vorbis_dec.c

来自「ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦」· C语言 代码 · 共 1,638 行 · 第 1/4 页

// Read and decode residue

static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fast8_t ch, uint_fast8_t *do_not_decode, float *vec, uint_fast16_t vlen) {
    GetBitContext *gb=&vc->gb;
    uint_fast8_t c_p_c=vc->codebooks[vr->classbook].dimensions;
    uint_fast16_t n_to_read=vr->end-vr->begin;
    uint_fast16_t ptns_to_read=n_to_read/vr->partition_size;
    uint_fast8_t classifs[ptns_to_read*vc->audio_channels];
    uint_fast8_t pass;
    uint_fast8_t ch_used;
    uint_fast8_t i,j,l;
    uint_fast16_t k;

    if (vr->type==2) {
        for(j=1;j<ch;++j) {
                do_not_decode[0]&=do_not_decode[j];  // FIXME - clobbering input
        }
        if (do_not_decode[0]) return 0;
        ch_used=1;
    } else {
        ch_used=ch;
    }

    AV_DEBUG(" residue type 0/1/2 decode begin, ch: %d  cpc %d  \n", ch, c_p_c);

    for(pass=0;pass<=vr->maxpass;++pass) { // FIXME OPTIMIZE?
        uint_fast16_t voffset;
        uint_fast16_t partition_count;
        uint_fast16_t j_times_ptns_to_read;

        voffset=vr->begin;
        for(partition_count=0;partition_count<ptns_to_read;) {  // SPEC        error
            if (!pass) {
                uint_fast32_t inverse_class = ff_inverse[vr->classifications];
                for(j_times_ptns_to_read=0, j=0;j<ch_used;++j) {
                    if (!do_not_decode[j]) {
                        uint_fast32_t temp=get_vlc2(gb, vc->codebooks[vr->classbook].vlc.table,
                        vc->codebooks[vr->classbook].nb_bits, 3);

                        AV_DEBUG("Classword: %d \n", temp);

                        assert(vr->classifications > 1 && temp<=65536); //needed for inverse[]
                        for(i=0;i<c_p_c;++i) {
                            uint_fast32_t temp2;

                            temp2=(((uint_fast64_t)temp) * inverse_class)>>32;
                            if (partition_count+c_p_c-1-i < ptns_to_read) {
                                classifs[j_times_ptns_to_read+partition_count+c_p_c-1-i]=temp-temp2*vr->classifications;
                            }
                            temp=temp2;
                        }
                    }
                    j_times_ptns_to_read+=ptns_to_read;
                }
            }
            for(i=0;(i<c_p_c) && (partition_count<ptns_to_read);++i) {
                for(j_times_ptns_to_read=0, j=0;j<ch_used;++j) {
                    uint_fast16_t voffs;

                    if (!do_not_decode[j]) {
                        uint_fast8_t vqclass=classifs[j_times_ptns_to_read+partition_count];
                        int_fast16_t vqbook=vr->books[vqclass][pass];

                        if (vqbook>=0 && vc->codebooks[vqbook].codevectors) {
                            uint_fast16_t coffs;
                            unsigned dim= vc->codebooks[vqbook].dimensions; // not uint_fast8_t: 64bit is slower here on amd64
                            uint_fast16_t step= dim==1 ? vr->partition_size
                                              : FASTDIV(vr->partition_size, dim);
                            vorbis_codebook codebook= vc->codebooks[vqbook];

                            if (vr->type==0) {

                                voffs=voffset+j*vlen;
                                for(k=0;k<step;++k) {
                                    coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
                                    for(l=0;l<dim;++l) {
                                        vec[voffs+k+l*step]+=codebook.codevectors[coffs+l];  // FPMATH
                                    }
                                }
                            }
                            else if (vr->type==1) {
                                voffs=voffset+j*vlen;
                                for(k=0;k<step;++k) {
                                    coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
                                    for(l=0;l<dim;++l, ++voffs) {
                                        vec[voffs]+=codebook.codevectors[coffs+l];  // FPMATH

                                        AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d  \n", pass, voffs, vec[voffs], codebook.codevectors[coffs+l], coffs);
                                    }
                                }
                            }
                            else if (vr->type==2 && ch==2 && (voffset&1)==0 && (dim&1)==0) { // most frequent case optimized
                                voffs=voffset>>1;

                                if(dim==2) {
                                    for(k=0;k<step;++k) {
                                        coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * 2;
                                        vec[voffs+k     ]+=codebook.codevectors[coffs  ];  // FPMATH
                                        vec[voffs+k+vlen]+=codebook.codevectors[coffs+1];  // FPMATH
                                    }
                                } else
                                for(k=0;k<step;++k) {
                                    coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
                                    for(l=0;l<dim;l+=2, voffs++) {
                                        vec[voffs     ]+=codebook.codevectors[coffs+l  ];  // FPMATH
                                        vec[voffs+vlen]+=codebook.codevectors[coffs+l+1];  // FPMATH

                                        AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d  \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], codebook.codevectors[coffs+l], coffs, l);
                                    }
                                }

                            }
                            else if (vr->type==2) {
                                voffs=voffset;

                                for(k=0;k<step;++k) {
                                    coffs=get_vlc2(gb, codebook.vlc.table, codebook.nb_bits, 3) * dim;
                                    for(l=0;l<dim;++l, ++voffs) {
                                        vec[voffs/ch+(voffs%ch)*vlen]+=codebook.codevectors[coffs+l];  // FPMATH FIXME use if and counter instead of / and %

                                        AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d  \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], codebook.codevectors[coffs+l], coffs, l);
                                    }
                                }
                            } else {
                                av_log(vc->avccontext, AV_LOG_ERROR, " Invalid residue type while residue decode?! \n");
                                return 1;
                            }
                        }
                    }
                    j_times_ptns_to_read+=ptns_to_read;
                }
                ++partition_count;
                voffset+=vr->partition_size;
            }
        }
    }
    return 0;
}

void vorbis_inverse_coupling(float *mag, float *ang, int blocksize)
{
    int i;
    for(i=0; i<blocksize; i++)
    {
        if (mag[i]>0.0) {
            if (ang[i]>0.0) {
                ang[i]=mag[i]-ang[i];
            } else {
                float temp=ang[i];
                ang[i]=mag[i];
                mag[i]+=temp;
            }
        } else {
            if (ang[i]>0.0) {
                ang[i]+=mag[i];
            } else {
                float temp=ang[i];
                ang[i]=mag[i];
                mag[i]-=temp;
            }
        }
    }
}

// Decode the audio packet using the functions above

static int vorbis_parse_audio_packet(vorbis_context *vc) {
    GetBitContext *gb=&vc->gb;

    uint_fast8_t previous_window=0,next_window=0;
    uint_fast8_t mode_number;
    uint_fast16_t blocksize;
    int_fast32_t i,j;
    uint_fast8_t no_residue[vc->audio_channels];
    uint_fast8_t do_not_decode[vc->audio_channels];
    vorbis_mapping *mapping;
    float *ch_res_ptr=vc->channel_residues;
    float *ch_floor_ptr=vc->channel_floors;
    uint_fast8_t res_chan[vc->audio_channels];
    uint_fast8_t res_num=0;
    int_fast16_t retlen=0;
    uint_fast16_t saved_start=0;
    float fadd_bias = vc->add_bias;

    if (get_bits1(gb)) {
        av_log(vc->avccontext, AV_LOG_ERROR, "Not a Vorbis I audio packet.\n");
        return -1; // packet type not audio
    }

    if (vc->mode_count==1) {
        mode_number=0;
    } else {
        mode_number=get_bits(gb, ilog(vc->mode_count-1));
    }
    vc->mode_number=mode_number;
    mapping=&vc->mappings[vc->modes[mode_number].mapping];

    AV_DEBUG(" Mode number: %d , mapping: %d , blocktype %d \n", mode_number, vc->modes[mode_number].mapping, vc->modes[mode_number].blockflag);

    if (vc->modes[mode_number].blockflag) {
        previous_window=get_bits1(gb);
        next_window=get_bits1(gb);
    }

    blocksize=vc->blocksize[vc->modes[mode_number].blockflag];
    memset(ch_res_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?
    memset(ch_floor_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?

// Decode floor

    for(i=0;i<vc->audio_channels;++i) {
        vorbis_floor *floor;
        if (mapping->submaps>1) {
            floor=&vc->floors[mapping->submap_floor[mapping->mux[i]]];
        } else {
            floor=&vc->floors[mapping->submap_floor[0]];
        }

        no_residue[i]=floor->decode(vc, &floor->data, ch_floor_ptr);
        ch_floor_ptr+=blocksize/2;
    }

// Nonzero vector propagate

    for(i=mapping->coupling_steps-1;i>=0;--i) {
        if (!(no_residue[mapping->magnitude[i]] & no_residue[mapping->angle[i]])) {
            no_residue[mapping->magnitude[i]]=0;
            no_residue[mapping->angle[i]]=0;
        }
    }

// Decode residue

    for(i=0;i<mapping->submaps;++i) {
        vorbis_residue *residue;
        uint_fast8_t ch=0;

        for(j=0;j<vc->audio_channels;++j) {
            if ((mapping->submaps==1) || (i=mapping->mux[j])) {
                res_chan[j]=res_num;
                if (no_residue[j]) {
                    do_not_decode[ch]=1;
                } else {
                    do_not_decode[ch]=0;
                }
                ++ch;
                ++res_num;
            }
        }
        residue=&vc->residues[mapping->submap_residue[i]];
        vorbis_residue_decode(vc, residue, ch, do_not_decode, ch_res_ptr, blocksize/2);

        ch_res_ptr+=ch*blocksize/2;
    }

// Inverse coupling

    for(i=mapping->coupling_steps-1;i>=0;--i) { //warning: i has to be signed
        float *mag, *ang;

        mag=vc->channel_residues+res_chan[mapping->magnitude[i]]*blocksize/2;
        ang=vc->channel_residues+res_chan[mapping->angle[i]]*blocksize/2;
        vc->dsp.vorbis_inverse_coupling(mag, ang, blocksize/2);
    }

// Dotproduct

    for(j=0, ch_floor_ptr=vc->channel_floors;j<vc->audio_channels;++j,ch_floor_ptr+=blocksize/2) {
        ch_res_ptr=vc->channel_residues+res_chan[j]*blocksize/2;
        vc->dsp.vector_fmul(ch_floor_ptr, ch_res_ptr, blocksize/2);
    }

// MDCT, overlap/add, save data for next overlapping  FPMATH

    for(j=0;j<vc->audio_channels;++j) {
        uint_fast8_t step=vc->audio_channels;
        uint_fast16_t k;
        float *saved=vc->saved+j*vc->blocksize[1]/2;
        float *ret=vc->ret;
        const float *lwin=vc->win[1];
        const float *swin=vc->win[0];
        float *buf=vc->buf;
        float *buf_tmp=vc->buf_tmp;

        ch_floor_ptr=vc->channel_floors+j*blocksize/2;

        saved_start=vc->saved_start;

        vc->mdct[0].fft.imdct_calc(&vc->mdct[vc->modes[mode_number].blockflag], buf, ch_floor_ptr, buf_tmp);

        //FIXME process channels together, to allow faster simd vector_fmul_add_add?
        if (vc->modes[mode_number].blockflag) {
            // -- overlap/add
            if (previous_window) {
                vc->dsp.vector_fmul_add_add(ret+j, buf, lwin, saved, vc->add_bias, vc->blocksize[1]/2, step);
                retlen=vc->blocksize[1]/2;
            } else {
                int len = (vc->blocksize[1]-vc->blocksize[0])/4;
                buf += len;
                vc->dsp.vector_fmul_add_add(ret+j, buf, swin, saved, vc->add_bias, vc->blocksize[0]/2, step);
                k = vc->blocksize[0]/2*step + j;
                buf += vc->blocksize[0]/2;
                if(vc->exp_bias){
                    for(i=0; i<len; i++, k+=step)
                        ((uint32_t*)ret)[k] = ((uint32_t*)buf)[i] + vc->exp_bias; // ret[k]=buf[i]*(1<<bias)
                } else {
                    for(i=0; i<len; i++, k+=step)
                        ret[k] = buf[i] + fadd_bias;
                }
                buf=vc->buf;
                retlen=vc->blocksize[0]/2+len;
            }
            // -- save
            if (next_window) {
                buf += vc->blocksize[1]/2;
                vc->dsp.vector_fmul_reverse(saved, buf, lwin, vc->blocksize[1]/2);
                saved_start=0;
            } else {
                saved_start=(vc->blocksize[1]-vc->blocksize[0])/4;
                buf += vc->blocksize[1]/2;
                for(i=0; i<saved_start; i++)
                    ((uint32_t*)saved)[i] = ((uint32_t*)buf)[i] + vc->exp_bias;
                vc->dsp.vector_fmul_reverse(saved+saved_start, buf+saved_start, swin, vc->blocksize[0]/2);
            }
        } else {
            // --overlap/add
            if(vc->add_bias) {
                for(k=j, i=0;i<saved_start;++i, k+=step)
                    ret[k] = saved[i] + fadd_bias;
            } else {
                for(k=j, i=0;i<saved_start;++i, k+=step)
                    ret[k] = saved[i];
            }
            vc->dsp.vector_fmul_add_add(ret+k, buf, swin, saved+saved_start, vc->add_bias, vc->blocksize[0]/2, step);
            retlen=saved_start+vc->blocksize[0]/2;
            // -- save
            buf += vc->blocksize[0]/2;
            vc->dsp.vector_fmul_reverse(saved, buf, swin, vc->blocksize[0]/2);
            saved_start=0;
        }
    }
    vc->saved_start=saved_start;

    return retlen*vc->audio_channels;
}

// Return the decoded audio packet through the standard api

static int vorbis_decode_frame(AVCodecContext *avccontext,
                        void *data, int *data_size,
                        uint8_t *buf, int buf_size)
{
    vorbis_context *vc = avccontext->priv_data ;
    GetBitContext *gb = &(vc->gb);

    int_fast16_t len;

    if(!buf_size){
        return 0;
    }

    AV_DEBUG("packet length %d \n", buf_size);

    init_get_bits(gb, buf, buf_size*8);

    len=vorbis_parse_audio_packet(vc);

    if (len<=0) {
        *data_size=0;
        return buf_size;
    }

    if (!vc->first_frame) {
        vc->first_frame=1;
        *data_size=0;
        return buf_size ;
    }

    AV_DEBUG("parsed %d bytes %d bits, returned %d samples (*ch*bits) \n", get_bits_count(gb)/8, get_bits_count(gb)%8, len);

    vc->dsp.float_to_int16(data, vc->ret, len);
    *data_size=len*2;

    return buf_size ;
}

// Close decoder

static int vorbis_decode_close(AVCodecContext *avccontext) {
    vorbis_context *vc = avccontext->priv_data;

    vorbis_free(vc);

    return 0 ;
}

AVCodec vorbis_decoder = {
    "vorbis",
    CODEC_TYPE_AUDIO,
    CODEC_ID_VORBIS,
    sizeof(vorbis_context),
    vorbis_decode_init,
    NULL,
    vorbis_decode_close,
    vorbis_decode_frame,
};