tak: demuxer, parser, and decoder

Signed-off-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Justin Ruggles <justin.ruggles@gmail.com>

1 files changed, 929 insertions, 0 deletions

diff --git a/libavcodec/takdec.c b/libavcodec/takdec.c
new file mode 100644
index 0000000000..87fcf832ba
--- /dev/null
+++ b/libavcodec/takdec.c
@@ -0,0 +1,929 @@
+/*
+ * TAK decoder
+ * Copyright (c) 2012 Paul B Mahol
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * TAK (Tom's lossless Audio Kompressor) decoder
+ * @author Paul B Mahol
+ */
+
+#include "libavutil/samplefmt.h"
+#include "tak.h"
+#include "avcodec.h"
+#include "dsputil.h"
+#include "internal.h"
+#include "unary.h"
+
+#define MAX_SUBFRAMES     8                         // max number of subframes per channel
+#define MAX_PREDICTORS  256
+
+typedef struct MCDParam {
+    int8_t present;                                 // decorrelation parameter availability for this channel
+    int8_t index;                                   // index into array of decorrelation types
+    int8_t chan1;
+    int8_t chan2;
+} MCDParam;
+
+typedef struct TAKDecContext {
+    AVCodecContext *avctx;                          // parent AVCodecContext
+    AVFrame         frame;                          // AVFrame for decoded output
+    DSPContext      dsp;
+    TAKStreamInfo   ti;
+    GetBitContext   gb;                             // bitstream reader initialized to start at the current frame
+
+    int             uval;
+    int             nb_samples;                     // number of samples in the current frame
+    uint8_t        *decode_buffer;
+    unsigned int    decode_buffer_size;
+    int32_t        *decoded[TAK_MAX_CHANNELS];      // decoded samples for each channel
+
+    int8_t          lpc_mode[TAK_MAX_CHANNELS];
+    int8_t          sample_shift[TAK_MAX_CHANNELS]; // shift applied to every sample in the channel
+    int             subframe_scale;
+
+    int8_t          dmode;                          // channel decorrelation type in the current frame
+
+    MCDParam        mcdparams[TAK_MAX_CHANNELS];    // multichannel decorrelation parameters
+
+    int16_t        *residues;
+    unsigned int    residues_buf_size;
+} TAKDecContext;
+
+static const int8_t mc_dmodes[] = { 1, 3, 4, 6, };
+
+static const uint16_t predictor_sizes[] = {
+    4, 8, 12, 16, 24, 32, 48, 64, 80, 96, 128, 160, 192, 224, 256, 0,
+};
+
+static const struct CParam {
+    int init;
+    int escape;
+    int scale;
+    int aescape;
+    int bias;
+} xcodes[50] = {
+    { 0x01, 0x0000001, 0x0000001, 0x0000003, 0x0000008 },
+    { 0x02, 0x0000003, 0x0000001, 0x0000007, 0x0000006 },
+    { 0x03, 0x0000005, 0x0000002, 0x000000E, 0x000000D },
+    { 0x03, 0x0000003, 0x0000003, 0x000000D, 0x0000018 },
+    { 0x04, 0x000000B, 0x0000004, 0x000001C, 0x0000019 },
+    { 0x04, 0x0000006, 0x0000006, 0x000001A, 0x0000030 },
+    { 0x05, 0x0000016, 0x0000008, 0x0000038, 0x0000032 },
+    { 0x05, 0x000000C, 0x000000C, 0x0000034, 0x0000060 },
+    { 0x06, 0x000002C, 0x0000010, 0x0000070, 0x0000064 },
+    { 0x06, 0x0000018, 0x0000018, 0x0000068, 0x00000C0 },
+    { 0x07, 0x0000058, 0x0000020, 0x00000E0, 0x00000C8 },
+    { 0x07, 0x0000030, 0x0000030, 0x00000D0, 0x0000180 },
+    { 0x08, 0x00000B0, 0x0000040, 0x00001C0, 0x0000190 },
+    { 0x08, 0x0000060, 0x0000060, 0x00001A0, 0x0000300 },
+    { 0x09, 0x0000160, 0x0000080, 0x0000380, 0x0000320 },
+    { 0x09, 0x00000C0, 0x00000C0, 0x0000340, 0x0000600 },
+    { 0x0A, 0x00002C0, 0x0000100, 0x0000700, 0x0000640 },
+    { 0x0A, 0x0000180, 0x0000180, 0x0000680, 0x0000C00 },
+    { 0x0B, 0x0000580, 0x0000200, 0x0000E00, 0x0000C80 },
+    { 0x0B, 0x0000300, 0x0000300, 0x0000D00, 0x0001800 },
+    { 0x0C, 0x0000B00, 0x0000400, 0x0001C00, 0x0001900 },
+    { 0x0C, 0x0000600, 0x0000600, 0x0001A00, 0x0003000 },
+    { 0x0D, 0x0001600, 0x0000800, 0x0003800, 0x0003200 },
+    { 0x0D, 0x0000C00, 0x0000C00, 0x0003400, 0x0006000 },
+    { 0x0E, 0x0002C00, 0x0001000, 0x0007000, 0x0006400 },
+    { 0x0E, 0x0001800, 0x0001800, 0x0006800, 0x000C000 },
+    { 0x0F, 0x0005800, 0x0002000, 0x000E000, 0x000C800 },
+    { 0x0F, 0x0003000, 0x0003000, 0x000D000, 0x0018000 },
+    { 0x10, 0x000B000, 0x0004000, 0x001C000, 0x0019000 },
+    { 0x10, 0x0006000, 0x0006000, 0x001A000, 0x0030000 },
+    { 0x11, 0x0016000, 0x0008000, 0x0038000, 0x0032000 },
+    { 0x11, 0x000C000, 0x000C000, 0x0034000, 0x0060000 },
+    { 0x12, 0x002C000, 0x0010000, 0x0070000, 0x0064000 },
+    { 0x12, 0x0018000, 0x0018000, 0x0068000, 0x00C0000 },
+    { 0x13, 0x0058000, 0x0020000, 0x00E0000, 0x00C8000 },
+    { 0x13, 0x0030000, 0x0030000, 0x00D0000, 0x0180000 },
+    { 0x14, 0x00B0000, 0x0040000, 0x01C0000, 0x0190000 },
+    { 0x14, 0x0060000, 0x0060000, 0x01A0000, 0x0300000 },
+    { 0x15, 0x0160000, 0x0080000, 0x0380000, 0x0320000 },
+    { 0x15, 0x00C0000, 0x00C0000, 0x0340000, 0x0600000 },
+    { 0x16, 0x02C0000, 0x0100000, 0x0700000, 0x0640000 },
+    { 0x16, 0x0180000, 0x0180000, 0x0680000, 0x0C00000 },
+    { 0x17, 0x0580000, 0x0200000, 0x0E00000, 0x0C80000 },
+    { 0x17, 0x0300000, 0x0300000, 0x0D00000, 0x1800000 },
+    { 0x18, 0x0B00000, 0x0400000, 0x1C00000, 0x1900000 },
+    { 0x18, 0x0600000, 0x0600000, 0x1A00000, 0x3000000 },
+    { 0x19, 0x1600000, 0x0800000, 0x3800000, 0x3200000 },
+    { 0x19, 0x0C00000, 0x0C00000, 0x3400000, 0x6000000 },
+    { 0x1A, 0x2C00000, 0x1000000, 0x7000000, 0x6400000 },
+    { 0x1A, 0x1800000, 0x1800000, 0x6800000, 0xC000000 },
+};
+
+static av_cold void tak_init_static_data(AVCodec *codec)
+{
+    ff_tak_init_crc();
+}
+
+static int set_bps_params(AVCodecContext *avctx)
+{
+    switch (avctx->bits_per_coded_sample) {
+    case 8:
+        avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
+        break;
+    case 16:
+        avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
+        break;
+    case 24:
+        avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
+        break;
+    default:
+        av_log(avctx, AV_LOG_ERROR, "unsupported bits per sample: %d\n",
+               avctx->bits_per_coded_sample);
+        return AVERROR_INVALIDDATA;
+    }
+    avctx->bits_per_raw_sample = avctx->bits_per_coded_sample;
+
+    return 0;
+}
+
+static void set_sample_rate_params(AVCodecContext *avctx)
+{
+    TAKDecContext *s  = avctx->priv_data;
+    int shift         = 3 - (avctx->sample_rate / 11025);
+    shift             = FFMAX(0, shift);
+    s->uval           = FFALIGN(avctx->sample_rate + 511 >> 9, 4) << shift;
+    s->subframe_scale = FFALIGN(avctx->sample_rate + 511 >> 9, 4) << 1;
+}
+
+static av_cold int tak_decode_init(AVCodecContext *avctx)
+{
+    TAKDecContext *s = avctx->priv_data;
+
+    ff_dsputil_init(&s->dsp, avctx);
+
+    s->avctx = avctx;
+    avcodec_get_frame_defaults(&s->frame);
+    avctx->coded_frame = &s->frame;
+
+    set_sample_rate_params(avctx);
+
+    return set_bps_params(avctx);
+}
+
+static void decode_lpc(int32_t *coeffs, int mode, int length)
+{
+    int i;
+
+    if (length < 2)
+        return;
+
+    if (mode == 1) {
+        int a1 = *coeffs++;
+        for (i = 0; i < length - 1 >> 1; i++) {
+            *coeffs   += a1;
+            coeffs[1] += *coeffs;
+            a1         = coeffs[1];
+            coeffs    += 2;
+        }
+        if (length - 1 & 1)
+            *coeffs += a1;
+    } else if (mode == 2) {
+        int a1    = coeffs[1];
+        int a2    = a1 + *coeffs;
+        coeffs[1] = a2;
+        if (length > 2) {
+            coeffs += 2;
+            for (i = 0; i < length - 2 >> 1; i++) {
+                int a3    = *coeffs + a1;
+                int a4    = a3 + a2;
+                *coeffs   = a4;
+                a1        = coeffs[1] + a3;
+                a2        = a1 + a4;
+                coeffs[1] = a2;
+                coeffs   += 2;
+            }
+            if (length & 1)
+                *coeffs += a1 + a2;
+        }
+    } else if (mode == 3) {
+        int a1    = coeffs[1];
+        int a2    = a1 + *coeffs;
+        coeffs[1] = a2;
+        if (length > 2) {
+            int a3  = coeffs[2];
+            int a4  = a3 + a1;
+            int a5  = a4 + a2;
+            coeffs += 3;
+            for (i = 0; i < length - 3; i++) {
+                a3     += *coeffs;
+                a4     += a3;
+                a5     += a4;
+                *coeffs = a5;
+                coeffs++;
+            }
+        }
+    }
+}
+
+static int decode_segment(GetBitContext *gb, int mode, int32_t *decoded,
+                          int len)
+{
+    struct CParam code;
+    int i;
+
+    if (!mode) {
+        memset(decoded, 0, len * sizeof(*decoded));
+        return 0;
+    }
+
+    if (mode > FF_ARRAY_ELEMS(xcodes))
+        return AVERROR_INVALIDDATA;
+    code = xcodes[mode - 1];
+
+    for (i = 0; i < len; i++) {
+        int x = get_bits_long(gb, code.init);
+        if (x >= code.escape && get_bits1(gb)) {
+            x |= 1 << code.init;
+            if (x >= code.aescape) {
+                int scale = get_unary(gb, 1, 9);
+                if (scale == 9) {
+                    int scale_bits = get_bits(gb, 3);
+                    if (scale_bits > 0) {
+                        if (scale_bits == 7) {
+                            scale_bits += get_bits(gb, 5);
+                            if (scale_bits > 29)
+                                return AVERROR_INVALIDDATA;
+                        }
+                        scale = get_bits_long(gb, scale_bits) + 1;
+                        x    += code.scale * scale;
+                    }
+                    x += code.bias;
+                } else
+                    x += code.scale * scale - code.escape;
+            } else
+                x -= code.escape;
+        }
+        decoded[i] = (x >> 1) ^ -(x & 1);
+    }
+
+    return 0;
+}
+
+static int decode_residues(TAKDecContext *s, int32_t *decoded, int length)
+{
+    GetBitContext *gb = &s->gb;
+    int i, mode, ret;
+
+    if (length > s->nb_samples)
+        return AVERROR_INVALIDDATA;
+
+    if (get_bits1(gb)) {
+        int wlength, rval;
+        int coding_mode[128];
+
+        wlength = length / s->uval;
+
+        rval = length - (wlength * s->uval);
+
+        if (rval < s->uval / 2)
+            rval += s->uval;
+        else
+            wlength++;
+
+        if (wlength <= 1 || wlength > 128)
+            return AVERROR_INVALIDDATA;
+
+        coding_mode[0] = mode = get_bits(gb, 6);
+
+        for (i = 1; i < wlength; i++) {
+            int c = get_unary(gb, 1, 6);
+
+            switch (c) {
+            case 6:
+                mode = get_bits(gb, 6);
+                break;
+            case 5:
+            case 4:
+            case 3: {
+                /* mode += sign ? (1 - c) : (c - 1) */
+                int sign = get_bits1(gb);
+                mode    += (-sign ^ (c - 1)) + sign;
+                break;
+            }
+            case 2:
+                mode++;
+                break;
+            case 1:
+                mode--;
+                break;
+            }
+            coding_mode[i] = mode;
+        }
+
+        i = 0;
+        while (i < wlength) {
+            int len = 0;
+
+            mode = coding_mode[i];
+            do {
+                if (i >= wlength - 1)
+                    len += rval;
+                else
+                    len += s->uval;
+                i++;
+
+                if (i == wlength)
+                    break;
+            } while (coding_mode[i] == mode);
+
+            if ((ret = decode_segment(gb, mode, decoded, len)) < 0)
+                return ret;
+            decoded += len;
+        }
+    } else {
+        mode = get_bits(gb, 6);
+        if ((ret = decode_segment(gb, mode, decoded, length)) < 0)
+            return ret;
+    }
+
+    return 0;
+}
+
+static int get_bits_esc4(GetBitContext *gb)
+{
+    if (get_bits1(gb))
+        return get_bits(gb, 4) + 1;
+    else
+        return 0;
+}
+
+static void decode_filter_coeffs(TAKDecContext *s, int filter_order, int size,
+                                 int filter_quant, int16_t *filter)
+{
+    GetBitContext *gb = &s->gb;
+    int i, j, a, b;
+    int filter_tmp[MAX_PREDICTORS];
+    int16_t predictors[MAX_PREDICTORS];
+
+    predictors[0] = get_sbits(gb, 10);
+    predictors[1] = get_sbits(gb, 10);
+    predictors[2] = get_sbits(gb, size) << (10 - size);
+    predictors[3] = get_sbits(gb, size) << (10 - size);
+    if (filter_order > 4) {
+        int av_uninit(code_size);
+        int code_size_base = size - get_bits1(gb);
+
+        for (i = 4; i < filter_order; i++) {
+            if (!(i & 3))
+                code_size = code_size_base - get_bits(gb, 2);
+            predictors[i] = get_sbits(gb, code_size) << (10 - size);
+        }
+    }
+
+    filter_tmp[0] = predictors[0] << 6;
+    for (i = 1; i < filter_order; i++) {
+        int *p1 = &filter_tmp[0];
+        int *p2 = &filter_tmp[i - 1];
+
+        for (j = 0; j < (i + 1) / 2; j++) {
+            int tmp = *p1 + (predictors[i] * *p2 + 256 >> 9);
+            *p2     = *p2 + (predictors[i] * *p1 + 256 >> 9);
+            *p1     = tmp;
+            p1++;
+            p2--;
+        }
+
+        filter_tmp[i] = predictors[i] << 6;
+    }
+
+    a = 1 << (32 - (15 - filter_quant));
+    b = 1 << ((15 - filter_quant) - 1);
+    for (i = 0, j = filter_order - 1; i < filter_order / 2; i++, j--) {
+        filter[j] = a - ((filter_tmp[i] + b) >> (15 - filter_quant));
+        filter[i] = a - ((filter_tmp[j] + b) >> (15 - filter_quant));
+    }
+}
+
+static int decode_subframe(TAKDecContext *s, int32_t *decoded,
+                           int subframe_size, int prev_subframe_size)
+{
+    LOCAL_ALIGNED_16(int16_t, filter, [MAX_PREDICTORS]) = { 0, };
+    GetBitContext *gb = &s->gb;
+    int i, ret;
+    int dshift, size, filter_quant, filter_order;
+
+    if (!get_bits1(gb))
+        return decode_residues(s, decoded, subframe_size);
+
+    filter_order = predictor_sizes[get_bits(gb, 4)];
+
+    if (prev_subframe_size > 0 && get_bits1(gb)) {
+        if (filter_order > prev_subframe_size)
+            return AVERROR_INVALIDDATA;
+
+        decoded       -= filter_order;
+        subframe_size += filter_order;
+
+        if (filter_order > subframe_size)
+            return AVERROR_INVALIDDATA;
+    } else {
+        int lpc_mode;
+
+        if (filter_order > subframe_size)
+            return AVERROR_INVALIDDATA;
+
+        lpc_mode = get_bits(gb, 2);
+        if (lpc_mode > 2)
+            return AVERROR_INVALIDDATA;
+
+        if ((ret = decode_residues(s, decoded, filter_order)) < 0)
+            return ret;
+
+        if (lpc_mode)
+            decode_lpc(decoded, lpc_mode, filter_order);
+    }
+
+    dshift = get_bits_esc4(gb);
+    size   = get_bits1(gb) + 6;
+
+    filter_quant = 10;
+    if (get_bits1(gb)) {
+        filter_quant -= get_bits(gb, 3) + 1;
+        if (filter_quant < 3)
+            return AVERROR_INVALIDDATA;
+    }
+
+    decode_filter_coeffs(s, filter_order, size, filter_quant, filter);
+
+    if ((ret = decode_residues(s, &decoded[filter_order],
+                               subframe_size - filter_order)) < 0)
+        return ret;
+
+    av_fast_malloc(&s->residues, &s->residues_buf_size,
+                   FFALIGN(subframe_size + 16, 16) * sizeof(*s->residues));
+    if (!s->residues)
+        return AVERROR(ENOMEM);
+    memset(s->residues, 0, s->residues_buf_size);
+
+    for (i = 0; i < filter_order; i++)
+        s->residues[i] = *decoded++ >> dshift;
+
+    for (i = 0; i < subframe_size - filter_order; i++) {
+        int v = 1 << (filter_quant - 1);
+
+        v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
+                                        FFALIGN(filter_order, 16));
+
+        v = (av_clip(v >> filter_quant, -8192, 8191) << dshift) - *decoded;
+        *decoded++ = v;
+        s->residues[filter_order + i] = v >> dshift;
+    }
+
+    emms_c();
+
+    return 0;
+}
+
+static int decode_channel(TAKDecContext *s, int chan)
+{
+    AVCodecContext *avctx = s->avctx;
+    GetBitContext *gb     = &s->gb;
+    int32_t *decoded      = s->decoded[chan];
+    int left              = s->nb_samples - 1;
+    int i, prev, ret, nb_subframes;
+    int subframe_len[MAX_SUBFRAMES];
+
+    s->sample_shift[chan] = get_bits_esc4(gb);
+    if (s->sample_shift[chan] >= avctx->bits_per_coded_sample)
+        return AVERROR_INVALIDDATA;
+
+    /* NOTE: TAK 2.2.0 appears to set the sample value to 0 if
+     *       bits_per_coded_sample - sample_shift is 1, but this produces
+     *       non-bit-exact output. Reading the 1 bit using get_sbits() instead
+     *       of skipping it produces bit-exact output. This has been reported
+     *       to the TAK author. */
+    *decoded++        = get_sbits(gb,
+                                  avctx->bits_per_coded_sample -
+                                  s->sample_shift[chan]);
+    s->lpc_mode[chan] = get_bits(gb, 2);
+    nb_subframes      = get_bits(gb, 3) + 1;
+
+    i = 0;
+    if (nb_subframes > 1) {
+        if (get_bits_left(gb) < (nb_subframes - 1) * 6)
+            return AVERROR_INVALIDDATA;
+
+        prev = 0;
+        for (; i < nb_subframes - 1; i++) {
+            int subframe_end = get_bits(gb, 6) * s->subframe_scale;
+            if (subframe_end <= prev)
+                return AVERROR_INVALIDDATA;
+            subframe_len[i] = subframe_end - prev;
+            left           -= subframe_len[i];
+            prev            = subframe_end;
+        }
+
+        if (left <= 0)
+            return AVERROR_INVALIDDATA;
+    }
+    subframe_len[i] = left;
+
+    prev = 0;
+    for (i = 0; i < nb_subframes; i++) {
+        if ((ret = decode_subframe(s, decoded, subframe_len[i], prev)) < 0)
+            return ret;
+        decoded += subframe_len[i];
+        prev     = subframe_len[i];
+    }
+
+    return 0;
+}
+
+static int decorrelate(TAKDecContext *s, int c1, int c2, int length)
+{
+    GetBitContext *gb = &s->gb;
+    int32_t *p1       = s->decoded[c1] + 1;
+    int32_t *p2       = s->decoded[c2] + 1;
+    int i;
+    int dshift, dfactor;
+
+    switch (s->dmode) {
+    case 1: /* left/side */
+        for (i = 0; i < length; i++) {
+            int32_t a = p1[i];
+            int32_t b = p2[i];
+            p2[i]     = a + b;
+        }
+        break;
+    case 2: /* side/right */
+        for (i = 0; i < length; i++) {
+            int32_t a = p1[i];
+            int32_t b = p2[i];
+            p1[i]     = b - a;
+        }
+        break;
+    case 3: /* side/mid */
+        for (i = 0; i < length; i++) {
+            int32_t a = p1[i];
+            int32_t b = p2[i];
+            a        -= b >> 1;
+            p1[i]     = a;
+            p2[i]     = a + b;
+        }
+        break;
+    case 4: /* side/left with scale factor */
+        FFSWAP(int32_t*, p1, p2);
+    case 5: /* side/right with scale factor */
+        dshift  = get_bits_esc4(gb);
+        dfactor = get_sbits(gb, 10);
+        for (i = 0; i < length; i++) {
+            int32_t a = p1[i];
+            int32_t b = p2[i];
+            b         = dfactor * (b >> dshift) + 128 >> 8 << dshift;
+            p1[i]     = b - a;
+        }
+        break;
+    case 6:
+        FFSWAP(int32_t*, p1, p2);
+    case 7: {
+        LOCAL_ALIGNED_16(int16_t, filter, [MAX_PREDICTORS]) = { 0 };
+        int length2, order_half, filter_order, dval1, dval2;
+        int av_uninit(code_size);
+
+        if (length < 256)
+            return AVERROR_INVALIDDATA;
+
+        dshift       = get_bits_esc4(gb);
+        filter_order = 8 << get_bits1(gb);
+        dval1        = get_bits1(gb);
+        dval2        = get_bits1(gb);
+
+        for (i = 0; i < filter_order; i++) {
+            if (!(i & 3))
+                code_size = 14 - get_bits(gb, 3);
+            filter[i] = get_sbits(gb, code_size);
+        }
+
+        order_half = filter_order / 2;
+        length2    = length - (filter_order - 1);
+
+        /* decorrelate beginning samples */
+        if (dval1) {
+            for (i = 0; i < order_half; i++) {
+                int32_t a = p1[i];
+                int32_t b = p2[i];
+                p1[i]     = a + b;
+            }
+        }
+
+        /* decorrelate ending samples */
+        if (dval2) {
+            for (i = length2 + order_half; i < length; i++) {
+                int32_t a = p1[i];
+                int32_t b = p2[i];
+                p1[i]     = a + b;
+            }
+        }
+
+        av_fast_malloc(&s->residues, &s->residues_buf_size,
+                       FFALIGN(length + 16, 16) * sizeof(*s->residues));
+        if (!s->residues)
+            return AVERROR(ENOMEM);
+        memset(s->residues, 0, s->residues_buf_size);
+
+        for (i = 0; i < length; i++)
+            s->residues[i] = p2[i] >> dshift;
+
+        p1 += order_half;
+
+        for (i = 0; i < length2; i++) {
+            int v = 1 << 9;
+
+            v += s->dsp.scalarproduct_int16(&s->residues[i], filter,
+                                            FFALIGN(filter_order, 16));
+
+            p1[i] = (av_clip(v >> 10, -8192, 8191) << dshift) - p1[i];
+        }
+
+        emms_c();
+        break;
+    }
+    }
+
+    return 0;
+}
+
+static int tak_decode_frame(AVCodecContext *avctx, void *data,
+                            int *got_frame_ptr, AVPacket *pkt)
+{
+    TAKDecContext *s  = avctx->priv_data;
+    GetBitContext *gb = &s->gb;
+    int chan, i, ret, hsize;
+
+    if (pkt->size < TAK_MIN_FRAME_HEADER_BYTES)
+        return AVERROR_INVALIDDATA;
+
+    init_get_bits(gb, pkt->data, pkt->size * 8);
+
+    if ((ret = ff_tak_decode_frame_header(avctx, gb, &s->ti, 0)) < 0)
+        return ret;
+
+    if (s->ti.flags & TAK_FRAME_FLAG_HAS_METADATA) {
+        av_log_missing_feature(avctx, "frame metadata", 1);
+        return AVERROR_PATCHWELCOME;
+    }
+
+    hsize = get_bits_count(gb) / 8;
+    if (avctx->err_recognition & AV_EF_CRCCHECK) {
+        if (ff_tak_check_crc(pkt->data, hsize)) {
+            av_log(avctx, AV_LOG_ERROR, "CRC error\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    if (s->ti.codec != TAK_CODEC_MONO_STEREO &&
+        s->ti.codec != TAK_CODEC_MULTICHANNEL) {
+        av_log(avctx, AV_LOG_ERROR, "unsupported codec: %d\n", s->ti.codec);
+        return AVERROR_PATCHWELCOME;
+    }
+    if (s->ti.data_type) {
+        av_log(avctx, AV_LOG_ERROR,
+               "unsupported data type: %d\n", s->ti.data_type);
+        return AVERROR_INVALIDDATA;
+    }
+    if (s->ti.codec == TAK_CODEC_MONO_STEREO && s->ti.channels > 2) {
+        av_log(avctx, AV_LOG_ERROR,
+               "invalid number of channels: %d\n", s->ti.channels);
+        return AVERROR_INVALIDDATA;
+    }
+    if (s->ti.channels > 6) {
+        av_log(avctx, AV_LOG_ERROR,
+               "unsupported number of channels: %d\n", s->ti.channels);
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (s->ti.frame_samples <= 0) {
+        av_log(avctx, AV_LOG_ERROR, "unsupported/invalid number of samples\n");
+        return AVERROR_INVALIDDATA;
+    }
+
+    if (s->ti.bps != avctx->bits_per_coded_sample) {
+        avctx->bits_per_coded_sample = s->ti.bps;
+        if ((ret = set_bps_params(avctx)) < 0)
+            return ret;
+    }
+    if (s->ti.sample_rate != avctx->sample_rate) {
+        avctx->sample_rate = s->ti.sample_rate;
+        set_sample_rate_params(avctx);
+    }
+    if (s->ti.ch_layout)
+        avctx->channel_layout = s->ti.ch_layout;
+    avctx->channels = s->ti.channels;
+
+    s->nb_samples = s->ti.last_frame_samples ? s->ti.last_frame_samples
+                                             : s->ti.frame_samples;
+
+    s->frame.nb_samples = s->nb_samples;
+    if ((ret = ff_get_buffer(avctx, &s->frame)) < 0)
+        return ret;
+
+    if (avctx->bits_per_coded_sample <= 16) {
+        int buf_size = av_samples_get_buffer_size(NULL, avctx->channels,
+                                                  s->nb_samples,
+                                                  AV_SAMPLE_FMT_S32P, 0);
+        av_fast_malloc(&s->decode_buffer, &s->decode_buffer_size, buf_size);
+        if (!s->decode_buffer)
+            return AVERROR(ENOMEM);
+        ret = av_samples_fill_arrays((uint8_t **)s->decoded, NULL,
+                                     s->decode_buffer, avctx->channels,
+                                     s->nb_samples, AV_SAMPLE_FMT_S32P, 0);
+        if (ret < 0)
+            return ret;
+    } else {
+        for (chan = 0; chan < avctx->channels; chan++)
+            s->decoded[chan] = (int32_t *)s->frame.extended_data[chan];
+    }
+
+    if (s->nb_samples < 16) {
+        for (chan = 0; chan < avctx->channels; chan++) {
+            int32_t *decoded = s->decoded[chan];
+            for (i = 0; i < s->nb_samples; i++)
+                decoded[i] = get_sbits(gb, avctx->bits_per_coded_sample);
+        }
+    } else {
+        if (s->ti.codec == TAK_CODEC_MONO_STEREO) {
+            for (chan = 0; chan < avctx->channels; chan++)
+                if (ret = decode_channel(s, chan))
+                    return ret;
+
+            if (avctx->channels == 2) {
+                if (get_bits1(gb)) {
+                    // some kind of subframe length, but it seems to be unused
+                    skip_bits(gb, 6);
+                }
+
+                s->dmode = get_bits(gb, 3);
+                if (ret = decorrelate(s, 0, 1, s->nb_samples - 1))
+                    return ret;
+            }
+        } else if (s->ti.codec == TAK_CODEC_MULTICHANNEL) {
+            if (get_bits1(gb)) {
+                int ch_mask = 0;
+
+                chan = get_bits(gb, 4) + 1;
+                if (chan > avctx->channels)
+                    return AVERROR_INVALIDDATA;
+
+                for (i = 0; i < chan; i++) {
+                    int nbit = get_bits(gb, 4);
+
+                    if (nbit >= avctx->channels)
+                        return AVERROR_INVALIDDATA;
+
+                    if (ch_mask & 1 << nbit)
+                        return AVERROR_INVALIDDATA;
+
+                    s->mcdparams[i].present = get_bits1(gb);
+                    if (s->mcdparams[i].present) {
+                        s->mcdparams[i].index = get_bits(gb, 2);
+                        s->mcdparams[i].chan2 = get_bits(gb, 4);
+                        if (s->mcdparams[i].index == 1) {
+                            if ((nbit == s->mcdparams[i].chan2) ||
+                                (ch_mask & 1 << s->mcdparams[i].chan2))
+                                return AVERROR_INVALIDDATA;
+
+                            ch_mask |= 1 << s->mcdparams[i].chan2;
+                        } else if (!(ch_mask & 1 << s->mcdparams[i].chan2)) {
+                            return AVERROR_INVALIDDATA;
+                        }
+                    }
+                    s->mcdparams[i].chan1 = nbit;
+
+                    ch_mask |= 1 << nbit;
+                }
+            } else {
+                chan = avctx->channels;
+                for (i = 0; i < chan; i++) {
+                    s->mcdparams[i].present = 0;
+                    s->mcdparams[i].chan1   = i;
+                }
+            }
+
+            for (i = 0; i < chan; i++) {
+                if (s->mcdparams[i].present && s->mcdparams[i].index == 1)
+                    if (ret = decode_channel(s, s->mcdparams[i].chan2))
+                        return ret;
+
+                if (ret = decode_channel(s, s->mcdparams[i].chan1))
+                    return ret;
+
+                if (s->mcdparams[i].present) {
+                    s->dmode = mc_dmodes[s->mcdparams[i].index];
+                    if (ret = decorrelate(s,
+                                          s->mcdparams[i].chan2,
+                                          s->mcdparams[i].chan1,
+                                          s->nb_samples - 1))
+                        return ret;
+                }
+            }
+        }
+
+        for (chan = 0; chan < avctx->channels; chan++) {
+            int32_t *decoded = s->decoded[chan];
+
+            if (s->lpc_mode[chan])
+                decode_lpc(decoded, s->lpc_mode[chan], s->nb_samples);
+
+            if (s->sample_shift[chan] > 0)
+                for (i = 0; i < s->nb_samples; i++)
+                    decoded[i] <<= s->sample_shift[chan];
+        }
+    }
+
+    align_get_bits(gb);
+    skip_bits(gb, 24);
+    if (get_bits_left(gb) < 0)
+        av_log(avctx, AV_LOG_DEBUG, "overread\n");
+    else if (get_bits_left(gb) > 0)
+        av_log(avctx, AV_LOG_DEBUG, "underread\n");
+
+    if (avctx->err_recognition & AV_EF_CRCCHECK) {
+        if (ff_tak_check_crc(pkt->data + hsize,
+                             get_bits_count(gb) / 8 - hsize)) {
+            av_log(avctx, AV_LOG_ERROR, "CRC error\n");
+            return AVERROR_INVALIDDATA;
+        }
+    }
+
+    /* convert to output buffer */
+    switch (avctx->sample_fmt) {
+    case AV_SAMPLE_FMT_U8P:
+        for (chan = 0; chan < avctx->channels; chan++) {
+            uint8_t *samples = (uint8_t *)s->frame.extended_data[chan];
+            int32_t *decoded = s->decoded[chan];
+            for (i = 0; i < s->nb_samples; i++)
+                samples[i] = decoded[i] + 0x80;
+        }
+        break;
+    case AV_SAMPLE_FMT_S16P:
+        for (chan = 0; chan < avctx->channels; chan++) {
+            int16_t *samples = (int16_t *)s->frame.extended_data[chan];
+            int32_t *decoded = s->decoded[chan];
+            for (i = 0; i < s->nb_samples; i++)
+                samples[i] = decoded[i];
+        }
+        break;
+    case AV_SAMPLE_FMT_S32P:
+        for (chan = 0; chan < avctx->channels; chan++) {
+            int32_t *samples = (int32_t *)s->frame.extended_data[chan];
+            for (i = 0; i < s->nb_samples; i++)
+                samples[i] <<= 8;
+        }
+        break;
+    }
+
+    *got_frame_ptr   = 1;
+    *(AVFrame *)data = s->frame;
+
+    return pkt->size;
+}
+
+static av_cold int tak_decode_close(AVCodecContext *avctx)
+{
+    TAKDecContext *s = avctx->priv_data;
+
+    av_freep(&s->decode_buffer);
+    av_freep(&s->residues);
+
+    return 0;
+}
+
+AVCodec ff_tak_decoder = {
+    .name             = "tak",
+    .type             = AVMEDIA_TYPE_AUDIO,
+    .id               = AV_CODEC_ID_TAK,
+    .priv_data_size   = sizeof(TAKDecContext),
+    .init             = tak_decode_init,
+    .init_static_data = tak_init_static_data,
+    .close            = tak_decode_close,
+    .decode           = tak_decode_frame,
+    .capabilities     = CODEC_CAP_DR1,
+    .long_name        = NULL_IF_CONFIG_SMALL("TAK (Tom's lossless Audio Kompressor)"),
+    .sample_fmts      = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
+                                                        AV_SAMPLE_FMT_S16P,
+                                                        AV_SAMPLE_FMT_S32P,
+                                                        AV_SAMPLE_FMT_NONE },
+};