mpegaudiodec.c

    return 0;
}

/* Reorder short blocks from bitstream order to interleaved order. It
   would be faster to do it in parsing, but the code would be far more
   complicated */
static void reorder_block(MPADecodeContext *s, GranuleDef *g)
{
    int i, j, len;
    INTFLOAT *ptr, *dst, *ptr1;
    INTFLOAT tmp[576];

    if (g->block_type != 2)
        return;

    if (g->switch_point) {
        if (s->sample_rate_index != 8)
            ptr = g->sb_hybrid + 36;
        else
            ptr = g->sb_hybrid + 48;
    } else {
        ptr = g->sb_hybrid;
    }

    for (i = g->short_start; i < 13; i++) {
        len  = band_size_short[s->sample_rate_index][i];
        ptr1 = ptr;
        dst  = tmp;
        for (j = len; j > 0; j--) {
            *dst++ = ptr[0*len];
            *dst++ = ptr[1*len];
            *dst++ = ptr[2*len];
            ptr++;
        }
        ptr += 2 * len;
        memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
    }
}

#define ISQRT2 FIXR(0.70710678118654752440)

static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
{
    int i, j, k, l;
    int sf_max, sf, len, non_zero_found;
    INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
    int non_zero_found_short[3];

    /* intensity stereo */
    if (s->mode_ext & MODE_EXT_I_STEREO) {
        if (!s->lsf) {
            is_tab = is_table;
            sf_max = 7;
        } else {
            is_tab = is_table_lsf[g1->scalefac_compress & 1];
            sf_max = 16;
        }

        tab0 = g0->sb_hybrid + 576;
        tab1 = g1->sb_hybrid + 576;

        non_zero_found_short[0] = 0;
        non_zero_found_short[1] = 0;
        non_zero_found_short[2] = 0;
        k = (13 - g1->short_start) * 3 + g1->long_end - 3;
        for (i = 12; i >= g1->short_start; i--) {
            /* for last band, use previous scale factor */
            if (i != 11)
                k -= 3;
            len = band_size_short[s->sample_rate_index][i];
            for (l = 2; l >= 0; l--) {
                tab0 -= len;
                tab1 -= len;
                if (!non_zero_found_short[l]) {
                    /* test if non zero band. if so, stop doing i-stereo */
                    for (j = 0; j < len; j++) {
                        if (tab1[j] != 0) {
                            non_zero_found_short[l] = 1;
                            goto found1;
                        }
                    }
                    sf = g1->scale_factors[k + l];
                    if (sf >= sf_max)
                        goto found1;

                    v1 = is_tab[0][sf];
                    v2 = is_tab[1][sf];
                    for (j = 0; j < len; j++) {
                        tmp0    = tab0[j];
                        tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
                        tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
                    }
                } else {
found1:
                    if (s->mode_ext & MODE_EXT_MS_STEREO) {
                        /* lower part of the spectrum : do ms stereo
                           if enabled */
                        for (j = 0; j < len; j++) {
                            tmp0    = tab0[j];
                            tmp1    = tab1[j];
                            tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
                            tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
                        }
                    }
                }
            }
        }

        non_zero_found = non_zero_found_short[0] |
                         non_zero_found_short[1] |
                         non_zero_found_short[2];

        for (i = g1->long_end - 1;i >= 0;i--) {
            len   = band_size_long[s->sample_rate_index][i];
            tab0 -= len;
            tab1 -= len;
            /* test if non zero band. if so, stop doing i-stereo */
            if (!non_zero_found) {
                for (j = 0; j < len; j++) {
                    if (tab1[j] != 0) {
                        non_zero_found = 1;
                        goto found2;
                    }
                }
                /* for last band, use previous scale factor */
                k  = (i == 21) ? 20 : i;
                sf = g1->scale_factors[k];
                if (sf >= sf_max)
                    goto found2;
                v1 = is_tab[0][sf];
                v2 = is_tab[1][sf];
                for (j = 0; j < len; j++) {
                    tmp0    = tab0[j];
                    tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
                    tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
                }
            } else {
found2:
                if (s->mode_ext & MODE_EXT_MS_STEREO) {
                    /* lower part of the spectrum : do ms stereo
                       if enabled */
                    for (j = 0; j < len; j++) {
                        tmp0    = tab0[j];
                        tmp1    = tab1[j];
                        tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
                        tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
                    }
                }
            }
        }
    } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
        /* ms stereo ONLY */
        /* NOTE: the 1/sqrt(2) normalization factor is included in the
           global gain */
        tab0 = g0->sb_hybrid;
        tab1 = g1->sb_hybrid;
        for (i = 0; i < 576; i++) {
            tmp0    = tab0[i];
            tmp1    = tab1[i];
            tab0[i] = tmp0 + tmp1;
            tab1[i] = tmp0 - tmp1;
        }
    }
}

#if CONFIG_FLOAT
#define AA(j) do {                                                      \
        float tmp0 = ptr[-1-j];                                         \
        float tmp1 = ptr[   j];                                         \
        ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1];    \
        ptr[   j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0];    \
    } while (0)
#else
#define AA(j) do {                                              \
        int tmp0 = ptr[-1-j];                                   \
        int tmp1 = ptr[   j];                                   \
        int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]);          \
        ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2]));   \
        ptr[   j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3]));   \
    } while (0)
#endif

static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
{
    INTFLOAT *ptr;
    int n, i;

    /* we antialias only "long" bands */
    if (g->block_type == 2) {
        if (!g->switch_point)
            return;
        /* XXX: check this for 8000Hz case */
        n = 1;
    } else {
        n = SBLIMIT - 1;
    }

    ptr = g->sb_hybrid + 18;
    for (i = n; i > 0; i--) {
        AA(0);
        AA(1);
        AA(2);
        AA(3);
        AA(4);
        AA(5);
        AA(6);
        AA(7);

        ptr += 18;
    }
}

static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
                          INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
{
    INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
    INTFLOAT out2[12];
    int i, j, mdct_long_end, sblimit;

    /* find last non zero block */
    ptr  = g->sb_hybrid + 576;
    ptr1 = g->sb_hybrid + 2 * 18;
    while (ptr >= ptr1) {
        int32_t *p;
        ptr -= 6;
        p    = (int32_t*)ptr;
        if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
            break;
    }
    sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

    if (g->block_type == 2) {
        /* XXX: check for 8000 Hz */
        if (g->switch_point)
            mdct_long_end = 2;
        else
            mdct_long_end = 0;
    } else {
        mdct_long_end = sblimit;
    }

    s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
                                     mdct_long_end, g->switch_point,
                                     g->block_type);

    buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
    ptr = g->sb_hybrid + 18 * mdct_long_end;

    for (j = mdct_long_end; j < sblimit; j++) {
        /* select frequency inversion */
        win     = RENAME(ff_mdct_win)[2 + (4  & -(j & 1))];
        out_ptr = sb_samples + j;

        for (i = 0; i < 6; i++) {
            *out_ptr = buf[4*i];
            out_ptr += SBLIMIT;
        }
        imdct12(out2, ptr + 0);
        for (i = 0; i < 6; i++) {
            *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*1)];
            buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
            out_ptr += SBLIMIT;
        }
        imdct12(out2, ptr + 1);
        for (i = 0; i < 6; i++) {
            *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*2)];
            buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
            out_ptr += SBLIMIT;
        }
        imdct12(out2, ptr + 2);
        for (i = 0; i < 6; i++) {
            buf[4*(i + 6*0)] = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*0)];
            buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
            buf[4*(i + 6*2)] = 0;
        }
        ptr += 18;
        buf += (j&3) != 3 ? 1 : (4*18-3);
    }
    /* zero bands */
    for (j = sblimit; j < SBLIMIT; j++) {
        /* overlap */
        out_ptr = sb_samples + j;
        for (i = 0; i < 18; i++) {
            *out_ptr = buf[4*i];
            buf[4*i]   = 0;
            out_ptr += SBLIMIT;
        }
        buf += (j&3) != 3 ? 1 : (4*18-3);
    }
}

/* main layer3 decoding function */
static int mp_decode_layer3(MPADecodeContext *s)
{
    int nb_granules, main_data_begin;
    int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
    GranuleDef *g;
    int16_t exponents[576]; //FIXME try INTFLOAT

    /* read side info */
    if (s->lsf) {
        main_data_begin = get_bits(&s->gb, 8);
        skip_bits(&s->gb, s->nb_channels);
        nb_granules = 1;
    } else {
        main_data_begin = get_bits(&s->gb, 9);
        if (s->nb_channels == 2)
            skip_bits(&s->gb, 3);
        else
            skip_bits(&s->gb, 5);
        nb_granules = 2;
        for (ch = 0; ch < s->nb_channels; ch++) {
            s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
            s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
        }
    }

    for (gr = 0; gr < nb_granules; gr++) {
        for (ch = 0; ch < s->nb_channels; ch++) {
            av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
            g = &s->granules[ch][gr];
            g->part2_3_length = get_bits(&s->gb, 12);
            g->big_values     = get_bits(&s->gb,  9);
            if (g->big_values > 288) {
                av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
                return AVERROR_INVALIDDATA;
            }

            g->global_gain = get_bits(&s->gb, 8);
            /* if MS stereo only is selected, we precompute the
               1/sqrt(2) renormalization factor */
            if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
                MODE_EXT_MS_STEREO)
                g->global_gain -= 2;
            if (s->lsf)
                g->scalefac_compress = get_bits(&s->gb, 9);
            else
                g->scalefac_compress = get_bits(&s->gb, 4);
            blocksplit_flag = get_bits1(&s->gb);
            if (blocksplit_flag) {
                g->block_type = get_bits(&s->gb, 2);
                if (g->block_type == 0) {
                    av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
                    return AVERROR_INVALIDDATA;
                }
                g->switch_point = get_bits1(&s->gb);
                for (i = 0; i < 2; i++)
                    g->table_select[i] = get_bits(&s->gb, 5);
                for (i = 0; i < 3; i++)
                    g->subblock_gain[i] = get_bits(&s->gb, 3);
                ff_init_short_region(s, g);
            } else {
                int region_address1, region_address2;
                g->block_type = 0;
                g->switch_point = 0;
                for (i = 0; i < 3; i++)
                    g->table_select[i] = get_bits(&s->gb, 5);
                /* compute huffman coded region sizes */
                region_address1 = get_bits(&s->gb, 4);
                region_address2 = get_bits(&s->gb, 3);
                av_dlog(s->avctx, "region1=%d region2=%d\n",
                        region_address1, region_address2);
                ff_init_long_region(s, g, region_address1, region_address2);
            }
            ff_region_offset2size(g);
            ff_compute_band_indexes(s, g);

            g->preflag = 0;
            if (!s->lsf)
                g->preflag = get_bits1(&s->gb);
            g->scalefac_scale     = get_bits1(&s->gb);
            g->count1table_select = get_bits1(&s->gb);
            av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
                    g->block_type, g->switch_point);
        }
    }

    if (!s->adu_mode) {
        int skip;
        const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
        int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0, EXTRABYTES);
        assert((get_bits_count(&s->gb) & 7) == 0);
        /* now we get bits from the main_data_begin offset */
        av_dlog(s->avctx, "seekback: %d\n", main_data_begin);
    //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);

        memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
        s->in_gb = s->gb;
        init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
#if !UNCHECKED_BITSTREAM_READER
        s->gb.size_in_bits_plus8 += extrasize * 8;
#endif
        s->last_buf_size <<= 3;
        for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
            for (ch = 0; ch < s->nb_channels; ch++) {
                g = &s->granules[ch][gr];
                s->last_buf_size += g->part2_3_length;
                memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
            }
        }
        skip = s->last_buf_size - 8 * main_data_begin;
        if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
            skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
            s->gb           = s->in_gb;
            s->in_gb.buffer = NULL;
        } else {
            skip_bits_long(&s->gb, skip);
        }
    } else {
        gr = 0;
    }

    for (; gr < nb_granules; gr++) {
        for (ch = 0; ch < s->nb_channels; ch++) {
            g = &s->granules[ch][gr];
            bits_pos = get_bits_count(&s->gb);

            if (!s->lsf) {
                uint8_t *sc;
                int slen, slen1, slen2;

                /* MPEG1 scale factors */
                slen1 = slen_table[0][g->scalefac_compress];
                slen2 = slen_table[1][g->scalefac_compress];
                av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
                if (g->block_type == 2) {
                    n = g->switch_point ? 17 : 18;
                    j = 0;
                    if (slen1) {
                        for (i = 0; i < n; i++)
                            g->scale_factors[j++] = get_bits(&s->gb, slen1);
                    } else {
                        for (i = 0; i < n; i++)
                            g->scale_factors[j++] = 0;
                    }
                    if (slen2) {
                        for (i = 0; i < 18; i++)
                            g->scale_factors[j++] = get_bits(&s->gb, slen2);
                        for (i = 0; i < 3; i++)
                            g->scale_factors[j++] = 0;
                    } else {
                        for (i = 0; i < 21; i++)
                            g->scale_factors[j++] = 0;
                    }
                } else {
                    sc = s->granules[ch][0].scale_factors;
                    j = 0;
                    for (k = 0; k < 4; k++) {
                        n = k == 0 ? 6 : 5;
                        if ((g->scfsi & (0x8 >> k)) == 0) {
                            slen = (k < 2) ? slen1 : slen2;
                            if (slen) {
                                for (i = 0; i < n; i++)
                                    g->scale_factors[j++] = get_bits(&s->gb, slen);
                            } else {
                                for (i = 0; i < n; i++)
                                    g->scale_factors[j++] = 0;
                            }
                        } else {
                            /* simply copy from last granule */
                            for (i = 0; i < n; i++) {
                                g->scale_factors[j] = sc[j];
                                j++;
                            }
                        }
                    }
                    g->scale_factors[j++] = 0;
                }
            } else {
                int tindex, tindex2, slen[4], sl, sf;

                /* LSF scale factors */
                if (g->block_type == 2)
                    tindex = g->switch_point ? 2 : 1;
                else
                    tindex = 0;

                sf = g->scalefac_compress;
                if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
                    /* intensity stereo case */
                    sf >>= 1;
                    if (sf < 180) {
                        lsf_sf_expand(slen, sf, 6, 6, 0);
                        tindex2 = 3;
                    } else if (sf < 244) {
                        lsf_sf_expand(slen, sf - 180, 4, 4, 0);
                        tindex2 = 4;
                    } else {
                        lsf_sf_expand(slen, sf - 244, 3, 0, 0);
                        tindex2 = 5;
                    }
                } else {
                    /* normal case */
                    if (sf < 400) {
                        lsf_sf_expand(slen, sf, 5, 4, 4);
                        tindex2 = 0;
                    } else if (sf < 500) {
                        lsf_sf_expand(slen, sf - 400, 5, 4, 0);
                        tindex2 = 1;
                    } else {
                        lsf_sf_expand(slen, sf - 500, 3, 0, 0);
                        tindex2 = 2;
                        g->preflag = 1;
                    }
                }

                j = 0;
                for (k = 0; k < 4; k++) {
                    n  = lsf_nsf_table[tindex2][tindex][k];
                    sl = slen[k];
                    if (sl) {
                        for (i = 0; i < n; i++)
                            g->scale_factors[j++] = get_bits(&s->gb, sl);
                    } else {
                        for (i = 0; i < n; i++)
                            g->scale_factors[j++] = 0;
                    }
                }
                /* XXX: should compute exact size */
                for (; j < 40; j++)
                    g->scale_factors[j] = 0;
            }

            exponents_from_scale_factors(s, g, exponents);

            /* read Huffman coded residue */
            huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
        } /* ch */

        if (s->nb_channels == 2)
            compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);

        for (ch = 0; ch < s->nb_channels; ch++) {
            g = &s->granules[ch][gr];

            reorder_block(s, g);
            compute_antialias(s, g);
            compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
        }
    } /* gr */
    if (get_bits_count(&s->gb) < 0)
        skip_bits_long(&s->gb, -get_bits_count(&s->gb));
    return nb_granules * 18;
}

static int mp_decode_frame(MPADecodeContext *s, OUT_INT *samples,
                           const uint8_t *buf, int buf_size)
{
    int i, nb_frames, ch, ret;
    OUT_INT *samples_ptr;

    init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);

    /* skip error protection field */
    if (s->error_protection)
        skip_bits(&s->gb, 16);

    switch(s->layer) {
    case 1:
        s->avctx->frame_size = 384;
        nb_frames = mp_decode_layer1(s);
        break;
    case 2:
        s->avctx->frame_size = 1152;
        nb_frames = mp_decode_layer2(s);
        break;
    case 3:
        s->avctx->frame_size = s->lsf ? 576 : 1152;
    default:
        nb_frames = mp_decode_layer3(s);

        s->last_buf_size=0;
        if (s->in_gb.buffer) {
            align_get_bits(&s->gb);
            i = get_bits_left(&s->gb)>>3;
            if (i >= 0 && i <= BACKSTEP_SIZE) {
                memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
                s->last_buf_size=i;
            } else
                av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
            s->gb           = s->in_gb;
            s->in_gb.buffer = NULL;
        }

        align_get_bits(&s->gb);
        assert((get_bits_count(&s->gb) & 7) == 0);
        i = get_bits_left(&s->gb) >> 3;

        if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
            if (i < 0)
                av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
            i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
        }
        assert(i <= buf_size - HEADER_SIZE && i >= 0);
        memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
        s->last_buf_size += i;
    }

    /* get output buffer */
    if (!samples) {
        s->frame.nb_samples = s->avctx->frame_size;
        if ((ret = s->avctx->get_buffer(s->avctx, &s->frame)) < 0) {
            av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
            return ret;
        }
        samples = (OUT_INT *)s->frame.data[0];
    }

    /* apply the synthesis filter */
    for (ch = 0; ch < s->nb_channels; ch++) {
        samples_ptr = samples + ch;
        for (i = 0; i < nb_frames; i++) {
            RENAME(ff_mpa_synth_filter)(
                         &s->mpadsp,
                         s->synth_buf[ch], &(s->synth_buf_offset[ch]),
                         RENAME(ff_mpa_synth_window), &s->dither_state,
                         samples_ptr, s->nb_channels,
                         s->sb_samples[ch][i]);
            samples_ptr += 32 * s->nb_channels;
        }
    }

    return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
}

static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
                        AVPacket *avpkt)
{
    const uint8_t *buf  = avpkt->data;
    int buf_size        = avpkt->size;
    MPADecodeContext *s = avctx->priv_data;
    uint32_t header;
    int out_size;

    if (buf_size < HEADER_SIZE)
        return AVERROR_INVALIDDATA;

    header = AV_RB32(buf);
    if (ff_mpa_check_header(header) < 0) {
        av_log(avctx, AV_LOG_ERROR, "Header missing\n");
        return AVERROR_INVALIDDATA;
    }

    if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
        /* free format: prepare to compute frame size */
        s->frame_size = -1;
        return AVERROR_INVALIDDATA;
    }
    /* update codec info */
    avctx->channels       = s->nb_channels;
    avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
    if (!avctx->bit_rate)
        avctx->bit_rate = s->bit_rate;
    avctx->sub_id = s->layer;

    if (s->frame_size <= 0 || s->frame_size > buf_size) {
        av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
        return AVERROR_INVALIDDATA;
    } else if (s->frame_size < buf_size) {
        av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");
        buf_size= s->frame_size;
    }

    out_size = mp_decode_frame(s, NULL, buf, buf_size);
    if (out_size >= 0) {
        *got_frame_ptr   = 1;
        *(AVFrame *)data = s->frame;
        avctx->sample_rate = s->sample_rate;
        //FIXME maybe move the other codec info stuff from above here too
    } else {
        av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
        /* Only return an error if the bad frame makes up the whole packet.
           If there is more data in the packet, just consume the bad frame
           instead of returning an error, which would discard the whole
           packet. */
        *got_frame_ptr = 0;
        if (buf_size == avpkt->size)
            return out_size;
    }
    s->frame_size = 0;
    return buf_size;
}

static void flush(AVCodecContext *avctx)
{
    MPADecodeContext *s = avctx->priv_data;
    memset(s->synth_buf, 0, sizeof(s->synth_buf));
    s->last_buf_size = 0;
}

#if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
static int decode_frame_adu(AVCodecContext *avctx, void *data,
                            int *got_frame_ptr, AVPacket *avpkt)
{
    const uint8_t *buf  = avpkt->data;
    int buf_size        = avpkt->size;
    MPADecodeContext *s = avctx->priv_data;
    uint32_t header;
    int len, out_size;

    len = buf_size;

    // Discard too short frames
    if (buf_size < HEADER_SIZE) {
        av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
        return AVERROR_INVALIDDATA;
    }


    if (len > MPA_MAX_CODED_FRAME_SIZE)
        len = MPA_MAX_CODED_FRAME_SIZE;

    // Get header and restore sync word
    header = AV_RB32(buf) | 0xffe00000;

    if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
        av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
        return AVERROR_INVALIDDATA;
    }

    avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
    /* update codec info */
    avctx->sample_rate = s->sample_rate;
    avctx->channels    = s->nb_channels;
    if (!avctx->bit_rate)
        avctx->bit_rate = s->bit_rate;
    avctx->sub_id = s->layer;

    s->frame_size = len;

    out_size = mp_decode_frame(s, NULL, buf, buf_size);

    *got_frame_ptr   = 1;
    *(AVFrame *)data = s->frame;

    return buf_size;
}
#endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */

#if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER

/**
 * Context for MP3On4 decoder
 */
typedef struct MP3On4DecodeContext {
    AVFrame *frame;
    int frames;                     ///< number of mp3 frames per block (number of mp3 decoder instances)
    int syncword;                   ///< syncword patch
    const uint8_t *coff;            ///< channel offsets in output buffer
    MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
    OUT_INT *decoded_buf;           ///< output buffer for decoded samples
} MP3On4DecodeContext;

#include "mpeg4audio.h"

/* Next 3 arrays are indexed by channel config number (passed via codecdata) */

/* number of mp3 decoder instances */
static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };

/* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
static const uint8_t chan_offset[8][5] = {
    { 0             },
    { 0             },  // C
    { 0             },  // FLR
    { 2, 0          },  // C FLR
    { 2, 0, 3       },  // C FLR BS
    { 2, 0, 3       },  // C FLR BLRS
    { 2, 0, 4, 3    },  // C FLR BLRS LFE
    { 2, 0, 6, 4, 3 },  // C FLR BLRS BLR LFE
};

/* mp3on4 channel layouts */
static const int16_t chan_layout[8] = {
    0,
    AV_CH_LAYOUT_MONO,
    AV_CH_LAYOUT_STEREO,
    AV_CH_LAYOUT_SURROUND,
    AV_CH_LAYOUT_4POINT0,
    AV_CH_LAYOUT_5POINT0,
    AV_CH_LAYOUT_5POINT1,
    AV_CH_LAYOUT_7POINT1
};

static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
{
    MP3On4DecodeContext *s = avctx->priv_data;
    int i;

    for (i = 0; i < s->frames; i++)
        av_free(s->mp3decctx[i]);

    av_freep(&s->decoded_buf);

    return 0;
}


static int decode_init_mp3on4(AVCodecContext * avctx)
{
    MP3On4DecodeContext *s = avctx->priv_data;
    MPEG4AudioConfig cfg;
    int i;

    if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
        av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
        return AVERROR_INVALIDDATA;
    }

    avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
                                 avctx->extradata_size * 8, 1);
    if (!cfg.chan_config || cfg.chan_config > 7) {
        av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
        return AVERROR_INVALIDDATA;
    }
    s->frames             = mp3Frames[cfg.chan_config];
    s->coff               = chan_offset[cfg.chan_config];
    avctx->channels       = ff_mpeg4audio_channels[cfg.chan_config];
    avctx->channel_layout = chan_layout[cfg.chan_config];

    if (cfg.sample_rate < 16000)
        s->syncword = 0xffe00000;
    else
        s->syncword = 0xfff00000;

    /* Init the first mp3 decoder in standard way, so that all tables get builded
     * We replace avctx->priv_data with the context of the first decoder so that
     * decode_init() does not have to be changed.
     * Other decoders will be initialized here copying data from the first context
     */
    // Allocate zeroed memory for the first decoder context
    s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
    if (!s->mp3decctx[0])
        goto alloc_fail;
    // Put decoder context in place to make init_decode() happy
    avctx->priv_data = s->mp3decctx[0];
    decode_init(avctx);
    s->frame = avctx->coded_frame;
    // Restore mp3on4 context pointer
    avctx->priv_data = s;
    s->mp3decctx[0]->adu_mode = 1; // Set adu mode

    /* Create a separate codec/context for each frame (first is already ok).
     * Each frame is 1 or 2 channels - up to 5 frames allowed
     */
    for (i = 1; i < s->frames; i++) {
        s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
        if (!s->mp3decctx[i])
            goto alloc_fail;
        s->mp3decctx[i]->adu_mode = 1;
        s->mp3decctx[i]->avctx = avctx;
        s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
    }

    /* Allocate buffer for multi-channel output if needed */
    if (s->frames > 1) {
        s->decoded_buf = av_malloc(MPA_FRAME_SIZE * MPA_MAX_CHANNELS *
                                   sizeof(*s->decoded_buf));
        if (!s->decoded_buf)
            goto alloc_fail;
    }

    return 0;
alloc_fail:
    decode_close_mp3on4(avctx);
    return AVERROR(ENOMEM);
}


static void flush_mp3on4(AVCodecContext *avctx)
{
    int i;
    MP3On4DecodeContext *s = avctx->priv_data;

    for (i = 0; i < s->frames; i++) {
        MPADecodeContext *m = s->mp3decctx[i];
        memset(m->synth_buf, 0, sizeof(m->synth_buf));
        m->last_buf_size = 0;
    }
}


static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
                               int *got_frame_ptr, AVPacket *avpkt)
{
    const uint8_t *buf     = avpkt->data;
    int buf_size           = avpkt->size;
    MP3On4DecodeContext *s = avctx->priv_data;
    MPADecodeContext *m;
    int fsize, len = buf_size, out_size = 0;
    uint32_t header;
    OUT_INT *out_samples;
    OUT_INT *outptr, *bp;
    int fr, j, n, ch, ret;

    /* get output buffer */
    s->frame->nb_samples = MPA_FRAME_SIZE;
    if ((ret = avctx->get_buffer(avctx, s->frame)) < 0) {
        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return ret;
    }
    out_samples = (OUT_INT *)s->frame->data[0];

    // Discard too short frames
    if (buf_size < HEADER_SIZE)
        return AVERROR_INVALIDDATA;

    // If only one decoder interleave is not needed
    outptr = s->frames == 1 ? out_samples : s->decoded_buf;

    avctx->bit_rate = 0;

    ch = 0;
    for (fr = 0; fr < s->frames; fr++) {
        fsize = AV_RB16(buf) >> 4;
        fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
        m     = s->mp3decctx[fr];
        assert(m != NULL);

        if (fsize < HEADER_SIZE) {
            av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
            return AVERROR_INVALIDDATA;
        }
        header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

        if (ff_mpa_check_header(header) < 0) // Bad header, discard block
            break;

        avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);

        if (ch + m->nb_channels > avctx->channels) {
            av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
                                        "channel count\n");
            return AVERROR_INVALIDDATA;
        }
        ch += m->nb_channels;

        out_size += mp_decode_frame(m, outptr, buf, fsize);
        buf      += fsize;
        len      -= fsize;

        if (s->frames > 1) {
            n = m->avctx->frame_size*m->nb_channels;
            /* interleave output data */
            bp = out_samples + s->coff[fr];
            if (m->nb_channels == 1) {
                for (j = 0; j < n; j++) {
                    *bp = s->decoded_buf[j];
                    bp += avctx->channels;
                }
            } else {
                for (j = 0; j < n; j++) {
                    bp[0] = s->decoded_buf[j++];
                    bp[1] = s->decoded_buf[j];
                    bp   += avctx->channels;
                }
            }
        }
        avctx->bit_rate += m->bit_rate;
    }

    /* update codec info */
    avctx->sample_rate = s->mp3decctx[0]->sample_rate;

    s->frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
    *got_frame_ptr   = 1;
    *(AVFrame *)data = *s->frame;

    return buf_size;
}
#endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */

#if !CONFIG_FLOAT
#if CONFIG_MP1_DECODER
AVCodec ff_mp1_decoder = {
    .name           = "mp1",
    .type           = AVMEDIA_TYPE_AUDIO,
    .id             = CODEC_ID_MP1,
    .priv_data_size = sizeof(MPADecodeContext),
    .init           = decode_init,
    .decode         = decode_frame,
    .capabilities   = CODEC_CAP_DR1,
    .flush          = flush,
    .long_name      = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
};
#endif
#if CONFIG_MP2_DECODER
AVCodec ff_mp2_decoder = {
    .name           = "mp2",
    .type           = AVMEDIA_TYPE_AUDIO,
    .id             = CODEC_ID_MP2,
    .priv_data_size = sizeof(MPADecodeContext),
    .init           = decode_init,
    .decode         = decode_frame,
    .capabilities   = CODEC_CAP_DR1,
    .flush          = flush,
    .long_name      = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
};
#endif
#if CONFIG_MP3_DECODER