (mask & ~DCA_SPEAKER_MASK_Rs) | DCA_SPEAKER_MASK_Rss;
+
+    if (mask != s->xxch_core_mask) {
+        av_log(s->avctx, AV_LOG_ERROR, "XXCH core speaker activity mask (%#x) disagrees with core (%#x)\n", s->xxch_core_mask, mask);
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Reserved
+    // Byte align
+    // CRC16 of XXCH frame header
+    if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XXCH frame header\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Parse XXCH channel set 0
+    if ((ret = parse_frame_data(s, HEADER_XXCH, s->nchannels)) < 0)
+        return ret;
+
+    if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8 + xxch_frame_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XXCH channel set\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static int parse_xbr_subframe(DCACoreDecoder *s, int xbr_base_ch, int xbr_nchannels,
+                              int *xbr_nsubbands, int xbr_transition_mode, int sf, int *sub_pos)
+{
+    int     xbr_nabits[DCA_CHANNELS];
+    int     xbr_bit_allocation[DCA_CHANNELS][DCA_SUBBANDS];
+    int     xbr_scale_nbits[DCA_CHANNELS];
+    int32_t xbr_scale_factors[DCA_CHANNELS][DCA_SUBBANDS][2];
+    int     ssf, ch, band, ofs;
+
+    // Check number of subband samples in this subframe
+    if (*sub_pos + s->nsubsubframes[sf] * DCA_SUBBAND_SAMPLES > s->npcmblocks) {
+        av_log(s->avctx, AV_LOG_ERROR, "Subband sample buffer overflow\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    // Number of bits for XBR bit allocation index
+    for (ch = xbr_base_ch; ch < xbr_nchannels; ch++)
+        xbr_nabits[ch] = get_bits(&s->gb, 2) + 2;
+
+    // XBR bit allocation index
+    for (ch = xbr_base_ch; ch < xbr_nchannels; ch++) {
+        for (band = 0; band < xbr_nsubbands[ch]; band++) {
+            xbr_bit_allocation[ch][band] = get_bits(&s->gb, xbr_nabits[ch]);
+            if (xbr_bit_allocation[ch][band] > DCA_ABITS_MAX) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid XBR bit allocation index\n");
+                return AVERROR_INVALIDDATA;
+            }
+        }
+    }
+
+    // Number of bits for scale indices
+    for (ch = xbr_base_ch; ch < xbr_nchannels; ch++) {
+        xbr_scale_nbits[ch] = get_bits(&s->gb, 3);
+        if (!xbr_scale_nbits[ch]) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid number of bits for XBR scale factor index\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // XBR scale factors
+    for (ch = xbr_base_ch; ch < xbr_nchannels; ch++) {
+        const uint32_t *scale_table;
+        int scale_size;
+
+        // Select the root square table
+        if (s->scale_factor_sel[ch] > 5) {
+            scale_table = ff_dca_scale_factor_quant7;
+            scale_size = FF_ARRAY_ELEMS(ff_dca_scale_factor_quant7);
+        } else {
+            scale_table = ff_dca_scale_factor_quant6;
+            scale_size = FF_ARRAY_ELEMS(ff_dca_scale_factor_quant6);
+        }
+
+        // Parse scale factor indices and look up scale factors from the root
+        // square table
+        for (band = 0; band < xbr_nsubbands[ch]; band++) {
+            if (xbr_bit_allocation[ch][band]) {
+                int scale_index = get_bits(&s->gb, xbr_scale_nbits[ch]);
+                if (scale_index >= scale_size) {
+                    av_log(s->avctx, AV_LOG_ERROR, "Invalid XBR scale factor index\n");
+                    return AVERROR_INVALIDDATA;
+                }
+                xbr_scale_factors[ch][band][0] = scale_table[scale_index];
+                if (xbr_transition_mode && s->transition_mode[sf][ch][band]) {
+                    scale_index = get_bits(&s->gb, xbr_scale_nbits[ch]);
+                    if (scale_index >= scale_size) {
+                        av_log(s->avctx, AV_LOG_ERROR, "Invalid XBR scale factor index\n");
+                        return AVERROR_INVALIDDATA;
+                    }
+                    xbr_scale_factors[ch][band][1] = scale_table[scale_index];
+                }
+            }
+        }
+    }
+
+    // Audio data
+    for (ssf = 0, ofs = *sub_pos; ssf < s->nsubsubframes[sf]; ssf++) {
+        for (ch = xbr_base_ch; ch < xbr_nchannels; ch++) {
+            if (get_bits_left(&s->gb) < 0)
+                return AVERROR_INVALIDDATA;
+
+            for (band = 0; band < xbr_nsubbands[ch]; band++) {
+                int ret, trans_ssf, abits = xbr_bit_allocation[ch][band];
+                int32_t audio[DCA_SUBBAND_SAMPLES], step_size, scale;
+
+                // Extract bits from the bit stream
+                if (abits > 7) {
+                    // No further encoding
+                    get_array(&s->gb, audio, DCA_SUBBAND_SAMPLES, abits - 3);
+                } else if (abits > 0) {
+                    // Block codes
+                    if ((ret = parse_block_codes(s, audio, abits)) < 0)
+                        return ret;
+                } else {
+                    // No bits allocated
+                    continue;
+                }
+
+                // Look up quantization step size
+                step_size = ff_dca_lossless_quant[abits];
+
+                // Identify transient location
+                if (xbr_transition_mode)
+                    trans_ssf = s->transition_mode[sf][ch][band];
+                else
+                    trans_ssf = 0;
+
+                // Determine proper scale factor
+                if (trans_ssf == 0 || ssf < trans_ssf)
+                    scale = xbr_scale_factors[ch][band][0];
+                else
+                    scale = xbr_scale_factors[ch][band][1];
+
+                dequantize(s->subband_samples[ch][band] + ofs,
+                           audio, step_size, scale, 1);
+            }
+        }
+
+        // DSYNC
+        if ((ssf == s->nsubsubframes[sf] - 1 || s->sync_ssf) && get_bits(&s->gb, 16) != 0xffff) {
+            av_log(s->avctx, AV_LOG_ERROR, "XBR-DSYNC check failed\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        ofs += DCA_SUBBAND_SAMPLES;
+    }
+
+    // Advance subband sample pointer for the next subframe
+    *sub_pos = ofs;
+    return 0;
+}
+
+static int parse_xbr_frame(DCACoreDecoder *s)
+{
+    int     xbr_frame_size[DCA_EXSS_CHSETS_MAX];
+    int     xbr_nchannels[DCA_EXSS_CHSETS_MAX];
+    int     xbr_nsubbands[DCA_EXSS_CHSETS_MAX * DCA_EXSS_CHANNELS_MAX];
+    int     xbr_nchsets, xbr_transition_mode, xbr_band_nbits, xbr_base_ch;
+    int     i, ch1, ch2, ret, header_size, header_pos = get_bits_count(&s->gb);
+
+    // XBR sync word
+    if (get_bits_long(&s->gb, 32) != DCA_SYNCWORD_XBR) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid XBR sync word\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // XBR frame header length
+    header_size = get_bits(&s->gb, 6) + 1;
+
+    // Check XBR frame header CRC
+    if ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&s->gb, header_pos + 32, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid XBR frame header checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Number of channel sets
+    xbr_nchsets = get_bits(&s->gb, 2) + 1;
+
+    // Channel set data byte size
+    for (i = 0; i < xbr_nchsets; i++)
+        xbr_frame_size[i] = get_bits(&s->gb, 14) + 1;
+
+    // Transition mode flag
+    xbr_transition_mode = get_bits1(&s->gb);
+
+    // Channel set headers
+    for (i = 0, ch2 = 0; i < xbr_nchsets; i++) {
+        xbr_nchannels[i] = get_bits(&s->gb, 3) + 1;
+        xbr_band_nbits = get_bits(&s->gb, 2) + 5;
+        for (ch1 = 0; ch1 < xbr_nchannels[i]; ch1++, ch2++) {
+            xbr_nsubbands[ch2] = get_bits(&s->gb, xbr_band_nbits) + 1;
+            if (xbr_nsubbands[ch2] > DCA_SUBBANDS) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid number of active XBR subbands (%d)\n", xbr_nsubbands[ch2]);
+                return AVERROR_INVALIDDATA;
+            }
+        }
+    }
+
+    // Reserved
+    // Byte align
+    // CRC16 of XBR frame header
+    if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of XBR frame header\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Channel set data
+    for (i = 0, xbr_base_ch = 0; i < xbr_nchsets; i++) {
+        header_pos = get_bits_count(&s->gb);
+
+        if (xbr_base_ch + xbr_nchannels[i] <= s->nchannels) {
+            int sf, sub_pos;
+
+            for (sf = 0, sub_pos = 0; sf < s->nsubframes; sf++) {
+                if ((ret = parse_xbr_subframe(s, xbr_base_ch,
+                                              xbr_base_ch + xbr_nchannels[i],
+                                              xbr_nsubbands, xbr_transition_mode,
+                                              sf, &sub_pos)) < 0)
+                    return ret;
+            }
+        }
+
+        xbr_base_ch += xbr_nchannels[i];
+
+        if (ff_dca_seek_bits(&s->gb, header_pos + xbr_frame_size[i] * 8)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Read past end of XBR channel set\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    return 0;
+}
+
+// Modified ISO/IEC 9899 linear congruential generator
+// Returns pseudorandom integer in range [-2^30, 2^30 - 1]
+static int rand_x96(DCACoreDecoder *s)
+{
+    s->x96_rand = 1103515245U * s->x96_rand + 12345U;
+    return (s->x96_rand & 0x7fffffff) - 0x40000000;
+}
+
+static int parse_x96_subframe_audio(DCACoreDecoder *s, int sf, int xch_base, int *sub_pos)
+{
+    int n, ssf, ch, band, ofs;
+
+    // Check number of subband samples in this subframe
+    int nsamples = s->nsubsubframes[sf] * DCA_SUBBAND_SAMPLES;
+    if (*sub_pos + nsamples > s->npcmblocks) {
+        av_log(s->avctx, AV_LOG_ERROR, "Subband sample buffer overflow\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    // VQ encoded or unallocated subbands
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        for (band = s->x96_subband_start; band < s->nsubbands[ch]; band++) {
+            // Get the sample pointer and scale factor
+            int32_t *samples = s->x96_subband_samples[ch][band] + *sub_pos;
+            int32_t scale    = s->scale_factors[ch][band >> 1][band & 1];
+
+            switch (s->bit_allocation[ch][band]) {
+            case 0: // No bits allocated for subband
+                if (scale <= 1)
+                    memset(samples, 0, nsamples * sizeof(int32_t));
+                else for (n = 0; n < nsamples; n++)
+                    // Generate scaled random samples
+                    samples[n] = mul31(rand_x96(s), scale);
+                break;
+
+            case 1: // VQ encoded subband
+                for (ssf = 0; ssf < (s->nsubsubframes[sf] + 1) / 2; ssf++) {
+                    // Extract the VQ address from the bit stream and look up
+                    // the VQ code book for up to 16 subband samples
+                    const int8_t *vq_samples = ff_dca_high_freq_vq[get_bits(&s->gb, 10)];
+                    // Scale and take the samples
+                    for (n = 0; n < FFMIN(nsamples - ssf * 16, 16); n++)
+                        *samples++ = clip23(vq_samples[n] * scale + (1 << 3) >> 4);
+                }
+                break;
+            }
+        }
+    }
+
+    // Audio data
+    for (ssf = 0, ofs = *sub_pos; ssf < s->nsubsubframes[sf]; ssf++) {
+        for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+            if (get_bits_left(&s->gb) < 0)
+                return AVERROR_INVALIDDATA;
+
+            for (band = s->x96_subband_start; band < s->nsubbands[ch]; band++) {
+                int ret, abits = s->bit_allocation[ch][band] - 1;
+                int32_t audio[DCA_SUBBAND_SAMPLES], step_size, scale;
+
+                // Not VQ encoded or unallocated subbands
+                if (abits < 1)
+                    continue;
+
+                // Extract bits from the bit stream
+                if ((ret = extract_audio(s, audio, abits, ch)) < 0)
+                    return ret;
+
+                // Select quantization step size table and look up quantization
+                // step size
+                if (s->bit_rate == 3)
+                    step_size = ff_dca_lossless_quant[abits];
+                else
+                    step_size = ff_dca_lossy_quant[abits];
+
+                // Get the scale factor
+                scale = s->scale_factors[ch][band >> 1][band & 1];
+
+                dequantize(s->x96_subband_samples[ch][band] + ofs,
+                           audio, step_size, scale, 0);
+            }
+        }
+
+        // DSYNC
+        if ((ssf == s->nsubsubframes[sf] - 1 || s->sync_ssf) && get_bits(&s->gb, 16) != 0xffff) {
+            av_log(s->avctx, AV_LOG_ERROR, "X96-DSYNC check failed\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        ofs += DCA_SUBBAND_SAMPLES;
+    }
+
+    // Inverse ADPCM
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        inverse_adpcm(s->x96_subband_samples[ch], s->prediction_vq_index[ch],
+                      s->prediction_mode[ch], s->x96_subband_start, s->nsubbands[ch],
+                      *sub_pos, nsamples);
+    }
+
+    // Joint subband coding
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        int src_ch = s->joint_intensity_index[ch] - 1;
+        if (src_ch >= 0) {
+            s->dcadsp->decode_joint(s->x96_subband_samples[ch], s->x96_subband_samples[src_ch],
+                                    s->joint_scale_factors[ch], s->nsubbands[ch],
+                                    s->nsubbands[src_ch], *sub_pos, nsamples);
+        }
+    }
+
+    // Advance subband sample pointer for the next subframe
+    *sub_pos = ofs;
+    return 0;
+}
+
+static void erase_x96_adpcm_history(DCACoreDecoder *s)
+{
+    int ch, band;
+
+    // Erase ADPCM history from previous frame if
+    // predictor history switch was disabled
+    for (ch = 0; ch < DCA_CHANNELS; ch++)
+        for (band = 0; band < DCA_SUBBANDS_X96; band++)
+            AV_ZERO128(s->x96_subband_samples[ch][band] - DCA_ADPCM_COEFFS);
+}
+
+static int alloc_x96_sample_buffer(DCACoreDecoder *s)
+{
+    int nchsamples = DCA_ADPCM_COEFFS + s->npcmblocks;
+    int nframesamples = nchsamples * DCA_CHANNELS * DCA_SUBBANDS_X96;
+    unsigned int size = s->x96_subband_size;
+    int ch, band;
+
+    // Reallocate subband sample buffer
+    av_fast_mallocz(&s->x96_subband_buffer, &s->x96_subband_size,
+                    nframesamples * sizeof(int32_t));
+    if (!s->x96_subband_buffer)
+        return AVERROR(ENOMEM);
+
+    if (size != s->x96_subband_size) {
+        for (ch = 0; ch < DCA_CHANNELS; ch++)
+            for (band = 0; band < DCA_SUBBANDS_X96; band++)
+                s->x96_subband_samples[ch][band] = s->x96_subband_buffer +
+                    (ch * DCA_SUBBANDS_X96 + band) * nchsamples + DCA_ADPCM_COEFFS;
+    }
+
+    if (!s->predictor_history)
+        erase_x96_adpcm_history(s);
+
+    return 0;
+}
+
+static int parse_x96_subframe_header(DCACoreDecoder *s, int xch_base)
+{
+    int ch, band, ret;
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    // Prediction mode
+    for (ch = xch_base; ch < s->x96_nchannels; ch++)
+        for (band = s->x96_subband_start; band < s->nsubbands[ch]; band++)
+            s->prediction_mode[ch][band] = get_bits1(&s->gb);
+
+    // Prediction coefficients VQ address
+    for (ch = xch_base; ch < s->x96_nchannels; ch++)
+        for (band = s->x96_subband_start; band < s->nsubbands[ch]; band++)
+            if (s->prediction_mode[ch][band])
+                s->prediction_vq_index[ch][band] = get_bits(&s->gb, 12);
+
+    // Bit allocation index
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        int sel = s->bit_allocation_sel[ch];
+        int abits = 0;
+
+        for (band = s->x96_subband_start; band < s->nsubbands[ch]; band++) {
+            // If Huffman code was used, the difference of abits was encoded
+            if (sel < 7)
+                abits += get_vlc(&s->gb, &vlc_quant_index[5 + 2 * s->x96_high_res], sel);
+            else
+                abits = get_bits(&s->gb, 3 + s->x96_high_res);
+
+            if (abits < 0 || abits > 7 + 8 * s->x96_high_res) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 bit allocation index\n");
+                return AVERROR_INVALIDDATA;
+            }
+
+            s->bit_allocation[ch][band] = abits;
+        }
+    }
+
+    // Scale factors
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        int sel = s->scale_factor_sel[ch];
+        int scale_index = 0;
+
+        // Extract scales for subbands which are transmitted even for
+        // unallocated subbands
+        for (band = s->x96_subband_start; band < s->nsubbands[ch]; band++) {
+            if ((ret = parse_scale(s, &scale_index, sel)) < 0)
+                return ret;
+            s->scale_factors[ch][band >> 1][band & 1] = ret;
+        }
+    }
+
+    // Joint subband codebook select
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        if (s->joint_intensity_index[ch]) {
+            s->joint_scale_sel[ch] = get_bits(&s->gb, 3);
+            if (s->joint_scale_sel[ch] == 7) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 joint scale factor code book\n");
+                return AVERROR_INVALIDDATA;
+            }
+        }
+    }
+
+    // Scale factors for joint subband coding
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        int src_ch = s->joint_intensity_index[ch] - 1;
+        if (src_ch >= 0) {
+            int sel = s->joint_scale_sel[ch];
+            for (band = s->nsubbands[ch]; band < s->nsubbands[src_ch]; band++) {
+                if ((ret = parse_joint_scale(s, sel)) < 0)
+                    return ret;
+                s->joint_scale_factors[ch][band] = ret;
+            }
+        }
+    }
+
+    // Side information CRC check word
+    if (s->crc_present)
+        skip_bits(&s->gb, 16);
+
+    return 0;
+}
+
+static int parse_x96_coding_header(DCACoreDecoder *s, int exss, int xch_base)
+{
+    int n, ch, header_size = 0, header_pos = get_bits_count(&s->gb);
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    if (exss) {
+        // Channel set header length
+        header_size = get_bits(&s->gb, 7) + 1;
+
+        // Check CRC
+        if (s->x96_crc_present
+            && (s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+            && ff_dca_check_crc(&s->gb, header_pos, header_pos + header_size * 8)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 channel set header checksum\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // High resolution flag
+    s->x96_high_res = get_bits1(&s->gb);
+
+    // First encoded subband
+    if (s->x96_rev_no < 8) {
+        s->x96_subband_start = get_bits(&s->gb, 5);
+        if (s->x96_subband_start > 27) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 subband start index (%d)\n", s->x96_subband_start);
+            return AVERROR_INVALIDDATA;
+        }
+    } else {
+        s->x96_subband_start = DCA_SUBBANDS;
+    }
+
+    // Subband activity count
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        s->nsubbands[ch] = get_bits(&s->gb, 6) + 1;
+        if (s->nsubbands[ch] < DCA_SUBBANDS) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 subband activity count (%d)\n", s->nsubbands[ch]);
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // Joint intensity coding index
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        if ((n = get_bits(&s->gb, 3)) && xch_base)
+            n += xch_base - 1;
+        if (n > s->x96_nchannels) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 joint intensity coding index\n");
+            return AVERROR_INVALIDDATA;
+        }
+        s->joint_intensity_index[ch] = n;
+    }
+
+    // Scale factor code book
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        s->scale_factor_sel[ch] = get_bits(&s->gb, 3);
+        if (s->scale_factor_sel[ch] >= 6) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 scale factor code book\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // Bit allocation quantizer select
+    for (ch = xch_base; ch < s->x96_nchannels; ch++)
+        s->bit_allocation_sel[ch] = get_bits(&s->gb, 3);
+
+    // Quantization index codebook select
+    for (n = 0; n < 6 + 4 * s->x96_high_res; n++)
+        for (ch = xch_base; ch < s->x96_nchannels; ch++)
+            s->quant_index_sel[ch][n] = get_bits(&s->gb, quant_index_sel_nbits[n]);
+
+    if (exss) {
+        // Reserved
+        // Byte align
+        // CRC16 of channel set header
+        if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Read past end of X96 channel set header\n");
+            return AVERROR_INVALIDDATA;
+        }
+    } else {
+        if (s->crc_present)
+            skip_bits(&s->gb, 16);
+    }
+
+    return 0;
+}
+
+static int parse_x96_frame_data(DCACoreDecoder *s, int exss, int xch_base)
+{
+    int sf, ch, ret, band, sub_pos;
+
+    if ((ret = parse_x96_coding_header(s, exss, xch_base)) < 0)
+        return ret;
+
+    for (sf = 0, sub_pos = 0; sf < s->nsubframes; sf++) {
+        if ((ret = parse_x96_subframe_header(s, xch_base)) < 0)
+            return ret;
+        if ((ret = parse_x96_subframe_audio(s, sf, xch_base, &sub_pos)) < 0)
+            return ret;
+    }
+
+    for (ch = xch_base; ch < s->x96_nchannels; ch++) {
+        // Determine number of active subbands for this channel
+        int nsubbands = s->nsubbands[ch];
+        if (s->joint_intensity_index[ch])
+            nsubbands = FFMAX(nsubbands, s->nsubbands[s->joint_intensity_index[ch] - 1]);
+
+        // Update history for ADPCM and clear inactive subbands
+        for (band = 0; band < DCA_SUBBANDS_X96; band++) {
+            int32_t *samples = s->x96_subband_samples[ch][band] - DCA_ADPCM_COEFFS;
+            if (band >= s->x96_subband_start && band < nsubbands)
+                AV_COPY128(samples, samples + s->npcmblocks);
+            else
+                memset(samples, 0, (DCA_ADPCM_COEFFS + s->npcmblocks) * sizeof(int32_t));
+        }
+    }
+
+    return 0;
+}
+
+static int parse_x96_frame(DCACoreDecoder *s)
+{
+    int ret;
+
+    // Revision number
+    s->x96_rev_no = get_bits(&s->gb, 4);
+    if (s->x96_rev_no < 1 || s->x96_rev_no > 8) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 revision (%d)\n", s->x96_rev_no);
+        return AVERROR_INVALIDDATA;
+    }
+
+    s->x96_crc_present = 0;
+    s->x96_nchannels = s->nchannels;
+
+    if ((ret = alloc_x96_sample_buffer(s)) < 0)
+        return ret;
+
+    if ((ret = parse_x96_frame_data(s, 0, 0)) < 0)
+        return ret;
+
+    // Seek to the end of core frame
+    if (ff_dca_seek_bits(&s->gb, s->frame_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of X96 frame\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static int parse_x96_frame_exss(DCACoreDecoder *s)
+{
+    int     x96_frame_size[DCA_EXSS_CHSETS_MAX];
+    int     x96_nchannels[DCA_EXSS_CHSETS_MAX];
+    int     x96_nchsets, x96_base_ch;
+    int     i, ret, header_size, header_pos = get_bits_count(&s->gb);
+
+    // X96 sync word
+    if (get_bits_long(&s->gb, 32) != DCA_SYNCWORD_X96) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 sync word\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // X96 frame header length
+    header_size = get_bits(&s->gb, 6) + 1;
+
+    // Check X96 frame header CRC
+    if ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&s->gb, header_pos + 32, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 frame header checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Revision number
+    s->x96_rev_no = get_bits(&s->gb, 4);
+    if (s->x96_rev_no < 1 || s->x96_rev_no > 8) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid X96 revision (%d)\n", s->x96_rev_no);
+        return AVERROR_INVALIDDATA;
+    }
+
+    // CRC presence flag for channel set header
+    s->x96_crc_present = get_bits1(&s->gb);
+
+    // Number of channel sets
+    x96_nchsets = get_bits(&s->gb, 2) + 1;
+
+    // Channel set data byte size
+    for (i = 0; i < x96_nchsets; i++)
+        x96_frame_size[i] = get_bits(&s->gb, 12) + 1;
+
+    // Number of channels in channel set
+    for (i = 0; i < x96_nchsets; i++)
+        x96_nchannels[i] = get_bits(&s->gb, 3) + 1;
+
+    // Reserved
+    // Byte align
+    // CRC16 of X96 frame header
+    if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of X96 frame header\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    if ((ret = alloc_x96_sample_buffer(s)) < 0)
+        return ret;
+
+    // Channel set data
+    for (i = 0, x96_base_ch = 0; i < x96_nchsets; i++) {
+        header_pos = get_bits_count(&s->gb);
+
+        if (x96_base_ch + x96_nchannels[i] <= s->nchannels) {
+            s->x96_nchannels = x96_base_ch + x96_nchannels[i];
+            if ((ret = parse_x96_frame_data(s, 1, x96_base_ch)) < 0)
+                return ret;
+        }
+
+        x96_base_ch += x96_nchannels[i];
+
+        if (ff_dca_seek_bits(&s->gb, header_pos + x96_frame_size[i] * 8)) {
+            av_log(s->avctx, AV_LOG_ERROR, "Read past end of X96 channel set\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    return 0;
+}
+
+static int parse_aux_data(DCACoreDecoder *s)
+{
+    int aux_pos;
+
+    if (get_bits_left(&s->gb) < 0)
+        return AVERROR_INVALIDDATA;
+
+    // Auxiliary data byte count (can't be trusted)
+    skip_bits(&s->gb, 6);
+
+    // 4-byte align
+    skip_bits_long(&s->gb, -get_bits_count(&s->gb) & 31);
+
+    // Auxiliary data sync word
+    if (get_bits_long(&s->gb, 32) != DCA_SYNCWORD_REV1AUX) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid auxiliary data sync word\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    aux_pos = get_bits_count(&s->gb);
+
+    // Auxiliary decode time stamp flag
+    if (get_bits1(&s->gb))
+        skip_bits_long(&s->gb, 47);
+
+    // Auxiliary dynamic downmix flag
+    if (s->prim_dmix_embedded = get_bits1(&s->gb)) {
+        int i, m, n;
+
+        // Auxiliary primary channel downmix type
+        s->prim_dmix_type = get_bits(&s->gb, 3);
+        if (s->prim_dmix_type >= DCA_DMIX_TYPE_COUNT) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid primary channel set downmix type\n");
+            return AVERROR_INVALIDDATA;
+        }
+
+        // Size of downmix coefficients matrix
+        m = ff_dca_dmix_primary_nch[s->prim_dmix_type];
+        n = ff_dca_channels[s->audio_mode] + !!s->lfe_present;
+
+        // Dynamic downmix code coefficients
+        for (i = 0; i < m * n; i++) {
+            int code = get_bits(&s->gb, 9);
+            int sign = (code >> 8) - 1;
+            unsigned int index = code & 0xff;
+            if (index >= FF_DCA_DMIXTABLE_SIZE) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid downmix coefficient index\n");
+                return AVERROR_INVALIDDATA;
+            }
+            s->prim_dmix_coeff[i] = (ff_dca_dmixtable[index] ^ sign) - sign;
+        }
+    }
+
+    // Byte align
+    skip_bits(&s->gb, -get_bits_count(&s->gb) & 7);
+
+    // CRC16 of auxiliary data
+    skip_bits(&s->gb, 16);
+
+    // Check CRC
+    if ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&s->gb, aux_pos, get_bits_count(&s->gb))) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid auxiliary data checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static int parse_optional_info(DCACoreDecoder *s)
+{
+    DCAContext *dca = s->avctx->priv_data;
+    int ret = -1;
+
+    // Time code stamp
+    if (s->ts_present)
+        skip_bits_long(&s->gb, 32);
+
+    // Auxiliary data
+    if (s->aux_present && (ret = parse_aux_data(s)) < 0
+        && (s->avctx->err_recognition & AV_EF_EXPLODE))
+        return ret;
+
+    if (ret < 0)
+        s->prim_dmix_embedded = 0;
+
+    // Core extensions
+    if (s->ext_audio_present && !dca->core_only) {
+        int sync_pos = FFMIN(s->frame_size / 4, s->gb.size_in_bits / 32) - 1;
+        int last_pos = get_bits_count(&s->gb) / 32;
+        int size, dist;
+
+        // Search for extension sync words aligned on 4-byte boundary. Search
+        // must be done backwards from the end of core frame to work around
+        // sync word aliasing issues.
+        switch (s->ext_audio_type) {
+        case EXT_AUDIO_XCH:
+            if (dca->request_channel_layout)
+                break;
+
+            // The distance between XCH sync word and end of the core frame
+            // must be equal to XCH frame size. Off by one error is allowed for
+            // compatibility with legacy bitstreams. Minimum XCH frame size is
+            // 96 bytes. AMODE and PCHS are further checked to reduce
+            // probability of alias sync detection.
+            for (; sync_pos >= last_pos; sync_pos--) {
+                if (AV_RB32(s->gb.buffer + sync_pos * 4) == DCA_SYNCWORD_XCH) {
+                    s->gb.index = (sync_pos + 1) * 32;
+                    size = get_bits(&s->gb, 10) + 1;
+                    dist = s->frame_size - sync_pos * 4;
+                    if (size >= 96
+                        && (size == dist || size - 1 == dist)
+                        && get_bits(&s->gb, 7) == 0x08) {
+                        s->xch_pos = get_bits_count(&s->gb);
+                        break;
+                    }
+                }
+            }
+
+            if (s->avctx->err_recognition & AV_EF_EXPLODE) {
+                av_log(s->avctx, AV_LOG_ERROR, "XCH sync word not found\n");
+                return AVERROR_INVALIDDATA;
+            }
+            break;
+
+        case EXT_AUDIO_X96:
+            // The distance between X96 sync word and end of the core frame
+            // must be equal to X96 frame size. Minimum X96 frame size is 96
+            // bytes.
+            for (; sync_pos >= last_pos; sync_pos--) {
+                if (AV_RB32(s->gb.buffer + sync_pos * 4) == DCA_SYNCWORD_X96) {
+                    s->gb.index = (sync_pos + 1) * 32;
+                    size = get_bits(&s->gb, 12) + 1;
+                    dist = s->frame_size - sync_pos * 4;
+                    if (size >= 96 && size == dist) {
+                        s->x96_pos = get_bits_count(&s->gb);
+                        break;
+                    }
+                }
+            }
+
+            if (s->avctx->err_recognition & AV_EF_EXPLODE) {
+                av_log(s->avctx, AV_LOG_ERROR, "X96 sync word not found\n");
+                return AVERROR_INVALIDDATA;
+            }
+            break;
+
+        case EXT_AUDIO_XXCH:
+            if (dca->request_channel_layout)
+                break;
+
+            // XXCH frame header CRC must be valid. Minimum XXCH frame header
+            // size is 11 bytes.
+            for (; sync_pos >= last_pos; sync_pos--) {
+                if (AV_RB32(s->gb.buffer + sync_pos * 4) == DCA_SYNCWORD_XXCH) {
+                    s->gb.index = (sync_pos + 1) * 32;
+                    size = get_bits(&s->gb, 6) + 1;
+                    if (size >= 11 &&
+                        !ff_dca_check_crc(&s->gb, (sync_pos + 1) * 32,
+                                          sync_pos * 32 + size * 8)) {
+                        s->xxch_pos = sync_pos * 32;
+                        break;
+                    }
+                }
+            }
+
+            if (s->avctx->err_recognition & AV_EF_EXPLODE) {
+                av_log(s->avctx, AV_LOG_ERROR, "XXCH sync word not found\n");
+                return AVERROR_INVALIDDATA;
+            }
+            break;
+        }
+    }
+
+    return 0;
+}
+
+int ff_dca_core_parse(DCACoreDecoder *s, uint8_t *data, int size)
+{
+    int ret;
+
+    s->ext_audio_mask = 0;
+    s->xch_pos = s->xxch_pos = s->x96_pos = 0;
+
+    if ((ret = init_get_bits8(&s->gb, data, size)) < 0)
+        return ret;
+
+    skip_bits_long(&s->gb, 32);
+    if ((ret = parse_frame_header(s)) < 0)
+        return ret;
+    if ((ret = alloc_sample_buffer(s)) < 0)
+        return ret;
+    if ((ret = parse_frame_data(s, HEADER_CORE, 0)) < 0)
+        return ret;
+    if ((ret = parse_optional_info(s)) < 0)
+        return ret;
+
+    // Workaround for DTS in WAV
+    if (s->frame_size > size && s->frame_size < size + 4) {
+        av_log(s->avctx, AV_LOG_DEBUG, "Working around excessive core frame size (%d > %d)\n", s->frame_size, size);
+        s->frame_size = size;
+    }
+
+    if (ff_dca_seek_bits(&s->gb, s->frame_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of core frame\n");
+        if (s->avctx->err_recognition & AV_EF_EXPLODE)
+            return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+int ff_dca_core_parse_exss(DCACoreDecoder *s, uint8_t *data, DCAExssAsset *asset)
+{
+    AVCodecContext *avctx = s->avctx;
+    DCAContext *dca = avctx->priv_data;
+    GetBitContext gb = s->gb;
+    int exss_mask = asset ? asset->extension_mask : 0;
+    int ret = 0, ext = 0;
+
+    // Parse (X)XCH unless downmixing
+    if (!dca->request_channel_layout) {
+        if (exss_mask & DCA_EXSS_XXCH) {
+            if ((ret = init_get_bits8(&s->gb, data + asset->xxch_offset, asset->xxch_size)) < 0)
+                return ret;
+            ret = parse_xxch_frame(s);
+            ext = DCA_EXSS_XXCH;
+        } else if (s->xxch_pos) {
+            s->gb.index = s->xxch_pos;
+            ret = parse_xxch_frame(s);
+            ext = DCA_CSS_XXCH;
+        } else if (s->xch_pos) {
+            s->gb.index = s->xch_pos;
+            ret = parse_xch_frame(s);
+            ext = DCA_CSS_XCH;
+        }
+
+        // Revert to primary channel set in case (X)XCH parsing fails
+        if (ret < 0) {
+            if (avctx->err_recognition & AV_EF_EXPLODE)
+                return ret;
+            s->nchannels = ff_dca_channels[s->audio_mode];
+            s->ch_mask = audio_mode_ch_mask[s->audio_mode];
+            if (s->lfe_present)
+                s->ch_mask |= DCA_SPEAKER_MASK_LFE1;
+        } else {
+            s->ext_audio_mask |= ext;
+        }
+    }
+
+    // Parse XBR
+    if (exss_mask & DCA_EXSS_XBR) {
+        if ((ret = init_get_bits8(&s->gb, data + asset->xbr_offset, asset->xbr_size)) < 0)
+            return ret;
+        if ((ret = parse_xbr_frame(s)) < 0) {
+            if (avctx->err_recognition & AV_EF_EXPLODE)
+                return ret;
+        } else {
+            s->ext_audio_mask |= DCA_EXSS_XBR;
+        }
+    }
+
+    // Parse X96 unless decoding XLL
+    if (!(dca->packet & DCA_PACKET_XLL)) {
+        if (exss_mask & DCA_EXSS_X96) {
+            if ((ret = init_get_bits8(&s->gb, data + asset->x96_offset, asset->x96_size)) < 0)
+                return ret;
+            if ((ret = parse_x96_frame_exss(s)) < 0) {
+                if (ret == AVERROR(ENOMEM) || (avctx->err_recognition & AV_EF_EXPLODE))
+                    return ret;
+            } else {
+                s->ext_audio_mask |= DCA_EXSS_X96;
+            }
+        } else if (s->x96_pos) {
+            s->gb = gb;
+            s->gb.index = s->x96_pos;
+            if ((ret = parse_x96_frame(s)) < 0) {
+                if (ret == AVERROR(ENOMEM) || (avctx->err_recognition & AV_EF_EXPLODE))
+                    return ret;
+            } else {
+                s->ext_audio_mask |= DCA_CSS_X96;
+            }
+        }
+    }
+
+    return 0;
+}
+
+static int map_prm_ch_to_spkr(DCACoreDecoder *s, int ch)
+{
+    int pos, spkr;
+
+    // Try to map this channel to core first
+    pos = ff_dca_channels[s->audio_mode];
+    if (ch < pos) {
+        spkr = prm_ch_to_spkr_map[s->audio_mode][ch];
+        if (s->ext_audio_mask & (DCA_CSS_XXCH | DCA_EXSS_XXCH)) {
+            if (s->xxch_core_mask & (1U << spkr))
+                return spkr;
+            if (spkr == DCA_SPEAKER_Ls && (s->xxch_core_mask & DCA_SPEAKER_MASK_Lss))
+                return DCA_SPEAKER_Lss;
+            if (spkr == DCA_SPEAKER_Rs && (s->xxch_core_mask & DCA_SPEAKER_MASK_Rss))
+                return DCA_SPEAKER_Rss;
+            return -1;
+        }
+        return spkr;
+    }
+
+    // Then XCH
+    if ((s->ext_audio_mask & DCA_CSS_XCH) && ch == pos)
+        return DCA_SPEAKER_Cs;
+
+    // Then XXCH
+    if (s->ext_audio_mask & (DCA_CSS_XXCH | DCA_EXSS_XXCH)) {
+        for (spkr = DCA_SPEAKER_Cs; spkr < s->xxch_mask_nbits; spkr++)
+            if (s->xxch_spkr_mask & (1U << spkr))
+                if (pos++ == ch)
+                    return spkr;
+    }
+
+    // No mapping
+    return -1;
+}
+
+static void erase_dsp_history(DCACoreDecoder *s)
+{
+    memset(s->dcadsp_data, 0, sizeof(s->dcadsp_data));
+    s->output_history_lfe_fixed = 0;
+    s->output_history_lfe_float = 0;
+}
+
+static void set_filter_mode(DCACoreDecoder *s, int mode)
+{
+    if (s->filter_mode != mode) {
+        erase_dsp_history(s);
+        s->filter_mode = mode;
+    }
+}
+
+int ff_dca_core_filter_fixed(DCACoreDecoder *s, int x96_synth)
+{
+    int n, ch, spkr, nsamples, x96_nchannels = 0;
+    const int32_t *filter_coeff;
+    int32_t *ptr;
+
+    // Externally set x96_synth flag implies that X96 synthesis should be
+    // enabled, yet actual X96 subband data should be discarded. This is a
+    // special case for lossless residual decoder that ignores X96 data if
+    // present.
+    if (!x96_synth && (s->ext_audio_mask & (DCA_CSS_X96 | DCA_EXSS_X96))) {
+        x96_nchannels = s->x96_nchannels;
+        x96_synth = 1;
+    }
+    if (x96_synth < 0)
+        x96_synth = 0;
+
+    s->output_rate = s->sample_rate << x96_synth;
+    s->npcmsamples = nsamples = (s->npcmblocks * DCA_PCMBLOCK_SAMPLES) << x96_synth;
+
+    // Reallocate PCM output buffer
+    av_fast_malloc(&s->output_buffer, &s->output_size,
+                   nsamples * av_popcount(s->ch_mask) * sizeof(int32_t));
+    if (!s->output_buffer)
+        return AVERROR(ENOMEM);
+
+    ptr = (int32_t *)s->output_buffer;
+    for (spkr = 0; spkr < DCA_SPEAKER_COUNT; spkr++) {
+        if (s->ch_mask & (1U << spkr)) {
+            s->output_samples[spkr] = ptr;
+            ptr += nsamples;
+        } else {
+            s->output_samples[spkr] = NULL;
+        }
+    }
+
+    // Handle change of filtering mode
+    set_filter_mode(s, x96_synth | DCA_FILTER_MODE_FIXED);
+
+    // Select filter
+    if (x96_synth)
+        filter_coeff = ff_dca_fir_64bands_fixed;
+    else if (s->filter_perfect)
+        filter_coeff = ff_dca_fir_32bands_perfect_fixed;
+    else
+        filter_coeff = ff_dca_fir_32bands_nonperfect_fixed;
+
+    // Filter primary channels
+    for (ch = 0; ch < s->nchannels; ch++) {
+        // Map this primary channel to speaker
+        spkr = map_prm_ch_to_spkr(s, ch);
+        if (spkr < 0)
+            return AVERROR(EINVAL);
+
+        // Filter bank reconstruction
+        s->dcadsp->sub_qmf_fixed[x96_synth](
+            &s->synth,
+            &s->dcadct,
+            s->output_samples[spkr],
+            s->subband_samples[ch],
+            ch < x96_nchannels ? s->x96_subband_samples[ch] : NULL,
+            s->dcadsp_data[ch].u.fix.hist1,
+            &s->dcadsp_data[ch].offset,
+            s->dcadsp_data[ch].u.fix.hist2,
+            filter_coeff,
+            s->npcmblocks);
+    }
+
+    // Filter LFE channel
+    if (s->lfe_present) {
+        int32_t *samples = s->output_samples[DCA_SPEAKER_LFE1];
+        int nlfesamples = s->npcmblocks >> 1;
+
+        // Check LFF
+        if (s->lfe_present == LFE_FLAG_128) {
+            av_log(s->avctx, AV_LOG_ERROR, "Fixed point mode doesn't support LFF=1\n");
+            return AVERROR(EINVAL);
+        }
+
+        // Offset intermediate buffer for X96
+        if (x96_synth)
+            samples += nsamples / 2;
+
+        // Interpolate LFE channel
+        s->dcadsp->lfe_fir_fixed(samples, s->lfe_samples + DCA_LFE_HISTORY,
+                                 ff_dca_lfe_fir_64_fixed, s->npcmblocks);
+
+        if (x96_synth) {
+            // Filter 96 kHz oversampled LFE PCM to attenuate high frequency
+            // (47.6 - 48.0 kHz) components of interpolation image
+            s->dcadsp->lfe_x96_fixed(s->output_samples[DCA_SPEAKER_LFE1],
+                                     samples, &s->output_history_lfe_fixed,
+                                     nsamples / 2);
+
+        }
+
+        // Update LFE history
+        for (n = DCA_LFE_HISTORY - 1; n >= 0; n--)
+            s->lfe_samples[n] = s->lfe_samples[nlfesamples + n];
+    }
+
+    return 0;
+}
+
+static int filter_frame_fixed(DCACoreDecoder *s, AVFrame *frame)
+{
+    AVCodecContext *avctx = s->avctx;
+    DCAContext *dca = avctx->priv_data;
+    int i, n, ch, ret, spkr, nsamples;
+
+    // Don't filter twice when falling back from XLL
+    if (!(dca->packet & DCA_PACKET_XLL) && (ret = ff_dca_core_filter_fixed(s, 0)) < 0)
+        return ret;
+
+    avctx->sample_rate = s->output_rate;
+    avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
+    avctx->bits_per_raw_sample = 24;
+
+    frame->nb_samples = nsamples = s->npcmsamples;
+    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
+        return ret;
+
+    // Undo embedded XCH downmix
+    if (s->es_format && (s->ext_audio_mask & DCA_CSS_XCH)
+        && s->audio_mode >= AMODE_2F2R) {
+        s->dcadsp->dmix_sub_xch(s->output_samples[DCA_SPEAKER_Ls],
+                                s->output_samples[DCA_SPEAKER_Rs],
+                                s->output_samples[DCA_SPEAKER_Cs],
+                                nsamples);
+
+    }
+
+    // Undo embedded XXCH downmix
+    if ((s->ext_audio_mask & (DCA_CSS_XXCH | DCA_EXSS_XXCH))
+        && s->xxch_dmix_embedded) {
+        int scale_inv   = s->xxch_dmix_scale_inv;
+        int *coeff_ptr  = s->xxch_dmix_coeff;
+        int xch_base    = ff_dca_channels[s->audio_mode];
+        av_assert1(s->nchannels - xch_base <= DCA_XXCH_CHANNELS_MAX);
+
+        // Undo embedded core downmix pre-scaling
+        for (spkr = 0; spkr < s->xxch_mask_nbits; spkr++) {
+            if (s->xxch_core_mask & (1U << spkr)) {
+                s->dcadsp->dmix_scale_inv(s->output_samples[spkr],
+                                          scale_inv, nsamples);
+            }
+        }
+
+        // Undo downmix
+        for (ch = xch_base; ch < s->nchannels; ch++) {
+            int src_spkr = map_prm_ch_to_spkr(s, ch);
+            if (src_spkr < 0)
+                return AVERROR(EINVAL);
+            for (spkr = 0; spkr < s->xxch_mask_nbits; spkr++) {
+                if (s->xxch_dmix_mask[ch - xch_base] & (1U << spkr)) {
+                    int coeff = mul16(*coeff_ptr++, scale_inv);
+                    if (coeff) {
+                        s->dcadsp->dmix_sub(s->output_samples[spkr    ],
+                                            s->output_samples[src_spkr],
+                                            coeff, nsamples);
+                    }
+                }
+            }
+        }
+    }
+
+    if (!(s->ext_audio_mask & (DCA_CSS_XXCH | DCA_CSS_XCH | DCA_EXSS_XXCH))) {
+        // Front sum/difference decoding
+        if ((s->sumdiff_front && s->audio_mode > AMODE_MONO)
+            || s->audio_mode == AMODE_STEREO_SUMDIFF) {
+            s->fixed_dsp->butterflies_fixed(s->output_samples[DCA_SPEAKER_L],
+                                            s->output_samples[DCA_SPEAKER_R],
+                                            nsamples);
+        }
+
+        // Surround sum/difference decoding
+        if (s->sumdiff_surround && s->audio_mode >= AMODE_2F2R) {
+            s->fixed_dsp->butterflies_fixed(s->output_samples[DCA_SPEAKER_Ls],
+                                            s->output_samples[DCA_SPEAKER_Rs],
+                                            nsamples);
+        }
+    }
+
+    // Downmix primary channel set to stereo
+    if (s->request_mask != s->ch_mask) {
+        ff_dca_downmix_to_stereo_fixed(s->dcadsp,
+                                       s->output_samples,
+                                       s->prim_dmix_coeff,
+                                       nsamples, s->ch_mask);
+    }
+
+    for (i = 0; i < avctx->channels; i++) {
+        int32_t *samples = s->output_samples[s->ch_remap[i]];
+        int32_t *plane = (int32_t *)frame->extended_data[i];
+        for (n = 0; n < nsamples; n++)
+            plane[n] = clip23(samples[n]) * (1 << 8);
+    }
+
+    return 0;
+}
+
+static int filter_frame_float(DCACoreDecoder *s, AVFrame *frame)
+{
+    AVCodecContext *avctx = s->avctx;
+    int x96_nchannels = 0, x96_synth = 0;
+    int i, n, ch, ret, spkr, nsamples, nchannels;
+    float *output_samples[DCA_SPEAKER_COUNT] = { NULL }, *ptr;
+    const float *filter_coeff;
+
+    if (s->ext_audio_mask & (DCA_CSS_X96 | DCA_EXSS_X96)) {
+        x96_nchannels = s->x96_nchannels;
+        x96_synth = 1;
+    }
+
+    avctx->sample_rate = s->sample_rate << x96_synth;
+    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
+    avctx->bits_per_raw_sample = 0;
+
+    frame->nb_samples = nsamples = (s->npcmblocks * DCA_PCMBLOCK_SAMPLES) << x96_synth;
+    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
+        return ret;
+
+    // Build reverse speaker to channel mapping
+    for (i = 0; i < avctx->channels; i++)
+        output_samples[s->ch_remap[i]] = (float *)frame->extended_data[i];
+
+    // Allocate space for extra channels
+    nchannels = av_popcount(s->ch_mask) - avctx->channels;
+    if (nchannels > 0) {
+        av_fast_malloc(&s->output_buffer, &s->output_size,
+                       nsamples * nchannels * sizeof(float));
+        if (!s->output_buffer)
+            return AVERROR(ENOMEM);
+
+        ptr = (float *)s->output_buffer;
+        for (spkr = 0; spkr < DCA_SPEAKER_COUNT; spkr++) {
+            if (!(s->ch_mask & (1U << spkr)))
+                continue;
+            if (output_samples[spkr])
+                continue;
+            output_samples[spkr] = ptr;
+            ptr += nsamples;
+        }
+    }
+
+    // Handle change of filtering mode
+    set_filter_mode(s, x96_synth);
+
+    // Select filter
+    if (x96_synth)
+        filter_coeff = ff_dca_fir_64bands;
+    else if (s->filter_perfect)
+        filter_coeff = ff_dca_fir_32bands_perfect;
+    else
+        filter_coeff = ff_dca_fir_32bands_nonperfect;
+
+    // Filter primary channels
+    for (ch = 0; ch < s->nchannels; ch++) {
+        // Map this primary channel to speaker
+        spkr = map_prm_ch_to_spkr(s, ch);
+        if (spkr < 0)
+            return AVERROR(EINVAL);
+
+        // Filter bank reconstruction
+        s->dcadsp->sub_qmf_float[x96_synth](
+            &s->synth,
+            &s->imdct[x96_synth],
+            output_samples[spkr],
+            s->subband_samples[ch],
+            ch < x96_nchannels ? s->x96_subband_samples[ch] : NULL,
+            s->dcadsp_data[ch].u.flt.hist1,
+            &s->dcadsp_data[ch].offset,
+            s->dcadsp_data[ch].u.flt.hist2,
+            filter_coeff,
+            s->npcmblocks,
+            1.0f / (1 << (17 - x96_synth)));
+    }
+
+    // Filter LFE channel
+    if (s->lfe_present) {
+        int dec_select = (s->lfe_present == LFE_FLAG_128);
+        float *samples = output_samples[DCA_SPEAKER_LFE1];
+        int nlfesamples = s->npcmblocks >> (dec_select + 1);
+
+        // Offset intermediate buffer for X96
+        if (x96_synth)
+            samples += nsamples / 2;
+
+        // Select filter
+        if (dec_select)
+            filter_coeff = ff_dca_lfe_fir_128;
+        else
+            filter_coeff = ff_dca_lfe_fir_64;
+
+        // Interpolate LFE channel
+        s->dcadsp->lfe_fir_float[dec_select](
+            samples, s->lfe_samples + DCA_LFE_HISTORY,
+            filter_coeff, s->npcmblocks);
+
+        if (x96_synth) {
+            // Filter 96 kHz oversampled LFE PCM to attenuate high frequency
+            // (47.6 - 48.0 kHz) components of interpolation image
+            s->dcadsp->lfe_x96_float(output_samples[DCA_SPEAKER_LFE1],
+                                     samples, &s->output_history_lfe_float,
+                                     nsamples / 2);
+        }
+
+        // Update LFE history
+        for (n = DCA_LFE_HISTORY - 1; n >= 0; n--)
+            s->lfe_samples[n] = s->lfe_samples[nlfesamples + n];
+    }
+
+    // Undo embedded XCH downmix
+    if (s->es_format && (s->ext_audio_mask & DCA_CSS_XCH)
+        && s->audio_mode >= AMODE_2F2R) {
+        s->float_dsp->vector_fmac_scalar(output_samples[DCA_SPEAKER_Ls],
+                                         output_samples[DCA_SPEAKER_Cs],
+                                         -M_SQRT1_2, nsamples);
+        s->float_dsp->vector_fmac_scalar(output_samples[DCA_SPEAKER_Rs],
+                                         output_samples[DCA_SPEAKER_Cs],
+                                         -M_SQRT1_2, nsamples);
+    }
+
+    // Undo embedded XXCH downmix
+    if ((s->ext_audio_mask & (DCA_CSS_XXCH | DCA_EXSS_XXCH))
+        && s->xxch_dmix_embedded) {
+        float scale_inv = s->xxch_dmix_scale_inv * (1.0f / (1 << 16));
+        int *coeff_ptr  = s->xxch_dmix_coeff;
+        int xch_base    = ff_dca_channels[s->audio_mode];
+        av_assert1(s->nchannels - xch_base <= DCA_XXCH_CHANNELS_MAX);
+
+        // Undo downmix
+        for (ch = xch_base; ch < s->nchannels; ch++) {
+            int src_spkr = map_prm_ch_to_spkr(s, ch);
+            if (src_spkr < 0)
+                return AVERROR(EINVAL);
+            for (spkr = 0; spkr < s->xxch_mask_nbits; spkr++) {
+                if (s->xxch_dmix_mask[ch - xch_base] & (1U << spkr)) {
+                    int coeff = *coeff_ptr++;
+                    if (coeff) {
+                        s->float_dsp->vector_fmac_scalar(output_samples[    spkr],
+                                                         output_samples[src_spkr],
+                                                         coeff * (-1.0f / (1 << 15)),
+                                                         nsamples);
+                    }
+                }
+            }
+        }
+
+        // Undo embedded core downmix pre-scaling
+        for (spkr = 0; spkr < s->xxch_mask_nbits; spkr++) {
+            if (s->xxch_core_mask & (1U << spkr)) {
+                s->float_dsp->vector_fmul_scalar(output_samples[spkr],
+                                                 output_samples[spkr],
+                                                 scale_inv, nsamples);
+            }
+        }
+    }
+
+    if (!(s->ext_audio_mask & (DCA_CSS_XXCH | DCA_CSS_XCH | DCA_EXSS_XXCH))) {
+        // Front sum/difference decoding
+        if ((s->sumdiff_front && s->audio_mode > AMODE_MONO)
+            || s->audio_mode == AMODE_STEREO_SUMDIFF) {
+            s->float_dsp->butterflies_float(output_samples[DCA_SPEAKER_L],
+                                            output_samples[DCA_SPEAKER_R],
+                                            nsamples);
+        }
+
+        // Surround sum/difference decoding
+        if (s->sumdiff_surround && s->audio_mode >= AMODE_2F2R) {
+            s->float_dsp->butterflies_float(output_samples[DCA_SPEAKER_Ls],
+                                            output_samples[DCA_SPEAKER_Rs],
+                                            nsamples);
+        }
+    }
+
+    // Downmix primary channel set to stereo
+    if (s->request_mask != s->ch_mask) {
+        ff_dca_downmix_to_stereo_float(s->float_dsp, output_samples,
+                                       s->prim_dmix_coeff,
+                                       nsamples, s->ch_mask);
+    }
+
+    return 0;
+}
+
+int ff_dca_core_filter_frame(DCACoreDecoder *s, AVFrame *frame)
+{
+    AVCodecContext *avctx = s->avctx;
+    DCAContext *dca = avctx->priv_data;
+    DCAExssAsset *asset = &dca->exss.assets[0];
+    enum AVMatrixEncoding matrix_encoding;
+    int ret;
+
+    // Handle downmixing to stereo request
+    if (dca->request_channel_layout == DCA_SPEAKER_LAYOUT_STEREO
+        && s->audio_mode > AMODE_MONO && s->prim_dmix_embedded
+        && (s->prim_dmix_type == DCA_DMIX_TYPE_LoRo ||
+            s->prim_dmix_type == DCA_DMIX_TYPE_LtRt))
+        s->request_mask = DCA_SPEAKER_LAYOUT_STEREO;
+    else
+        s->request_mask = s->ch_mask;
+    if (!ff_dca_set_channel_layout(avctx, s->ch_remap, s->request_mask))
+        return AVERROR(EINVAL);
+
+    // Force fixed point mode when falling back from XLL
+    if ((avctx->flags & AV_CODEC_FLAG_BITEXACT) || ((dca->packet & DCA_PACKET_EXSS)
+                                                    && (asset->extension_mask & DCA_EXSS_XLL)))
+        ret = filter_frame_fixed(s, frame);
+    else
+        ret = filter_frame_float(s, frame);
+    if (ret < 0)
+        return ret;
+
+    // Set profile, bit rate, etc
+    if (s->ext_audio_mask & DCA_EXSS_MASK)
+        avctx->profile = FF_PROFILE_DTS_HD_HRA;
+    else if (s->ext_audio_mask & (DCA_CSS_XXCH | DCA_CSS_XCH))
+        avctx->profile = FF_PROFILE_DTS_ES;
+    else if (s->ext_audio_mask & DCA_CSS_X96)
+        avctx->profile = FF_PROFILE_DTS_96_24;
+    else
+        avctx->profile = FF_PROFILE_DTS;
+
+    if (s->bit_rate > 3 && !(s->ext_audio_mask & DCA_EXSS_MASK))
+        avctx->bit_rate = s->bit_rate;
+    else
+        avctx->bit_rate = 0;
+
+    if (s->audio_mode == AMODE_STEREO_TOTAL || (s->request_mask != s->ch_mask &&
+                                                s->prim_dmix_type == DCA_DMIX_TYPE_LtRt))
+        matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
+    else
+        matrix_encoding = AV_MATRIX_ENCODING_NONE;
+    if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
+        return ret;
+
+    return 0;
+}
+
+av_cold void ff_dca_core_flush(DCACoreDecoder *s)
+{
+    if (s->subband_buffer) {
+        erase_adpcm_history(s);
+        memset(s->lfe_samples, 0, DCA_LFE_HISTORY * sizeof(int32_t));
+    }
+
+    if (s->x96_subband_buffer)
+        erase_x96_adpcm_history(s);
+
+    erase_dsp_history(s);
+}
+
+av_cold int ff_dca_core_init(DCACoreDecoder *s)
+{
+    dca_init_vlcs();
+
+    if (!(s->float_dsp = avpriv_float_dsp_alloc(0)))
+        return -1;
+    if (!(s->fixed_dsp = avpriv_alloc_fixed_dsp(0)))
+        return -1;
+
+    ff_dcadct_init(&s->dcadct);
+    if (ff_mdct_init(&s->imdct[0], 6, 1, 1.0) < 0)
+        return -1;
+    if (ff_mdct_init(&s->imdct[1], 7, 1, 1.0) < 0)
+        return -1;
+    ff_synth_filter_init(&s->synth);
+
+    s->x96_rand = 1;
+    return 0;
+}
+
+av_cold void ff_dca_core_close(DCACoreDecoder *s)
+{
+    av_freep(&s->float_dsp);
+    av_freep(&s->fixed_dsp);
+
+    ff_mdct_end(&s->imdct[0]);
+    ff_mdct_end(&s->imdct[1]);
+
+    av_freep(&s->subband_buffer);
+    s->subband_size = 0;
+
+    av_freep(&s->x96_subband_buffer);
+    s->x96_subband_size = 0;
+
+    av_freep(&s->output_buffer);
+    s->output_size = 0;
+}

libavcodec/dca_core.h

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_DCA_CORE_H
+#define AVCODEC_DCA_CORE_H
+
+#include "libavutil/common.h"
+#include "libavutil/float_dsp.h"
+#include "libavutil/fixed_dsp.h"
+#include "libavutil/mem.h"
+
+#include "avcodec.h"
+#include "internal.h"
+#include "get_bits.h"
+#include "dca.h"
+#include "dca_exss.h"
+#include "dcadsp.h"
+#include "dcadct.h"
+#include "fft.h"
+#include "synth_filter.h"
+
+#define DCA_CHANNELS            7
+#define DCA_SUBBANDS            32
+#define DCA_SUBBANDS_X96        64
+#define DCA_SUBFRAMES           16
+#define DCA_SUBBAND_SAMPLES     8
+#define DCA_PCMBLOCK_SAMPLES    32
+#define DCA_ADPCM_COEFFS        4
+#define DCA_LFE_HISTORY         8
+#define DCA_CODE_BOOKS          10
+#define DCA_ABITS_MAX           26
+
+#define DCA_CORE_CHANNELS_MAX       6
+#define DCA_DMIX_CHANNELS_MAX       4
+#define DCA_XXCH_CHANNELS_MAX       2
+#define DCA_EXSS_CHANNELS_MAX       8
+#define DCA_EXSS_CHSETS_MAX         4
+
+#define DCA_FILTER_MODE_X96     0x01
+#define DCA_FILTER_MODE_FIXED   0x02
+
+typedef struct DCADSPData {
+    union {
+        struct {
+            DECLARE_ALIGNED(32, float, hist1)[1024];
+            DECLARE_ALIGNED(32, float, hist2)[64];
+        } flt;
+        struct {
+            DECLARE_ALIGNED(32, int32_t, hist1)[1024];
+            DECLARE_ALIGNED(32, int32_t, hist2)[64];
+        } fix;
+    } u;
+    int offset;
+} DCADSPData;
+
+typedef struct DCACoreDecoder {
+    AVCodecContext  *avctx;
+    GetBitContext   gb;
+
+    // Bit stream header
+    int     crc_present;        ///< CRC present flag
+    int     npcmblocks;         ///< Number of PCM sample blocks
+    int     frame_size;         ///< Primary frame byte size
+    int     audio_mode;         ///< Audio channel arrangement
+    int     sample_rate;        ///< Core audio sampling frequency
+    int     bit_rate;           ///< Transmission bit rate
+    int     drc_present;        ///< Embedded dynamic range flag
+    int     ts_present;         ///< Embedded time stamp flag
+    int     aux_present;        ///< Auxiliary data flag
+    int     ext_audio_type;     ///< Extension audio descriptor flag
+    int     ext_audio_present;  ///< Extended coding flag
+    int     sync_ssf;           ///< Audio sync word insertion flag
+    int     lfe_present;        ///< Low frequency effects flag
+    int     predictor_history;  ///< Predictor history flag switch
+    int     filter_perfect;     ///< Multirate interpolator switch
+    int     source_pcm_res;     ///< Source PCM resolution
+    int     es_format;          ///< Extended surround (ES) mastering flag
+    int     sumdiff_front;      ///< Front sum/difference flag
+    int     sumdiff_surround;   ///< Surround sum/difference flag
+
+    // Primary audio coding header
+    int         nsubframes;     ///< Number of subframes
+    int         nchannels;      ///< Number of primary audio channels (incl. extension channels)
+    int         ch_mask;        ///< Speaker layout mask (incl. LFE and extension channels)
+    int8_t      nsubbands[DCA_CHANNELS];                ///< Subband activity count
+    int8_t      subband_vq_start[DCA_CHANNELS];         ///< High frequency VQ start subband
+    int8_t      joint_intensity_index[DCA_CHANNELS];    ///< Joint intensity coding index
+    int8_t      transition_mode_sel[DCA_CHANNELS];      ///< Transient mode code book
+    int8_t      scale_factor_sel[DCA_CHANNELS];         ///< Scale factor code book
+    int8_t      bit_allocation_sel[DCA_CHANNELS];       ///< Bit allocation quantizer select
+    int8_t      quant_index_sel[DCA_CHANNELS][DCA_CODE_BOOKS];  ///< Quantization index codebook select
+    int32_t     scale_factor_adj[DCA_CHANNELS][DCA_CODE_BOOKS]; ///< Scale factor adjustment
+
+    // Primary audio coding side information
+    int8_t      nsubsubframes[DCA_SUBFRAMES];   ///< Subsubframe count for each subframe
+    int8_t      prediction_mode[DCA_CHANNELS][DCA_SUBBANDS_X96];            ///< Prediction mode
+    int16_t     prediction_vq_index[DCA_CHANNELS][DCA_SUBBANDS_X96];        ///< Prediction coefficients VQ address
+    int8_t      bit_allocation[DCA_CHANNELS][DCA_SUBBANDS_X96];             ///< Bit allocation index
+    int8_t      transition_mode[DCA_SUBFRAMES][DCA_CHANNELS][DCA_SUBBANDS]; ///< Transition mode
+    int32_t     scale_factors[DCA_CHANNELS][DCA_SUBBANDS][2];               ///< Scale factors (2x for transients and X96)
+    int8_t      joint_scale_sel[DCA_CHANNELS];                              ///< Joint subband codebook select
+    int32_t     joint_scale_factors[DCA_CHANNELS][DCA_SUBBANDS_X96];        ///< Scale factors for joint subband coding
+
+    // Auxiliary data
+    int     prim_dmix_embedded; ///< Auxiliary dynamic downmix flag
+    int     prim_dmix_type;     ///< Auxiliary primary channel downmix type
+    int     prim_dmix_coeff[DCA_DMIX_CHANNELS_MAX * DCA_CORE_CHANNELS_MAX]; ///< Dynamic downmix code coefficients
+
+    // Core extensions
+    int     ext_audio_mask;     ///< Bit mask of fully decoded core extensions
+
+    // XCH extension data
+    int     xch_pos;    ///< Bit position of XCH frame in core substream
+
+    // XXCH extension data
+    int     xxch_crc_present;       ///< CRC presence flag for XXCH channel set header
+    int     xxch_mask_nbits;        ///< Number of bits for loudspeaker mask
+    int     xxch_core_mask;         ///< Core loudspeaker activity mask
+    int     xxch_spkr_mask;         ///< Loudspeaker layout mask
+    int     xxch_dmix_embedded;     ///< Downmix already performed by encoder
+    int     xxch_dmix_scale_inv;    ///< Downmix scale factor
+    int     xxch_dmix_mask[DCA_XXCH_CHANNELS_MAX];  ///< Downmix channel mapping mask
+    int     xxch_dmix_coeff[DCA_XXCH_CHANNELS_MAX * DCA_CORE_CHANNELS_MAX];     ///< Downmix coefficients
+    int     xxch_pos;   ///< Bit position of XXCH frame in core substream
+
+    // X96 extension data
+    int     x96_rev_no;         ///< X96 revision number
+    int     x96_crc_present;    ///< CRC presence flag for X96 channel set header
+    int     x96_nchannels;      ///< Number of primary channels in X96 extension
+    int     x96_high_res;       ///< X96 high resolution flag
+    int     x96_subband_start;  ///< First encoded subband in X96 extension
+    int     x96_rand;           ///< Random seed for generating samples for unallocated X96 subbands
+    int     x96_pos;            ///< Bit position of X96 frame in core substream
+
+    // Sample buffers
+    unsigned int    x96_subband_size;
+    int32_t         *x96_subband_buffer;    ///< X96 subband sample buffer base
+    int32_t         *x96_subband_samples[DCA_CHANNELS][DCA_SUBBANDS_X96];   ///< X96 subband samples
+
+    unsigned int    subband_size;
+    int32_t         *subband_buffer;    ///< Subband sample buffer base
+    int32_t         *subband_samples[DCA_CHANNELS][DCA_SUBBANDS];   ///< Subband samples
+    int32_t         *lfe_samples;    ///< Decimated LFE samples
+
+    // DSP contexts
+    DCADSPData              dcadsp_data[DCA_CHANNELS];    ///< FIR history buffers
+    DCADSPContext           *dcadsp;
+    DCADCTContext           dcadct;
+    FFTContext              imdct[2];
+    SynthFilterContext      synth;
+    AVFloatDSPContext       *float_dsp;
+    AVFixedDSPContext       *fixed_dsp;
+
+    // PCM output data
+    unsigned int    output_size;
+    void            *output_buffer;                         ///< PCM output buffer base
+    int32_t         *output_samples[DCA_SPEAKER_COUNT];     ///< PCM output for fixed point mode
+    int32_t         output_history_lfe_fixed;               ///< LFE PCM history for X96 filter
+    float           output_history_lfe_float;               ///< LFE PCM history for X96 filter
+
+    int     ch_remap[DCA_SPEAKER_COUNT];   ///< Channel to speaker map
+    int     request_mask;   ///< Requested channel layout (for stereo downmix)
+
+    int     npcmsamples;    ///< Number of PCM samples per channel
+    int     output_rate;    ///< Output sample rate (1x or 2x header rate)
+
+    int     filter_mode;    ///< Previous filtering mode for detecting changes
+} DCACoreDecoder;
+
+static inline int ff_dca_core_map_spkr(DCACoreDecoder *core, int spkr)
+{
+    if (core->ch_mask & (1U << spkr))
+        return spkr;
+    if (spkr == DCA_SPEAKER_Lss && (core->ch_mask & DCA_SPEAKER_MASK_Ls))
+        return DCA_SPEAKER_Ls;
+    if (spkr == DCA_SPEAKER_Rss && (core->ch_mask & DCA_SPEAKER_MASK_Rs))
+        return DCA_SPEAKER_Rs;
+    return -1;
+}
+
+int ff_dca_core_parse(DCACoreDecoder *s, uint8_t *data, int size);
+int ff_dca_core_parse_exss(DCACoreDecoder *s, uint8_t *data, DCAExssAsset *asset);
+int ff_dca_core_filter_fixed(DCACoreDecoder *s, int x96_synth);
+int ff_dca_core_filter_frame(DCACoreDecoder *s, AVFrame *frame);
+av_cold void ff_dca_core_flush(DCACoreDecoder *s);
+av_cold int ff_dca_core_init(DCACoreDecoder *s);
+av_cold void ff_dca_core_close(DCACoreDecoder *s);
+
+#endif

libavcodec/dca_exss.c

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "dcadec.h"
+#include "dcadata.h"
+
+static int count_chs_for_mask(int mask)
+{
+    return av_popcount(mask) + av_popcount(mask & 0xae66);
+}
+
+static void parse_xll_parameters(DCAExssParser *s, DCAExssAsset *asset)
+{
+    // Size of XLL data in extension substream
+    asset->xll_size = get_bits(&s->gb, s->exss_size_nbits) + 1;
+
+    // XLL sync word present flag
+    if (asset->xll_sync_present = get_bits1(&s->gb)) {
+        int xll_delay_nbits;
+
+        // Peak bit rate smoothing buffer size
+        skip_bits(&s->gb, 4);
+
+        // Number of bits for XLL decoding delay
+        xll_delay_nbits = get_bits(&s->gb, 5) + 1;
+
+        // Initial XLL decoding delay in frames
+        asset->xll_delay_nframes = get_bits_long(&s->gb, xll_delay_nbits);
+
+        // Number of bytes offset to XLL sync
+        asset->xll_sync_offset = get_bits(&s->gb, s->exss_size_nbits);
+    } else {
+        asset->xll_delay_nframes = 0;
+        asset->xll_sync_offset = 0;
+    }
+}
+
+static void parse_lbr_parameters(DCAExssParser *s, DCAExssAsset *asset)
+{
+    // Size of LBR component in extension substream
+    asset->lbr_size = get_bits(&s->gb, 14) + 1;
+
+    // LBR sync word present flag
+    if (get_bits1(&s->gb))
+        // LBR sync distance
+        skip_bits(&s->gb, 2);
+}
+
+static int parse_descriptor(DCAExssParser *s, DCAExssAsset *asset)
+{
+    int i, j, drc_present, descr_size, descr_pos = get_bits_count(&s->gb);
+
+    // Size of audio asset descriptor in bytes
+    descr_size = get_bits(&s->gb, 9) + 1;
+
+    // Audio asset identifier
+    asset->asset_index = get_bits(&s->gb, 3);
+
+    //
+    // Per stream static metadata
+    //
+
+    if (s->static_fields_present) {
+        // Asset type descriptor presence
+        if (get_bits1(&s->gb))
+            // Asset type descriptor
+            skip_bits(&s->gb, 4);
+
+        // Language descriptor presence
+        if (get_bits1(&s->gb))
+            // Language descriptor
+            skip_bits(&s->gb, 24);
+
+        // Additional textual information presence
+        if (get_bits1(&s->gb)) {
+            // Byte size of additional text info
+            int text_size = get_bits(&s->gb, 10) + 1;
+
+            // Sanity check available size
+            if (get_bits_left(&s->gb) < text_size * 8)
+                return AVERROR_INVALIDDATA;
+
+            // Additional textual information string
+            skip_bits_long(&s->gb, text_size * 8);
+        }
+
+        // PCM bit resolution
+        asset->pcm_bit_res = get_bits(&s->gb, 5) + 1;
+
+        // Maximum sample rate
+        asset->max_sample_rate = ff_dca_sampling_freqs[get_bits(&s->gb, 4)];
+
+        // Total number of channels
+        asset->nchannels_total = get_bits(&s->gb, 8) + 1;
+
+        // One to one map channel to speakers
+        if (asset->one_to_one_map_ch_to_spkr = get_bits1(&s->gb)) {
+            int spkr_mask_nbits = 0;
+            int spkr_remap_nsets;
+            int nspeakers[8];
+
+            // Embedded stereo flag
+            if (asset->nchannels_total > 2)
+                asset->embedded_stereo = get_bits1(&s->gb);
+
+            // Embedded 6 channels flag
+            if (asset->nchannels_total > 6)
+                asset->embedded_6ch = get_bits1(&s->gb);
+
+            // Speaker mask enabled flag
+            if (asset->spkr_mask_enabled = get_bits1(&s->gb)) {
+                // Number of bits for speaker activity mask
+                spkr_mask_nbits = (get_bits(&s->gb, 2) + 1) << 2;
+
+                // Loudspeaker activity mask
+                asset->spkr_mask = get_bits(&s->gb, spkr_mask_nbits);
+            }
+
+            // Number of speaker remapping sets
+            if ((spkr_remap_nsets = get_bits(&s->gb, 3)) && !spkr_mask_nbits) {
+                av_log(s->avctx, AV_LOG_ERROR, "Speaker mask disabled yet there are remapping sets\n");
+                return AVERROR_INVALIDDATA;
+            }
+
+            // Standard loudspeaker layout mask
+            for (i = 0; i < spkr_remap_nsets; i++)
+                nspeakers[i] = count_chs_for_mask(get_bits(&s->gb, spkr_mask_nbits));
+
+            for (i = 0; i < spkr_remap_nsets; i++) {
+                // Number of channels to be decoded for speaker remapping
+                int nch_for_remaps = get_bits(&s->gb, 5) + 1;
+
+                for (j = 0; j < nspeakers[i]; j++) {
+                    // Decoded channels to output speaker mapping mask
+                    int remap_ch_mask = get_bits_long(&s->gb, nch_for_remaps);
+
+                    // Loudspeaker remapping codes
+                    skip_bits_long(&s->gb, av_popcount(remap_ch_mask) * 5);
+                }
+            }
+        } else {
+            asset->embedded_stereo = 0;
+            asset->embedded_6ch = 0;
+            asset->spkr_mask_enabled = 0;
+            asset->spkr_mask = 0;
+
+            // Representation type
+            asset->representation_type = get_bits(&s->gb, 3);
+        }
+    }
+
+    //
+    // DRC, DNC and mixing metadata
+    //
+
+    // Dynamic range coefficient presence flag
+    drc_present = get_bits1(&s->gb);
+
+    // Code for dynamic range coefficient
+    if (drc_present)
+        skip_bits(&s->gb, 8);
+
+    // Dialog normalization presence flag
+    if (get_bits1(&s->gb))
+        // Dialog normalization code
+        skip_bits(&s->gb, 5);
+
+    // DRC for stereo downmix
+    if (drc_present && asset->embedded_stereo)
+        skip_bits(&s->gb, 8);
+
+    // Mixing metadata presence flag
+    if (s->mix_metadata_enabled && get_bits1(&s->gb)) {
+        int nchannels_dmix;
+
+        // External mixing flag
+        skip_bits1(&s->gb);
+
+        // Post mixing / replacement gain adjustment
+        skip_bits(&s->gb, 6);
+
+        // DRC prior to mixing
+        if (get_bits(&s->gb, 2) == 3)
+            // Custom code for mixing DRC
+            skip_bits(&s->gb, 8);
+        else
+            // Limit for mixing DRC
+            skip_bits(&s->gb, 3);
+
+        // Scaling type for channels of main audio
+        // Scaling parameters of main audio
+        if (get_bits1(&s->gb))
+            for (i = 0; i < s->nmixoutconfigs; i++)
+                skip_bits_long(&s->gb, 6 * s->nmixoutchs[i]);
+        else
+            skip_bits_long(&s->gb, 6 * s->nmixoutconfigs);
+
+        nchannels_dmix = asset->nchannels_total;
+        if (asset->embedded_6ch)
+            nchannels_dmix += 6;
+        if (asset->embedded_stereo)
+            nchannels_dmix += 2;
+
+        for (i = 0; i < s->nmixoutconfigs; i++) {
+            if (!s->nmixoutchs[i]) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid speaker layout mask for mixing configuration\n");
+                return AVERROR_INVALIDDATA;
+            }
+            for (j = 0; j < nchannels_dmix; j++) {
+                // Mix output mask
+                int mix_map_mask = get_bits(&s->gb, s->nmixoutchs[i]);
+
+                // Mixing coefficients
+                skip_bits_long(&s->gb, av_popcount(mix_map_mask) * 6);
+            }
+        }
+    }
+
+    //
+    // Decoder navigation data
+    //
+
+    // Coding mode for the asset
+    asset->coding_mode = get_bits(&s->gb, 2);
+
+    // Coding components used in asset
+    switch (asset->coding_mode) {
+    case 0: // Coding mode that may contain multiple coding components
+        asset->extension_mask = get_bits(&s->gb, 12);
+
+        if (asset->extension_mask & DCA_EXSS_CORE) {
+            // Size of core component in extension substream
+            asset->core_size = get_bits(&s->gb, 14) + 1;
+            // Core sync word present flag
+            if (get_bits1(&s->gb))
+                // Core sync distance
+                skip_bits(&s->gb, 2);
+        }
+
+        if (asset->extension_mask & DCA_EXSS_XBR)
+            // Size of XBR extension in extension substream
+            asset->xbr_size = get_bits(&s->gb, 14) + 1;
+
+        if (asset->extension_mask & DCA_EXSS_XXCH)
+            // Size of XXCH extension in extension substream
+            asset->xxch_size = get_bits(&s->gb, 14) + 1;
+
+        if (asset->extension_mask & DCA_EXSS_X96)
+            // Size of X96 extension in extension substream
+            asset->x96_size = get_bits(&s->gb, 12) + 1;
+
+        if (asset->extension_mask & DCA_EXSS_LBR)
+            parse_lbr_parameters(s, asset);
+
+        if (asset->extension_mask & DCA_EXSS_XLL)
+            parse_xll_parameters(s, asset);
+
+        if (asset->extension_mask & DCA_EXSS_RSV1)
+            skip_bits(&s->gb, 16);
+
+        if (asset->extension_mask & DCA_EXSS_RSV2)
+            skip_bits(&s->gb, 16);
+        break;
+
+    case 1: // Loss-less coding mode without CBR component
+        asset->extension_mask = DCA_EXSS_XLL;
+        parse_xll_parameters(s, asset);
+        break;
+
+    case 2: // Low bit rate mode
+        asset->extension_mask = DCA_EXSS_LBR;
+        parse_lbr_parameters(s, asset);
+        break;
+
+    case 3: // Auxiliary coding mode
+        asset->extension_mask = 0;
+
+        // Size of auxiliary coded data
+        skip_bits(&s->gb, 14);
+
+        // Auxiliary codec identification
+        skip_bits(&s->gb, 8);
+
+        // Aux sync word present flag
+        if (get_bits1(&s->gb))
+            // Aux sync distance
+            skip_bits(&s->gb, 3);
+        break;
+    }
+
+    if (asset->extension_mask & DCA_EXSS_XLL)
+        // DTS-HD stream ID
+        asset->hd_stream_id = get_bits(&s->gb, 3);
+
+    // One to one mixing flag
+    // Per channel main audio scaling flag
+    // Main audio scaling codes
+    // Decode asset in secondary decoder flag
+    // Revision 2 DRC metadata
+    // Reserved
+    // Zero pad
+    if (ff_dca_seek_bits(&s->gb, descr_pos + descr_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of EXSS asset descriptor\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}
+
+static int set_exss_offsets(DCAExssAsset *asset)
+{
+    int offs = asset->asset_offset;
+    int size = asset->asset_size;
+
+    if (asset->extension_mask & DCA_EXSS_CORE) {
+        asset->core_offset = offs;
+        if (asset->core_size > size)
+            return AVERROR_INVALIDDATA;
+        offs += asset->core_size;
+        size -= asset->core_size;
+    }
+
+    if (asset->extension_mask & DCA_EXSS_XBR) {
+        asset->xbr_offset = offs;
+        if (asset->xbr_size > size)
+            return AVERROR_INVALIDDATA;
+        offs += asset->xbr_size;
+        size -= asset->xbr_size;
+    }
+
+    if (asset->extension_mask & DCA_EXSS_XXCH) {
+        asset->xxch_offset = offs;
+        if (asset->xxch_size > size)
+            return AVERROR_INVALIDDATA;
+        offs += asset->xxch_size;
+        size -= asset->xxch_size;
+    }
+
+    if (asset->extension_mask & DCA_EXSS_X96) {
+        asset->x96_offset = offs;
+        if (asset->x96_size > size)
+            return AVERROR_INVALIDDATA;
+        offs += asset->x96_size;
+        size -= asset->x96_size;
+    }
+
+    if (asset->extension_mask & DCA_EXSS_LBR) {
+        asset->lbr_offset = offs;
+        if (asset->lbr_size > size)
+            return AVERROR_INVALIDDATA;
+        offs += asset->lbr_size;
+        size -= asset->lbr_size;
+    }
+
+    if (asset->extension_mask & DCA_EXSS_XLL) {
+        asset->xll_offset = offs;
+        if (asset->xll_size > size)
+            return AVERROR_INVALIDDATA;
+        offs += asset->xll_size;
+        size -= asset->xll_size;
+    }
+
+    return 0;
+}
+
+int ff_dca_exss_parse(DCAExssParser *s, uint8_t *data, int size)
+{
+    int i, ret, offset, wide_hdr, header_size;
+
+    if ((ret = init_get_bits8(&s->gb, data, size)) < 0)
+        return ret;
+
+    // Extension substream sync word
+    skip_bits_long(&s->gb, 32);
+
+    // User defined bits
+    skip_bits(&s->gb, 8);
+
+    // Extension substream index
+    s->exss_index = get_bits(&s->gb, 2);
+
+    // Flag indicating short or long header size
+    wide_hdr = get_bits1(&s->gb);
+
+    // Extension substream header length
+    header_size = get_bits(&s->gb, 8 + 4 * wide_hdr) + 1;
+
+    // Check CRC
+    if ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&s->gb, 32 + 8, header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid EXSS header checksum\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    s->exss_size_nbits = 16 + 4 * wide_hdr;
+
+    // Number of bytes of extension substream
+    s->exss_size = get_bits(&s->gb, s->exss_size_nbits) + 1;
+    if (s->exss_size > size) {
+        av_log(s->avctx, AV_LOG_ERROR, "Packet too short for EXSS frame\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    // Per stream static fields presence flag
+    if (s->static_fields_present = get_bits1(&s->gb)) {
+        int active_exss_mask[8];
+
+        // Reference clock code
+        skip_bits(&s->gb, 2);
+
+        // Extension substream frame duration
+        skip_bits(&s->gb, 3);
+
+        // Timecode presence flag
+        if (get_bits1(&s->gb))
+            // Timecode data
+            skip_bits_long(&s->gb, 36);
+
+        // Number of defined audio presentations
+        s->npresents = get_bits(&s->gb, 3) + 1;
+        if (s->npresents > 1) {
+            avpriv_request_sample(s->avctx, "%d audio presentations", s->npresents);
+            return AVERROR_PATCHWELCOME;
+        }
+
+        // Number of audio assets in extension substream
+        s->nassets = get_bits(&s->gb, 3) + 1;
+        if (s->nassets > 1) {
+            avpriv_request_sample(s->avctx, "%d audio assets", s->nassets);
+            return AVERROR_PATCHWELCOME;
+        }
+
+        // Active extension substream mask for audio presentation
+        for (i = 0; i < s->npresents; i++)
+            active_exss_mask[i] = get_bits(&s->gb, s->exss_index + 1);
+
+        // Active audio asset mask
+        for (i = 0; i < s->npresents; i++)
+            skip_bits_long(&s->gb, av_popcount(active_exss_mask[i]) * 8);
+
+        // Mixing metadata enable flag
+        if (s->mix_metadata_enabled = get_bits1(&s->gb)) {
+            int spkr_mask_nbits;
+
+            // Mixing metadata adjustment level
+            skip_bits(&s->gb, 2);
+
+            // Number of bits for mixer output speaker activity mask
+            spkr_mask_nbits = (get_bits(&s->gb, 2) + 1) << 2;
+
+            // Number of mixing configurations
+            s->nmixoutconfigs = get_bits(&s->gb, 2) + 1;
+
+            // Speaker layout mask for mixer output channels
+            for (i = 0; i < s->nmixoutconfigs; i++)
+                s->nmixoutchs[i] = count_chs_for_mask(get_bits(&s->gb, spkr_mask_nbits));
+        }
+    } else {
+        s->npresents = 1;
+        s->nassets = 1;
+    }
+
+    // Size of encoded asset data in bytes
+    offset = header_size;
+    for (i = 0; i < s->nassets; i++) {
+        s->assets[i].asset_offset = offset;
+        s->assets[i].asset_size = get_bits(&s->gb, s->exss_size_nbits) + 1;
+        offset += s->assets[i].asset_size;
+        if (offset > s->exss_size) {
+            av_log(s->avctx, AV_LOG_ERROR, "EXSS asset out of bounds\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    // Audio asset descriptor
+    for (i = 0; i < s->nassets; i++) {
+        if ((ret = parse_descriptor(s, &s->assets[i])) < 0)
+            return ret;
+        if ((ret = set_exss_offsets(&s->assets[i])) < 0) {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid extension size in EXSS asset descriptor\n");
+            return ret;
+        }
+    }
+
+    // Backward compatible core present
+    // Backward compatible core substream index
+    // Backward compatible core asset index
+    // Reserved
+    // Byte align
+    // CRC16 of extension substream header
+    if (ff_dca_seek_bits(&s->gb, header_size * 8)) {
+        av_log(s->avctx, AV_LOG_ERROR, "Read past end of EXSS header\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    return 0;
+}

libavcodec/dca_exss.h

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_DCA_EXSS_H
+#define AVCODEC_DCA_EXSS_H
+
+#include "libavutil/common.h"
+
+#include "avcodec.h"
+#include "get_bits.h"
+
+typedef struct DCAExssAsset {
+    int     asset_offset;   ///< Offset to asset data from start of substream
+    int     asset_size;     ///< Size of encoded asset data
+    int     asset_index;    ///< Audio asset identifier
+
+    int     pcm_bit_res;                ///< PCM bit resolution
+    int     max_sample_rate;            ///< Maximum sample rate
+    int     nchannels_total;            ///< Total number of channels
+    int     one_to_one_map_ch_to_spkr;  ///< One to one channel to speaker mapping flag
+    int     embedded_stereo;            ///< Embedded stereo flag
+    int     embedded_6ch;               ///< Embedded 6 channels flag
+    int     spkr_mask_enabled;          ///< Speaker mask enabled flag
+    int     spkr_mask;                  ///< Loudspeaker activity mask
+    int     representation_type;        ///< Representation type
+
+    int     coding_mode;        ///< Coding mode for the asset
+    int     extension_mask;     ///< Coding components used in asset
+
+    int     core_offset;    ///< Offset to core component from start of substream
+    int     core_size;      ///< Size of core component in extension substream
+
+    int     xbr_offset;     ///< Offset to XBR extension from start of substream
+    int     xbr_size;       ///< Size of XBR extension in extension substream
+
+    int     xxch_offset;    ///< Offset to XXCH extension from start of substream
+    int     xxch_size;      ///< Size of XXCH extension in extension substream
+
+    int     x96_offset;     ///< Offset to X96 extension from start of substream
+    int     x96_size;       ///< Size of X96 extension in extension substream
+
+    int     lbr_offset;     ///< Offset to LBR component from start of substream
+    int     lbr_size;       ///< Size of LBR component in extension substream
+
+    int     xll_offset;         ///< Offset to XLL data from start of substream
+    int     xll_size;           ///< Size of XLL data in extension substream
+    int     xll_sync_present;   ///< XLL sync word present flag
+    int     xll_delay_nframes;  ///< Initial XLL decoding delay in frames
+    int     xll_sync_offset;    ///< Number of bytes offset to XLL sync
+
+    int     hd_stream_id;   ///< DTS-HD stream ID
+} DCAExssAsset;
+
+typedef struct DCAExssParser {
+    AVCodecContext  *avctx;
+    GetBitContext   gb;
+
+    int     exss_index;         ///< Extension substream index
+    int     exss_size_nbits;    ///< Number of bits for extension substream size
+    int     exss_size;          ///< Number of bytes of extension substream
+
+    int     static_fields_present;  ///< Per stream static fields presence flag
+    int     npresents;  ///< Number of defined audio presentations
+    int     nassets;    ///< Number of audio assets in extension substream
+
+    int     mix_metadata_enabled;   ///< Mixing metadata enable flag
+    int     nmixoutconfigs;         ///< Number of mixing configurations
+    int     nmixoutchs[4];          ///< Speaker layout mask for mixer output channels
+
+    DCAExssAsset   assets[1];    ///< Audio asset descriptors
+} DCAExssParser;
+
+int ff_dca_exss_parse(DCAExssParser *s, uint8_t *data, int size);
+
+#endif

libavcodec/dca_xll.c

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "dcadec.h"
+#include "dcadata.h"
+#include "dcamath.h"
+#include "dca_syncwords.h"
+#include "unary.h"
+
+static int get_linear(GetBitContext *gb, int n)
+{
+    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, 128);
+    return (v << k) | get_bits_long(gb, k);
+}
+
+static int get_rice(GetBitContext *gb, int k)
+{
+    unsigned int v = get_rice_un(gb, k);
+    return (v >> 1) ^ -(v & 1);
+}
+
+static void get_array(GetBitContext *gb, int32_t *array, int size, int n)
+{
+    int i;
+
+    for (i = 0; i < size; i++)
+        array[i] = get_bits(gb, n);
+}
+
+static void get_linear_array(GetBitContext *gb, int32_t *array, int size, int n)
+{
+    int i;
+
+    if (n == 0)
+        memset(array, 0, sizeof(*array) * size);
+    else for (i = 0; i < size; i++)
+        array[i] = get_linear(gb, n);
+}
+
+static void get_rice_array(GetBitContext *gb, int32_t *array, int size, int k)
+{
+    int i;
+
+    for (i = 0; i < size; i++)
+        array[i] = get_rice(gb, k);
+}
+
+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;
+        }
+
+        // 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;
+        }
+    }
+
+    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;
+
+    // Size of channel set sub-header
+    header_size = get_bits(&s->gb, 10) + 1;
+
+    // Check CRC
+    if ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&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 != 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;
+            }
+        }
+
+        // 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;
+                }
+            }
+
+            // 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 {
+            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;
+        }
+
+        // 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;
+        }
+
+        // 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;
+                }
+                if (k < 0)
+                    b->adapt_refl_coeff[i][j] = -(int)ff_dca_xll_refl_coeff[-k];
+                else
+                    b->adapt_refl_coeff[i][j] =  (int)ff_dca_xll_refl_coeff[ k];
+            }
+        }
+
+        // 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;
+            }
+
+            // 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;
+
+            // 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;
+        }
+    }
+
+    // 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;
+}
+
+static int chs_alloc_msb_band_data(DCAXllDecoder *s, DCAXllChSet *c)
+{
+    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);
+
+    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;
+        }
+    }
+
+    return 0;
+}
+
+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);
+
+    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;
+            }
+        } 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);
+            if (!c->seg_common && c->rice_code_flag[i]) {
+                // Unpack Hybrid Rice coding flag
+                // 0 - Rice code, 1 - Hybrid Rice code
+                if (get_bits1(&s->gb))
+                    // Unpack binary code length for isolated samples
+                    c->bitalloc_hybrid_linear[i] = get_bits(&s->gb, c->nabits) + 1;
+                else
+                    // 0 indicates no Hybrid Rice coding
+                    c->bitalloc_hybrid_linear[i] = 0;
+            } else {
+                // 0 indicates no Hybrid Rice coding
+                c->bitalloc_hybrid_linear[i] = 0;
+            }
+        }
+
+        // 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);
+
+                // Adjust for the linear code
+                if (!c->rice_code_flag[i] && c->bitalloc_part_a[i])
+                    c->bitalloc_part_a[i]++;
+
+                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;
+                }
+
+                // 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
+                        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]);
+            }
+        }
+    }
+
+    // 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]);
+            }
+        }
+    }
+
+    // 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;
+    }
+
+    return 0;
+}
+
+static void av_cold 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] -= 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] += buf[k - 1];
+        }
+    }
+
+    // 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);
+            }
+        }
+
+        // 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] * (1 << shift) + (lsb[n] << adj);
+            } else {
+                for (n = 0; n < nsamples; n++)
+                    msb[n] = msb[n] * (1 << shift);
+            }
+        }
+    }
+}
+
+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 ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&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++;
+                }
+                *navi_ptr++ = size;
+            }
+        }
+    }
+
+    // Byte align
+    // CRC16
+    skip_bits(&s->gb, -get_bits_count(&s->gb) & 7);
+    skip_bits(&s->gb, 16);
+
+    // Check CRC
+    if ((s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL))
+        && ff_dca_check_crc(&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;
+    }
+
+    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);
+                    }
+                    s->gb.index = navi_pos;
+                }
+                navi_ptr++;
+            }
+        }
+    }
+
+    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 + DCA_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);
+                }
+            }
+        }
+
+        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 (nchannels >= o->hier_ofs)
+            break;
+    }
+}
+
+// Clear all band data and replace non-residual encoded channels with lossy
+// counterparts
+static void av_cold 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] += clip23((mul16(src[n], scale_inv) + round) >> shift);
+        } else {
+            // No downmix scaling
+            for (n = 0; n < nsamples; n++)
+                dst[n] += (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;
+        break;
+    case 24:
+        avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
+        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);
+    }
+
+    shift = p->storage_bit_res - p->pcm_bit_res;
+    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] * (1 << shift));
+        } else {
+            int32_t *plane = (int32_t *)frame->extended_data[i];
+            for (k = 0; k < nsamples; k++)
+                plane[k] = clip23(samples[k] * (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);
+}

libavcodec/dca_xll.h

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_DCA_XLL_H
+#define AVCODEC_DCA_XLL_H
+
+#include "libavutil/common.h"
+#include "libavutil/mem.h"
+
+#include "avcodec.h"
+#include "internal.h"
+#include "get_bits.h"
+#include "dca.h"
+#include "dcadsp.h"
+#include "dca_exss.h"
+
+#define DCA_XLL_CHSETS_MAX              3
+#define DCA_XLL_CHANNELS_MAX            8
+#define DCA_XLL_BANDS_MAX               2
+#define DCA_XLL_ADAPT_PRED_ORDER_MAX    16
+#define DCA_XLL_DECI_HISTORY_MAX        8
+#define DCA_XLL_DMIX_SCALES_MAX         ((DCA_XLL_CHSETS_MAX - 1) * DCA_XLL_CHANNELS_MAX)
+#define DCA_XLL_DMIX_COEFFS_MAX         (DCA_XLL_DMIX_SCALES_MAX * DCA_XLL_CHANNELS_MAX)
+#define DCA_XLL_PBR_BUFFER_MAX          (240 << 10)
+#define DCA_XLL_SAMPLE_BUFFERS_MAX      3
+
+typedef struct DCAXllBand {
+    int     decor_enabled;                          ///< Pairwise channel decorrelation flag
+    int     orig_order[DCA_XLL_CHANNELS_MAX];       ///< Original channel order
+    int     decor_coeff[DCA_XLL_CHANNELS_MAX / 2];  ///< Pairwise channel coefficients
+
+    int     adapt_pred_order[DCA_XLL_CHANNELS_MAX]; ///< Adaptive predictor order
+    int     highest_pred_order;                     ///< Highest adaptive predictor order
+    int     fixed_pred_order[DCA_XLL_CHANNELS_MAX]; ///< Fixed predictor order
+    int     adapt_refl_coeff[DCA_XLL_CHANNELS_MAX][DCA_XLL_ADAPT_PRED_ORDER_MAX];   ///< Adaptive predictor reflection coefficients
+
+    int     dmix_embedded;  ///< Downmix performed by encoder in frequency band
+
+    int     lsb_section_size;                       ///< Size of LSB section in any segment
+    int     nscalablelsbs[DCA_XLL_CHANNELS_MAX];    ///< Number of bits to represent the samples in LSB part
+    int     bit_width_adjust[DCA_XLL_CHANNELS_MAX]; ///< Number of bits discarded by authoring
+
+    int32_t *msb_sample_buffer[DCA_XLL_CHANNELS_MAX];   ///< MSB sample buffer pointers
+    int32_t *lsb_sample_buffer[DCA_XLL_CHANNELS_MAX];   ///< LSB sample buffer pointers or NULL
+} DCAXllBand;
+
+typedef struct DCAXllChSet {
+    // Channel set header
+    int     nchannels;          ///< Number of channels in the channel set (N)
+    int     residual_encode;    ///< Residual encoding mask (0 - residual, 1 - full channel)
+    int     pcm_bit_res;        ///< PCM bit resolution (variable)
+    int     storage_bit_res;    ///< Storage bit resolution (16 or 24)
+    int     freq;               ///< Original sampling frequency (max. 96000 Hz)
+
+    int     primary_chset;          ///< Primary channel set flag
+    int     dmix_coeffs_present;    ///< Downmix coefficients present in stream
+    int     dmix_embedded;          ///< Downmix already performed by encoder
+    int     dmix_type;              ///< Primary channel set downmix type
+    int     hier_chset;             ///< Whether the channel set is part of a hierarchy
+    int     hier_ofs;               ///< Number of preceding channels in a hierarchy (M)
+    int     dmix_coeff[DCA_XLL_DMIX_COEFFS_MAX];       ///< Downmixing coefficients
+    int     dmix_scale[DCA_XLL_DMIX_SCALES_MAX];       ///< Downmixing scales
+    int     dmix_scale_inv[DCA_XLL_DMIX_SCALES_MAX];   ///< Inverse downmixing scales
+    int     ch_mask;                ///< Channel mask for set
+    int     ch_remap[DCA_XLL_CHANNELS_MAX];    ///< Channel to speaker map
+
+    int     nfreqbands; ///< Number of frequency bands (1 or 2)
+    int     nabits;     ///< Number of bits to read bit allocation coding parameter
+
+    DCAXllBand     bands[DCA_XLL_BANDS_MAX];   ///< Frequency bands
+
+    // Frequency band coding parameters
+    int     seg_common;                                     ///< Segment type
+    int     rice_code_flag[DCA_XLL_CHANNELS_MAX];           ///< Rice coding flag
+    int     bitalloc_hybrid_linear[DCA_XLL_CHANNELS_MAX];   ///< Binary code length for isolated samples
+    int     bitalloc_part_a[DCA_XLL_CHANNELS_MAX];          ///< Coding parameter for part A of segment
+    int     bitalloc_part_b[DCA_XLL_CHANNELS_MAX];          ///< Coding parameter for part B of segment
+    int     nsamples_part_a[DCA_XLL_CHANNELS_MAX];          ///< Number of samples in part A of segment
+
+    // Decimator history
+    DECLARE_ALIGNED(32, int32_t, deci_history)[DCA_XLL_CHANNELS_MAX][DCA_XLL_DECI_HISTORY_MAX]; ///< Decimator history for frequency band 1
+
+    // Sample buffers
+    unsigned int    sample_size[DCA_XLL_SAMPLE_BUFFERS_MAX];
+    int32_t         *sample_buffer[DCA_XLL_SAMPLE_BUFFERS_MAX];
+} DCAXllChSet;
+
+typedef struct DCAXllDecoder {
+    AVCodecContext  *avctx;
+    GetBitContext   gb;
+
+    int     frame_size;             ///< Number of bytes in a lossless frame
+    int     nchsets;                ///< Number of channels sets per frame
+    int     nframesegs;             ///< Number of segments per frame
+    int     nsegsamples_log2;       ///< log2(nsegsamples)
+    int     nsegsamples;            ///< Samples in segment per one frequency band
+    int     nframesamples_log2;     ///< log2(nframesamples)
+    int     nframesamples;          ///< Samples in frame per one frequency band
+    int     seg_size_nbits;         ///< Number of bits used to read segment size
+    int     band_crc_present;       ///< Presence of CRC16 within each frequency band
+    int     scalable_lsbs;          ///< MSB/LSB split flag
+    int     ch_mask_nbits;          ///< Number of bits used to read channel mask
+    int     fixed_lsb_width;        ///< Fixed LSB width
+
+    DCAXllChSet    chset[DCA_XLL_CHSETS_MAX]; ///< Channel sets
+
+    int             *navi;          ///< NAVI table
+    unsigned int    navi_size;
+
+    int     nfreqbands;     ///< Highest number of frequency bands
+    int     nchannels;      ///< Total number of channels in a hierarchy
+    int     nreschsets;     ///< Number of channel sets that have residual encoded channels
+    int     nactivechsets;  ///< Number of active channel sets to decode
+
+    int     hd_stream_id;   ///< Previous DTS-HD stream ID for detecting changes
+
+    uint8_t     *pbr_buffer;        ///< Peak bit rate (PBR) smoothing buffer
+    int         pbr_length;         ///< Length in bytes of data currently buffered
+    int         pbr_delay;          ///< Delay in frames before decoding buffered data
+
+    DCADSPContext   *dcadsp;
+
+    int     output_mask;
+    int32_t *output_samples[DCA_SPEAKER_COUNT];
+} DCAXllDecoder;
+
+int ff_dca_xll_parse(DCAXllDecoder *s, uint8_t *data, DCAExssAsset *asset);
+int ff_dca_xll_filter_frame(DCAXllDecoder *s, AVFrame *frame);
+av_cold void ff_dca_xll_flush(DCAXllDecoder *s);
+av_cold void ff_dca_xll_close(DCAXllDecoder *s);
+
+#endif

libavcodec/dcadec.c

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/opt.h"
+#include "libavutil/channel_layout.h"
+
+#include "dcadec.h"
+#include "dcamath.h"
+#include "dca_syncwords.h"
+#include "profiles.h"
+
+#define MIN_PACKET_SIZE     16
+#define MAX_PACKET_SIZE     0x104000
+
+int ff_dca_set_channel_layout(AVCodecContext *avctx, int *ch_remap, int dca_mask)
+{
+    static const uint8_t dca2wav_norm[28] = {
+         2,  0, 1, 9, 10,  3,  8,  4,  5,  9, 10, 6, 7, 12,
+        13, 14, 3, 6,  7, 11, 12, 14, 16, 15, 17, 8, 4,  5,
+    };
+
+    static const uint8_t dca2wav_wide[28] = {
+         2,  0, 1, 4,  5,  3,  8,  4,  5,  9, 10, 6, 7, 12,
+        13, 14, 3, 9, 10, 11, 12, 14, 16, 15, 17, 8, 4,  5,
+    };
+
+    int dca_ch, wav_ch, nchannels = 0;
+
+    if (avctx->request_channel_layout & AV_CH_LAYOUT_NATIVE) {
+        for (dca_ch = 0; dca_ch < DCA_SPEAKER_COUNT; dca_ch++)
+            if (dca_mask & (1U << dca_ch))
+                ch_remap[nchannels++] = dca_ch;
+        avctx->channel_layout = dca_mask;
+    } else {
+        int wav_mask = 0;
+        int wav_map[18];
+        const uint8_t *dca2wav;
+        if (dca_mask == DCA_SPEAKER_LAYOUT_7POINT0_WIDE ||
+            dca_mask == DCA_SPEAKER_LAYOUT_7POINT1_WIDE)
+            dca2wav = dca2wav_wide;
+        else
+            dca2wav = dca2wav_norm;
+        for (dca_ch = 0; dca_ch < 28; dca_ch++) {
+            if (dca_mask & (1 << dca_ch)) {
+                wav_ch = dca2wav[dca_ch];
+                if (!(wav_mask & (1 << wav_ch))) {
+                    wav_map[wav_ch] = dca_ch;
+                    wav_mask |= 1 << wav_ch;
+                }
+            }
+        }
+        for (wav_ch = 0; wav_ch < 18; wav_ch++)
+            if (wav_mask & (1 << wav_ch))
+                ch_remap[nchannels++] = wav_map[wav_ch];
+        avctx->channel_layout = wav_mask;
+    }
+
+    avctx->channels = nchannels;
+    return nchannels;
+}
+
+static uint16_t crc16(const uint8_t *data, int size)
+{
+    static const uint16_t crctab[16] = {
+        0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50a5, 0x60c6, 0x70e7,
+        0x8108, 0x9129, 0xa14a, 0xb16b, 0xc18c, 0xd1ad, 0xe1ce, 0xf1ef,
+    };
+
+    uint16_t res = 0xffff;
+    int i;
+
+    for (i = 0; i < size; i++) {
+        res = (res << 4) ^ crctab[(data[i] >> 4) ^ (res >> 12)];
+        res = (res << 4) ^ crctab[(data[i] & 15) ^ (res >> 12)];
+    }
+
+    return res;
+}
+
+int ff_dca_check_crc(GetBitContext *s, int p1, int p2)
+{
+    if (((p1 | p2) & 7) || p1 < 0 || p2 > s->size_in_bits || p2 - p1 < 16)
+        return -1;
+    if (crc16(s->buffer + p1 / 8, (p2 - p1) / 8))
+        return -1;
+    return 0;
+}
+
+void ff_dca_downmix_to_stereo_fixed(DCADSPContext *dcadsp, int32_t **samples,
+                                    int *coeff_l, int nsamples, int ch_mask)
+{
+    int pos, spkr, max_spkr = av_log2(ch_mask);
+    int *coeff_r = coeff_l + av_popcount(ch_mask);
+
+    av_assert0(DCA_HAS_STEREO(ch_mask));
+
+    // Scale left and right channels
+    pos = (ch_mask & DCA_SPEAKER_MASK_C);
+    dcadsp->dmix_scale(samples[DCA_SPEAKER_L], coeff_l[pos    ], nsamples);
+    dcadsp->dmix_scale(samples[DCA_SPEAKER_R], coeff_r[pos + 1], nsamples);
+
+    // Downmix remaining channels
+    for (spkr = 0; spkr <= max_spkr; spkr++) {
+        if (!(ch_mask & (1U << spkr)))
+            continue;
+
+        if (*coeff_l && spkr != DCA_SPEAKER_L)
+            dcadsp->dmix_add(samples[DCA_SPEAKER_L], samples[spkr],
+                             *coeff_l, nsamples);
+
+        if (*coeff_r && spkr != DCA_SPEAKER_R)
+            dcadsp->dmix_add(samples[DCA_SPEAKER_R], samples[spkr],
+                             *coeff_r, nsamples);
+
+        coeff_l++;
+        coeff_r++;
+    }
+}
+
+void ff_dca_downmix_to_stereo_float(AVFloatDSPContext *fdsp, float **samples,
+                                    int *coeff_l, int nsamples, int ch_mask)
+{
+    int pos, spkr, max_spkr = av_log2(ch_mask);
+    int *coeff_r = coeff_l + av_popcount(ch_mask);
+    const float scale = 1.0f / (1 << 15);
+
+    av_assert0(DCA_HAS_STEREO(ch_mask));
+
+    // Scale left and right channels
+    pos = (ch_mask & DCA_SPEAKER_MASK_C);
+    fdsp->vector_fmul_scalar(samples[DCA_SPEAKER_L], samples[DCA_SPEAKER_L],
+                             coeff_l[pos    ] * scale, nsamples);
+    fdsp->vector_fmul_scalar(samples[DCA_SPEAKER_R], samples[DCA_SPEAKER_R],
+                             coeff_r[pos + 1] * scale, nsamples);
+
+    // Downmix remaining channels
+    for (spkr = 0; spkr <= max_spkr; spkr++) {
+        if (!(ch_mask & (1U << spkr)))
+            continue;
+
+        if (*coeff_l && spkr != DCA_SPEAKER_L)
+            fdsp->vector_fmac_scalar(samples[DCA_SPEAKER_L], samples[spkr],
+                                     *coeff_l * scale, nsamples);
+
+        if (*coeff_r && spkr != DCA_SPEAKER_R)
+            fdsp->vector_fmac_scalar(samples[DCA_SPEAKER_R], samples[spkr],
+                                     *coeff_r * scale, nsamples);
+
+        coeff_l++;
+        coeff_r++;
+    }
+}
+
+static int convert_bitstream(const uint8_t *src, int src_size, uint8_t *dst, int max_size)
+{
+    switch (AV_RB32(src)) {
+    case DCA_SYNCWORD_CORE_BE:
+    case DCA_SYNCWORD_SUBSTREAM:
+        memcpy(dst, src, src_size);
+        return src_size;
+    case DCA_SYNCWORD_CORE_LE:
+    case DCA_SYNCWORD_CORE_14B_BE:
+    case DCA_SYNCWORD_CORE_14B_LE:
+        return avpriv_dca_convert_bitstream(src, src_size, dst, max_size);
+    default:
+        return AVERROR_INVALIDDATA;
+    }
+}
+
+static int dcadec_decode_frame(AVCodecContext *avctx, void *data,
+                               int *got_frame_ptr, AVPacket *avpkt)
+{
+    DCAContext *s = avctx->priv_data;
+    AVFrame *frame = data;
+    uint8_t *input = avpkt->data;
+    int input_size = avpkt->size;
+    int i, ret, prev_packet = s->packet;
+
+    if (input_size < MIN_PACKET_SIZE || input_size > MAX_PACKET_SIZE) {
+        av_log(avctx, AV_LOG_ERROR, "Invalid packet size\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    av_fast_malloc(&s->buffer, &s->buffer_size,
+                   FFALIGN(input_size, 4096) + DCA_BUFFER_PADDING_SIZE);
+    if (!s->buffer)
+        return AVERROR(ENOMEM);
+
+    for (i = 0, ret = AVERROR_INVALIDDATA; i < input_size - MIN_PACKET_SIZE + 1 && ret < 0; i++)
+        ret = convert_bitstream(input + i, input_size - i, s->buffer, s->buffer_size);
+
+    if (ret < 0)
+        return ret;
+
+    input      = s->buffer;
+    input_size = ret;
+
+    s->packet = 0;
+
+    // Parse backward compatible core sub-stream
+    if (AV_RB32(input) == DCA_SYNCWORD_CORE_BE) {
+        int frame_size;
+
+        if ((ret = ff_dca_core_parse(&s->core, input, input_size)) < 0) {
+            s->core_residual_valid = 0;
+            return ret;
+        }
+
+        s->packet |= DCA_PACKET_CORE;
+
+        // EXXS data must be aligned on 4-byte boundary
+        frame_size = FFALIGN(s->core.frame_size, 4);
+        if (input_size - 4 > frame_size) {
+            input      += frame_size;
+            input_size -= frame_size;
+        }
+    }
+
+    if (!s->core_only) {
+        DCAExssAsset *asset = NULL;
+
+        // Parse extension sub-stream (EXSS)
+        if (AV_RB32(input) == DCA_SYNCWORD_SUBSTREAM) {
+            if ((ret = ff_dca_exss_parse(&s->exss, input, input_size)) < 0) {
+                if (avctx->err_recognition & AV_EF_EXPLODE)
+                    return ret;
+            } else {
+                s->packet |= DCA_PACKET_EXSS;
+                asset = &s->exss.assets[0];
+            }
+        }
+
+        // Parse XLL component in EXSS
+        if (asset && (asset->extension_mask & DCA_EXSS_XLL)) {
+            if ((ret = ff_dca_xll_parse(&s->xll, input, asset)) < 0) {
+                // Conceal XLL synchronization error
+                if (ret == AVERROR(EAGAIN)
+                    && (prev_packet & DCA_PACKET_XLL)
+                    && (s->packet & DCA_PACKET_CORE))
+                    s->packet |= DCA_PACKET_XLL | DCA_PACKET_RECOVERY;
+                else if (ret == AVERROR(ENOMEM) || (avctx->err_recognition & AV_EF_EXPLODE))
+                    return ret;
+            } else {
+                s->packet |= DCA_PACKET_XLL;
+            }
+        }
+
+        // Parse core extensions in EXSS or backward compatible core sub-stream
+        if ((s->packet & DCA_PACKET_CORE)
+            && (ret = ff_dca_core_parse_exss(&s->core, input, asset)) < 0)
+            return ret;
+    }
+
+    // Filter the frame
+    if (s->packet & DCA_PACKET_XLL) {
+        if (s->packet & DCA_PACKET_CORE) {
+            int x96_synth = -1;
+
+            // Enable X96 synthesis if needed
+            if (s->xll.chset[0].freq == 96000 && s->core.sample_rate == 48000)
+                x96_synth = 1;
+
+            if ((ret = ff_dca_core_filter_fixed(&s->core, x96_synth)) < 0) {
+                s->core_residual_valid = 0;
+                return ret;
+            }
+
+            // Force lossy downmixed output on the first core frame filtered.
+            // This prevents audible clicks when seeking and is consistent with
+            // what reference decoder does when there are multiple channel sets.
+            if (!s->core_residual_valid) {
+                if (s->xll.nreschsets > 0 && s->xll.nchsets > 1)
+                    s->packet |= DCA_PACKET_RECOVERY;
+                s->core_residual_valid = 1;
+            }
+        }
+
+        if ((ret = ff_dca_xll_filter_frame(&s->xll, frame)) < 0) {
+            // Fall back to core unless hard error
+            if (!(s->packet & DCA_PACKET_CORE))
+                return ret;
+            if (ret != AVERROR_INVALIDDATA || (avctx->err_recognition & AV_EF_EXPLODE))
+                return ret;
+            if ((ret = ff_dca_core_filter_frame(&s->core, frame)) < 0) {
+                s->core_residual_valid = 0;
+                return ret;
+            }
+        }
+    } else if (s->packet & DCA_PACKET_CORE) {
+        if ((ret = ff_dca_core_filter_frame(&s->core, frame)) < 0) {
+            s->core_residual_valid = 0;
+            return ret;
+        }
+        s->core_residual_valid = !!(s->core.filter_mode & DCA_FILTER_MODE_FIXED);
+    } else {
+        return AVERROR_INVALIDDATA;
+    }
+
+    *got_frame_ptr = 1;
+
+    return avpkt->size;
+}
+
+static av_cold void dcadec_flush(AVCodecContext *avctx)
+{
+    DCAContext *s = avctx->priv_data;
+
+    ff_dca_core_flush(&s->core);
+    ff_dca_xll_flush(&s->xll);
+
+    s->core_residual_valid = 0;
+}
+
+static av_cold int dcadec_close(AVCodecContext *avctx)
+{
+    DCAContext *s = avctx->priv_data;
+
+    ff_dca_core_close(&s->core);
+    ff_dca_xll_close(&s->xll);
+
+    av_freep(&s->buffer);
+    s->buffer_size = 0;
+
+    return 0;
+}
+
+static av_cold int dcadec_init(AVCodecContext *avctx)
+{
+    DCAContext *s = avctx->priv_data;
+
+    s->avctx = avctx;
+    s->core.avctx = avctx;
+    s->exss.avctx = avctx;
+    s->xll.avctx = avctx;
+
+    if (ff_dca_core_init(&s->core) < 0)
+        return AVERROR(ENOMEM);
+
+    ff_dcadsp_init(&s->dcadsp);
+    s->core.dcadsp = &s->dcadsp;
+    s->xll.dcadsp = &s->dcadsp;
+
+    switch (avctx->request_channel_layout & ~AV_CH_LAYOUT_NATIVE) {
+    case 0:
+        s->request_channel_layout = 0;
+        break;
+    case AV_CH_LAYOUT_STEREO:
+    case AV_CH_LAYOUT_STEREO_DOWNMIX:
+        s->request_channel_layout = DCA_SPEAKER_LAYOUT_STEREO;
+        break;
+    case AV_CH_LAYOUT_5POINT0:
+        s->request_channel_layout = DCA_SPEAKER_LAYOUT_5POINT0;
+        break;
+    case AV_CH_LAYOUT_5POINT1:
+        s->request_channel_layout = DCA_SPEAKER_LAYOUT_5POINT1;
+        break;
+    default:
+        av_log(avctx, AV_LOG_WARNING, "Invalid request_channel_layout\n");
+        break;
+    }
+
+    avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
+    avctx->bits_per_raw_sample = 24;
+
+    return 0;
+}
+
+#define OFFSET(x) offsetof(DCAContext, x)
+#define PARAM AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM
+
+static const AVOption dcadec_options[] = {
+    { "core_only", "Decode core only without extensions", OFFSET(core_only), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, PARAM },
+    { NULL }
+};
+
+static const AVClass dcadec_class = {
+    .class_name = "DCA decoder",
+    .item_name  = av_default_item_name,
+    .option     = dcadec_options,
+    .version    = LIBAVUTIL_VERSION_INT,
+    .category   = AV_CLASS_CATEGORY_DECODER,
+};
+
+AVCodec ff_dca_decoder = {
+    .name           = "dca",
+    .long_name      = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),
+    .type           = AVMEDIA_TYPE_AUDIO,
+    .id             = AV_CODEC_ID_DTS,
+    .priv_data_size = sizeof(DCAContext),
+    .init           = dcadec_init,
+    .decode         = dcadec_decode_frame,
+    .close          = dcadec_close,
+    .flush          = dcadec_flush,
+    .capabilities   = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S32P,
+                                                      AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE },
+    .priv_class     = &dcadec_class,
+    .profiles       = NULL_IF_CONFIG_SMALL(ff_dca_profiles),
+    .caps_internal  = FF_CODEC_CAP_INIT_CLEANUP,
+};

libavcodec/dcadec.h

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_DCADEC_H
+#define AVCODEC_DCADEC_H
+
+#include "libavutil/common.h"
+#include "libavutil/float_dsp.h"
+
+#include "avcodec.h"
+#include "get_bits.h"
+#include "dca.h"
+#include "dcadsp.h"
+#include "dca_core.h"
+#include "dca_exss.h"
+#include "dca_xll.h"
+
+#define DCA_BUFFER_PADDING_SIZE     1024
+
+#define DCA_PACKET_CORE         0x01
+#define DCA_PACKET_EXSS         0x02
+#define DCA_PACKET_XLL          0x04
+#define DCA_PACKET_RECOVERY     0x08
+
+typedef struct DCAContext {
+    const AVClass   *class;       ///< class for AVOptions
+    AVCodecContext  *avctx;
+
+    DCACoreDecoder core;  ///< Core decoder context
+    DCAExssParser  exss;  ///< EXSS parser context
+    DCAXllDecoder  xll;   ///< XLL decoder context
+
+    DCADSPContext   dcadsp;
+
+    uint8_t         *buffer;    ///< Packet buffer
+    unsigned int    buffer_size;
+
+    int     packet; ///< Packet flags
+
+    int     core_residual_valid;    ///< Core valid for residual decoding
+
+    int     request_channel_layout; ///< Converted from avctx.request_channel_layout
+    int     core_only;              ///< Core only decoding flag
+} DCAContext;
+
+int ff_dca_set_channel_layout(AVCodecContext *avctx, int *ch_remap, int dca_mask);
+
+int ff_dca_check_crc(GetBitContext *s, int p1, int p2);
+
+void ff_dca_downmix_to_stereo_fixed(DCADSPContext *dcadsp, int32_t **samples,
+                                    int *coeff_l, int nsamples, int ch_mask);
+void ff_dca_downmix_to_stereo_float(AVFloatDSPContext *fdsp, float **samples,
+                                    int *coeff_l, int nsamples, int ch_mask);
+
+static inline int ff_dca_seek_bits(GetBitContext *s, int p)
+{
+    if (p < s->index || p > s->size_in_bits)
+        return -1;
+    s->index = p;
+    return 0;
+}
+
+#endif

libavcodec/dcadsp.c

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/mem.h"
+
+#include "dcadsp.h"
+#include "dcamath.h"
+
+static void decode_hf_c(int32_t **dst,
+                        const int32_t *vq_index,
+                        const int8_t hf_vq[1024][32],
+                        int32_t scale_factors[32][2],
+                        intptr_t sb_start, intptr_t sb_end,
+                        intptr_t ofs, intptr_t len)
+{
+    int i, j;
+
+    for (i = sb_start; i < sb_end; i++) {
+        const int8_t *coeff = hf_vq[vq_index[i]];
+        int32_t scale = scale_factors[i][0];
+        for (j = 0; j < len; j++)
+            dst[i][j + ofs] = clip23(coeff[j] * scale + (1 << 3) >> 4);
+    }
+}
+
+static void decode_joint_c(int32_t **dst, int32_t **src,
+                           const int32_t *scale_factors,
+                           intptr_t sb_start, intptr_t sb_end,
+                           intptr_t ofs, intptr_t len)
+{
+    int i, j;
+
+    for (i = sb_start; i < sb_end; i++) {
+        int32_t scale = scale_factors[i];
+        for (j = 0; j < len; j++)
+            dst[i][j + ofs] = clip23(mul17(src[i][j + ofs], scale));
+    }
+}
+
+static void lfe_fir_float_c(float *pcm_samples, int32_t *lfe_samples,
+                            const float *filter_coeff, intptr_t npcmblocks,
+                            int dec_select)
+{
+    // Select decimation factor
+    int factor = 64 << dec_select;
+    int ncoeffs = 8 >> dec_select;
+    int nlfesamples = npcmblocks >> (dec_select + 1);
+    int i, j, k;
+
+    for (i = 0; i < nlfesamples; i++) {
+        // One decimated sample generates 64 or 128 interpolated ones
+        for (j = 0; j < factor / 2; j++) {
+            float a = 0;
+            float b = 0;
+
+            for (k = 0; k < ncoeffs; k++) {
+                a += filter_coeff[      j * ncoeffs + k] * lfe_samples[-k];
+                b += filter_coeff[255 - j * ncoeffs - k] * lfe_samples[-k];
+            }
+
+            pcm_samples[             j] = a;
+            pcm_samples[factor / 2 + j] = b;
+        }
+
+        lfe_samples++;
+        pcm_samples += factor;
+    }
+}
+
+static void lfe_fir1_float_c(float *pcm_samples, int32_t *lfe_samples,
+                             const float *filter_coeff, intptr_t npcmblocks)
+{
+    lfe_fir_float_c(pcm_samples, lfe_samples, filter_coeff, npcmblocks, 0);
+}
+
+static void lfe_fir2_float_c(float *pcm_samples, int32_t *lfe_samples,
+                             const float *filter_coeff, intptr_t npcmblocks)
+{
+    lfe_fir_float_c(pcm_samples, lfe_samples, filter_coeff, npcmblocks, 1);
+}
+
+static void lfe_x96_float_c(float *dst, const float *src,
+                            float *hist, intptr_t len)
+{
+    float prev = *hist;
+    int i;
+
+    for (i = 0; i < len; i++) {
+        float a = 0.25f * src[i] + 0.75f * prev;
+        float b = 0.75f * src[i] + 0.25f * prev;
+        prev = src[i];
+        *dst++ = a;
+        *dst++ = b;
+    }
+
+    *hist = prev;
+}
+
+static void sub_qmf32_float_c(SynthFilterContext *synth,
+                              FFTContext *imdct,
+                              float *pcm_samples,
+                              int32_t **subband_samples_lo,
+                              int32_t **subband_samples_hi,
+                              float *hist1, int *offset, float *hist2,
+                              const float *filter_coeff, intptr_t npcmblocks,
+                              float scale)
+{
+    LOCAL_ALIGNED(32, float, input, [32]);
+    int i, j;
+
+    for (j = 0; j < npcmblocks; j++) {
+        // Load in one sample from each subband
+        for (i = 0; i < 32; i++) {
+            if ((i - 1) & 2)
+                input[i] = -subband_samples_lo[i][j];
+            else
+                input[i] =  subband_samples_lo[i][j];
+        }
+
+        // One subband sample generates 32 interpolated ones
+        synth->synth_filter_float(imdct, hist1, offset,
+                                  hist2, filter_coeff,
+                                  pcm_samples, input, scale);
+        pcm_samples += 32;
+    }
+}
+
+static void sub_qmf64_float_c(SynthFilterContext *synth,
+                              FFTContext *imdct,
+                              float *pcm_samples,
+                              int32_t **subband_samples_lo,
+                              int32_t **subband_samples_hi,
+                              float *hist1, int *offset, float *hist2,
+                              const float *filter_coeff, intptr_t npcmblocks,
+                              float scale)
+{
+    LOCAL_ALIGNED(32, float, input, [64]);
+    int i, j;
+
+    if (!subband_samples_hi)
+        memset(&input[32], 0, sizeof(input[0]) * 32);
+
+    for (j = 0; j < npcmblocks; j++) {
+        // Load in one sample from each subband
+        if (subband_samples_hi) {
+            // Full 64 subbands, first 32 are residual coded
+            for (i =  0; i < 32; i++) {
+                if ((i - 1) & 2)
+                    input[i] = -subband_samples_lo[i][j] - subband_samples_hi[i][j];
+                else
+                    input[i] =  subband_samples_lo[i][j] + subband_samples_hi[i][j];
+            }
+            for (i = 32; i < 64; i++) {
+                if ((i - 1) & 2)
+                    input[i] = -subband_samples_hi[i][j];
+                else
+                    input[i] =  subband_samples_hi[i][j];
+            }
+        } else {
+            // Only first 32 subbands
+            for (i =  0; i < 32; i++) {
+                if ((i - 1) & 2)
+                    input[i] = -subband_samples_lo[i][j];
+                else
+                    input[i] =  subband_samples_lo[i][j];
+            }
+        }
+
+        // One subband sample generates 64 interpolated ones
+        synth->synth_filter_float_64(imdct, hist1, offset,
+                                     hist2, filter_coeff,
+                                     pcm_samples, input, scale);
+        pcm_samples += 64;
+    }
+}
+
+static void lfe_fir_fixed_c(int32_t *pcm_samples, int32_t *lfe_samples,
+                            const int32_t *filter_coeff, intptr_t npcmblocks)
+{
+    // Select decimation factor
+    int nlfesamples = npcmblocks >> 1;
+    int i, j, k;
+
+    for (i = 0; i < nlfesamples; i++) {
+        // One decimated sample generates 64 interpolated ones
+        for (j = 0; j < 32; j++) {
+            int64_t a = 0;
+            int64_t b = 0;
+
+            for (k = 0; k < 8; k++) {
+                a += (int64_t)filter_coeff[      j * 8 + k] * lfe_samples[-k];
+                b += (int64_t)filter_coeff[255 - j * 8 - k] * lfe_samples[-k];
+            }
+
+            pcm_samples[     j] = clip23(norm23(a));
+            pcm_samples[32 + j] = clip23(norm23(b));
+        }
+
+        lfe_samples++;
+        pcm_samples += 64;
+    }
+}
+
+static void lfe_x96_fixed_c(int32_t *dst, const int32_t *src,
+                            int32_t *hist, intptr_t len)
+{
+    int32_t prev = *hist;
+    int i;
+
+    for (i = 0; i < len; i++) {
+        int64_t a = INT64_C(2097471) * src[i] + INT64_C(6291137) * prev;
+        int64_t b = INT64_C(6291137) * src[i] + INT64_C(2097471) * prev;
+        prev = src[i];
+        *dst++ = clip23(norm23(a));
+        *dst++ = clip23(norm23(b));
+    }
+
+    *hist = prev;
+}
+
+static void sub_qmf32_fixed_c(SynthFilterContext *synth,
+                              DCADCTContext *imdct,
+                              int32_t *pcm_samples,
+                              int32_t **subband_samples_lo,
+                              int32_t **subband_samples_hi,
+                              int32_t *hist1, int *offset, int32_t *hist2,
+                              const int32_t *filter_coeff, intptr_t npcmblocks)
+{
+    LOCAL_ALIGNED(32, int32_t, input, [32]);
+    int i, j;
+
+    for (j = 0; j < npcmblocks; j++) {
+        // Load in one sample from each subband
+        for (i = 0; i < 32; i++)
+            input[i] = subband_samples_lo[i][j];
+
+        // One subband sample generates 32 interpolated ones
+        synth->synth_filter_fixed(imdct, hist1, offset,
+                                  hist2, filter_coeff,
+                                  pcm_samples, input);
+        pcm_samples += 32;
+    }
+}
+
+static void sub_qmf64_fixed_c(SynthFilterContext *synth,
+                              DCADCTContext *imdct,
+                              int32_t *pcm_samples,
+                              int32_t **subband_samples_lo,
+                              int32_t **subband_samples_hi,
+                              int32_t *hist1, int *offset, int32_t *hist2,
+                              const int32_t *filter_coeff, intptr_t npcmblocks)
+{
+    LOCAL_ALIGNED(32, int32_t, input, [64]);
+    int i, j;
+
+    if (!subband_samples_hi)
+        memset(&input[32], 0, sizeof(input[0]) * 32);
+
+    for (j = 0; j < npcmblocks; j++) {
+        // Load in one sample from each subband
+        if (subband_samples_hi) {
+            // Full 64 subbands, first 32 are residual coded
+            for (i =  0; i < 32; i++)
+                input[i] = subband_samples_lo[i][j] + subband_samples_hi[i][j];
+            for (i = 32; i < 64; i++)
+                input[i] = subband_samples_hi[i][j];
+        } else {
+            // Only first 32 subbands
+            for (i =  0; i < 32; i++)
+                input[i] = subband_samples_lo[i][j];
+        }
+
+        // One subband sample generates 64 interpolated ones
+        synth->synth_filter_fixed_64(imdct, hist1, offset,
+                                     hist2, filter_coeff,
+                                     pcm_samples, input);
+        pcm_samples += 64;
+    }
+}
+
+static void decor_c(int32_t *dst, const int32_t *src, intptr_t coeff, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++)
+        dst[i] += src[i] * coeff + (1 << 2) >> 3;
+}
+
+static void dmix_sub_xch_c(int32_t *dst1, int32_t *dst2,
+                           const int32_t *src, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++) {
+        int32_t cs = mul23(src[i], 5931520 /* M_SQRT1_2 * (1 << 23) */);
+        dst1[i] -= cs;
+        dst2[i] -= cs;
+    }
+}
+
+static void dmix_sub_c(int32_t *dst, const int32_t *src, intptr_t coeff, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++)
+        dst[i] -= mul15(src[i], coeff);
+}
+
+static void dmix_add_c(int32_t *dst, const int32_t *src, intptr_t coeff, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++)
+        dst[i] += mul15(src[i], coeff);
+}
+
+static void dmix_scale_c(int32_t *dst, intptr_t scale, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++)
+        dst[i] = mul15(dst[i], scale);
+}
+
+static void dmix_scale_inv_c(int32_t *dst, intptr_t scale_inv, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++)
+        dst[i] = mul16(dst[i], scale_inv);
+}
+
+static void filter0(int32_t *dst, const int32_t *src, int32_t coeff, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++)
+        dst[i] -= mul22(src[i], coeff);
+}
+
+static void filter1(int32_t *dst, const int32_t *src, int32_t coeff, intptr_t len)
+{
+    int i;
+
+    for (i = 0; i < len; i++)
+        dst[i] -= mul23(src[i], coeff);
+}
+
+static void assemble_freq_bands_c(int32_t *dst, int32_t *src0, int32_t *src1,
+                                  const int32_t *coeff, intptr_t len)
+{
+    int i;
+
+    filter0(src0, src1, coeff[0], len);
+    filter0(src1, src0, coeff[1], len);
+    filter0(src0, src1, coeff[2], len);
+    filter0(src1, src0, coeff[3], len);
+
+    for (i = 0; i < 8; i++, src0--) {
+        filter1(src0, src1, coeff[i +  4], len);
+        filter1(src1, src0, coeff[i + 12], len);
+        filter1(src0, src1, coeff[i +  4], len);
+    }
+
+    for (i = 0; i < len; i++) {
+        *dst++ = *src1++;
+        *dst++ = *++src0;
+    }
+}
+
+av_cold void ff_dcadsp_init(DCADSPContext *s)
+{
+    s->decode_hf     = decode_hf_c;
+    s->decode_joint  = decode_joint_c;
+
+    s->lfe_fir_float[0] = lfe_fir1_float_c;
+    s->lfe_fir_float[1] = lfe_fir2_float_c;
+    s->lfe_x96_float    = lfe_x96_float_c;
+    s->sub_qmf_float[0] = sub_qmf32_float_c;
+    s->sub_qmf_float[1] = sub_qmf64_float_c;
+
+    s->lfe_fir_fixed    = lfe_fir_fixed_c;
+    s->lfe_x96_fixed    = lfe_x96_fixed_c;
+    s->sub_qmf_fixed[0] = sub_qmf32_fixed_c;
+    s->sub_qmf_fixed[1] = sub_qmf64_fixed_c;
+
+    s->decor   = decor_c;
+
+    s->dmix_sub_xch   = dmix_sub_xch_c;
+    s->dmix_sub       = dmix_sub_c;
+    s->dmix_add       = dmix_add_c;
+    s->dmix_scale     = dmix_scale_c;
+    s->dmix_scale_inv = dmix_scale_inv_c;
+
+    s->assemble_freq_bands = assemble_freq_bands_c;
+}

libavcodec/dcadsp.h

+/*
+ * Copyright (C) 2016 foo86
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_DCADSP_H
+#define AVCODEC_DCADSP_H
+
+#include "libavutil/common.h"
+
+#include "fft.h"
+#include "dcadct.h"
+#include "synth_filter.h"
+
+typedef struct DCADSPContext {
+    void (*decode_hf)(int32_t **dst,
+                      const int32_t *vq_index,
+                      const int8_t hf_vq[1024][32],
+                      int32_t scale_factors[32][2],
+                      intptr_t sb_start, intptr_t sb_end,
+                      intptr_t ofs, intptr_t len);
+
+    void (*decode_joint)(int32_t **dst, int32_t **src,
+                         const int32_t *scale_factors,
+                         intptr_t sb_start, intptr_t sb_end,
+                         intptr_t ofs, intptr_t len);
+
+    void (*lfe_fir_float[2])(float *pcm_samples, int32_t *lfe_samples,
+                             const float *filter_coeff, intptr_t npcmblocks);
+
+    void (*lfe_x96_float)(float *dst, const float *src,
+                          float *hist, intptr_t len);
+
+    void (*sub_qmf_float[2])(SynthFilterContext *synth,
+                             FFTContext *imdct,
+                             float *pcm_samples,
+                             int32_t **subband_samples_lo,
+                             int32_t **subband_samples_hi,
+                             float *hist1, int *offset, float *hist2,
+                             const float *filter_coeff, intptr_t npcmblocks,
+                             float scale);
+
+    void (*lfe_fir_fixed)(int32_t *pcm_samples, int32_t *lfe_samples,
+                          const int32_t *filter_coeff, intptr_t npcmblocks);
+
+    void (*lfe_x96_fixed)(int32_t *dst, const int32_t *src,
+                          int32_t *hist, intptr_t len);
+
+    void (*sub_qmf_fixed[2])(SynthFilterContext *synth,
+                             DCADCTContext *imdct,
+                             int32_t *pcm_samples,
+                             int32_t **subband_samples_lo,
+                             int32_t **subband_samples_hi,
+                             int32_t *hist1, int *offset, int32_t *hist2,
+                             const int32_t *filter_coeff, intptr_t npcmblocks);
+
+    void (*decor)(int32_t *dst, const int32_t *src, intptr_t coeff, intptr_t len);
+
+    void (*dmix_sub_xch)(int32_t *dst1, int32_t *dst2,
+                         const int32_t *src, intptr_t len);
+
+    void (*dmix_sub)(int32_t *dst, const int32_t *src, intptr_t coeff, intptr_t len);
+
+    void (*dmix_add)(int32_t *dst, const int32_t *src, intptr_t coeff, intptr_t len);
+
+    void (*dmix_scale)(int32_t *dst, intptr_t scale, intptr_t len);
+
+    void (*dmix_scale_inv)(int32_t *dst, intptr_t scale_inv, intptr_t len);
+
+    void (*assemble_freq_bands)(int32_t *dst, int32_t *src0, int32_t *src1,
+                                const int32_t *coeff, intptr_t len);
+} DCADSPContext;
+
+av_cold void ff_dcadsp_init(DCADSPContext *s);
+
+#endif

libavcodec/version.h

@@ -30,7 +30,7 @@
 
 #define LIBAVCODEC_VERSION_MAJOR  57
 #define LIBAVCODEC_VERSION_MINOR  24
-#define LIBAVCODEC_VERSION_MICRO 100
+#define LIBAVCODEC_VERSION_MICRO 101
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
                                                LIBAVCODEC_VERSION_MINOR, \

libavcodec/x86/Makefile

@@ -44,7 +44,7 @@ OBJS-$(CONFIG_ADPCM_G722_ENCODER)      += x86/g722dsp_init.o
 OBJS-$(CONFIG_ALAC_DECODER)            += x86/alacdsp_init.o
 OBJS-$(CONFIG_APNG_DECODER)            += x86/pngdsp_init.o
 OBJS-$(CONFIG_CAVS_DECODER)            += x86/cavsdsp.o
-#OBJS-$(CONFIG_DCA_DECODER)             += x86/synth_filter_init.o
+OBJS-$(CONFIG_DCA_DECODER)             += x86/synth_filter_init.o
 OBJS-$(CONFIG_DNXHD_ENCODER)           += x86/dnxhdenc_init.o
 OBJS-$(CONFIG_HEVC_DECODER)            += x86/hevcdsp_init.o
 OBJS-$(CONFIG_JPEG2000_DECODER)        += x86/jpeg2000dsp_init.o
@@ -132,7 +132,7 @@ YASM-OBJS-$(CONFIG_ADPCM_G722_DECODER) += x86/g722dsp.o
 YASM-OBJS-$(CONFIG_ADPCM_G722_ENCODER) += x86/g722dsp.o
 YASM-OBJS-$(CONFIG_ALAC_DECODER)       += x86/alacdsp.o
 YASM-OBJS-$(CONFIG_APNG_DECODER)       += x86/pngdsp.o
-#YASM-OBJS-$(CONFIG_DCA_DECODER)        += x86/synth_filter.o
+YASM-OBJS-$(CONFIG_DCA_DECODER)        += x86/synth_filter.o
 YASM-OBJS-$(CONFIG_DIRAC_DECODER)      += x86/diracdsp_mmx.o x86/diracdsp_yasm.o \
                                           x86/dwt_yasm.o
 YASM-OBJS-$(CONFIG_DNXHD_ENCODER)      += x86/dnxhdenc.o

tests/checkasm/Makefile

 # libavcodec tests
 AVCODECOBJS-$(CONFIG_ALAC_DECODER) += alacdsp.o
 AVCODECOBJS-$(CONFIG_BSWAPDSP) += bswapdsp.o
-#AVCODECOBJS-$(CONFIG_DCA_DECODER) += synth_filter.o
+AVCODECOBJS-$(CONFIG_DCA_DECODER) += synth_filter.o
 AVCODECOBJS-$(CONFIG_FLACDSP)  += flacdsp.o
 AVCODECOBJS-$(CONFIG_FMTCONVERT)   += fmtconvert.o
 AVCODECOBJS-$(CONFIG_H264PRED) += h264pred.o

tests/checkasm/checkasm.c

@@ -71,9 +71,9 @@ static const struct {
     #if CONFIG_BSWAPDSP
         { "bswapdsp", checkasm_check_bswapdsp },
     #endif
-/*    #if CONFIG_DCA_DECODER
+    #if CONFIG_DCA_DECODER
         { "synth_filter", checkasm_check_synth_filter },
-    #endif*/
+    #endif
     #if CONFIG_FLACDSP
         { "flacdsp", checkasm_check_flacdsp },
     #endif

tests/fate/acodec.mak

@@ -99,14 +99,14 @@ FATE_ACODEC-$(call ENCDEC, ALAC, MOV) += fate-acodec-alac
 fate-acodec-alac: FMT = mov
 fate-acodec-alac: CODEC = alac -compression_level 1
 
-#FATE_ACODEC-$(call ENCDEC, DCA, DTS) += fate-acodec-dca
+FATE_ACODEC-$(call ENCDEC, DCA, DTS) += fate-acodec-dca
 fate-acodec-dca: tests/data/asynth-44100-2.wav
 fate-acodec-dca: SRC = tests/data/asynth-44100-2.wav
 fate-acodec-dca: CMD = md5 -i $(TARGET_PATH)/$(SRC) -c:a dca -strict -2 -f dts -flags +bitexact
 fate-acodec-dca: CMP = oneline
 fate-acodec-dca: REF = 7ffdefdf47069289990755c79387cc90
 
-#FATE_ACODEC-$(call ENCDEC, DCA, WAV) += fate-acodec-dca2
+FATE_ACODEC-$(call ENCDEC, DCA, WAV) += fate-acodec-dca2
 fate-acodec-dca2: CMD = enc_dec_pcm dts wav s16le $(SRC) -c:a dca -strict -2 -flags +bitexact
 fate-acodec-dca2: REF = $(SRC)
 fate-acodec-dca2: CMP = stddev

tests/fate/audio.mak

@@ -21,7 +21,7 @@ fate-dca-core: CMD = pcm -i $(TARGET_SAMPLES)/dts/dts.ts
 fate-dca-core: CMP = oneoff
 fate-dca-core: REF = $(SAMPLES)/dts/dts.pcm
 
-#FATE_SAMPLES_AUDIO-$(CONFIG_DCA_DECODER) += $(FATE_DCA-yes)
+FATE_SAMPLES_AUDIO-$(CONFIG_DCA_DECODER) += $(FATE_DCA-yes)
 fate-dca: $(FATE_DCA-yes)
 
 FATE_SAMPLES_AUDIO-$(call DEMDEC, DSICIN, DSICINAUDIO) += fate-delphine-cin-audio
@@ -31,7 +31,7 @@ FATE_SAMPLES_AUDIO-$(call DEMDEC, DSS, DSS_SP) += fate-dss-lp fate-dss-sp
 fate-dss-lp: CMD = framecrc -i $(TARGET_SAMPLES)/dss/lp.dss -frames 30
 fate-dss-sp: CMD = framecrc -i $(TARGET_SAMPLES)/dss/sp.dss -frames 30
 
-#FATE_SAMPLES_AUDIO-$(call DEMDEC, DTS, DCA) += fate-dts_es
+FATE_SAMPLES_AUDIO-$(call DEMDEC, DTS, DCA) += fate-dts_es
 fate-dts_es: CMD = pcm -i $(TARGET_SAMPLES)/dts/dts_es.dts
 fate-dts_es: CMP = oneoff
 fate-dts_es: REF = $(SAMPLES)/dts/dts_es_2.pcm