res0.c

OGG文件格式音频数解压缩SDK.现在OGG文件格式在一些游戏开发中使用的比较多.

/********************************************************************
 *                                                                  *
 * THIS FILE IS PART OF THE OggVorbis SOFTWARE CODEC SOURCE CODE.   *
 * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS     *
 * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
 * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING.       *
 *                                                                  *
 * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2002             *
 * by the XIPHOPHORUS Company http://www.xiph.org/                  *
 *                                                                  *
 ********************************************************************

 function: residue backend 0, 1 and 2 implementation
 last mod: $Id: res0.c,v 1.48 2002/07/11 06:40:50 xiphmont Exp $

 ********************************************************************/

/* Slow, slow, slow, simpleminded and did I mention it was slow?  The
   encode/decode loops are coded for clarity and performance is not
   yet even a nagging little idea lurking in the shadows.  Oh and BTW,
   it's slow. */

#include <stdlib.h>
#include <string.h>
#include <math.h>
#include <ogg/ogg.h>
#include "vorbis/codec.h"
#include "codec_internal.h"
#include "registry.h"
#include "codebook.h"
#include "misc.h"
#include "os.h"

#ifdef TRAIN_RES
#include <stdio.h>
#endif 

typedef struct {
  vorbis_info_residue0 *info;
  
  int         parts;
  int         stages;
  codebook   *fullbooks;
  codebook   *phrasebook;
  codebook ***partbooks;

  int         partvals;
  int       **decodemap;

  ogg_int32_t      postbits;
  ogg_int32_t      phrasebits;
  ogg_int32_t      frames;

#ifdef TRAIN_RES
  int        train_seq;
  ogg_int32_t      *training_data[8][64];
  float      training_max[8][64];
  float      training_min[8][64];
  float     tmin;
  float     tmax;
#endif

} vorbis_look_residue0;

void res0_free_info(vorbis_info_residue *i){
  vorbis_info_residue0 *info=(vorbis_info_residue0 *)i;
  if(info){
    memset(info,0,sizeof(*info));
    _ogg_free(info);
  }
}

void res0_free_look(vorbis_look_residue *i){
  int j;
  if(i){

    vorbis_look_residue0 *look=(vorbis_look_residue0 *)i;

#ifdef TRAIN_RES
    {
      int j,k,l;
      for(j=0;j<look->parts;j++){
	/*fprintf(stderr,"partition %d: ",j);*/
	for(k=0;k<8;k++)
	  if(look->training_data[k][j]){
	    char buffer[80];
	    FILE *of;
	    codebook *statebook=look->partbooks[j][k];
	    
	    /* ogg_int32_t and short into the same bucket by current convention */
	    sprintf(buffer,"res_part%d_pass%d.vqd",j,k);
	    of=fopen(buffer,"a");

	    for(l=0;l<statebook->entries;l++)
	      fprintf(of,"%d:%ld\n",l,look->training_data[k][j][l]);
	    
	    fclose(of);
	    
	    /*fprintf(stderr,"%d(%.2f|%.2f) ",k,
	      look->training_min[k][j],look->training_max[k][j]);*/

	    _ogg_free(look->training_data[k][j]);
	  }
	/*fprintf(stderr,"\n");*/
      }
    }
    fprintf(stderr,"min/max residue: %g::%g\n",look->tmin,look->tmax);

    /*fprintf(stderr,"residue bit usage %f:%f (%f total)\n",
	    (float)look->phrasebits/look->frames,
	    (float)look->postbits/look->frames,
	    (float)(look->postbits+look->phrasebits)/look->frames);*/
#endif


    /*vorbis_info_residue0 *info=look->info;

    fprintf(stderr,
	    "%ld frames encoded in %ld phrasebits and %ld residue bits "
	    "(%g/frame) \n",look->frames,look->phrasebits,
	    look->resbitsflat,
	    (look->phrasebits+look->resbitsflat)/(float)look->frames);
    
    for(j=0;j<look->parts;j++){
      ogg_int32_t acc=0;
      fprintf(stderr,"\t[%d] == ",j);
      for(k=0;k<look->stages;k++)
	if((info->secondstages[j]>>k)&1){
	  fprintf(stderr,"%ld,",look->resbits[j][k]);
	  acc+=look->resbits[j][k];
	}

      fprintf(stderr,":: (%ld vals) %1.2fbits/sample\n",look->resvals[j],
	      acc?(float)acc/(look->resvals[j]*info->grouping):0);
    }
    fprintf(stderr,"\n");*/

    for(j=0;j<look->parts;j++)
      if(look->partbooks[j])_ogg_free(look->partbooks[j]);
    _ogg_free(look->partbooks);
    for(j=0;j<look->partvals;j++)
      _ogg_free(look->decodemap[j]);
    _ogg_free(look->decodemap);

    memset(look,0,sizeof(*look));
    _ogg_free(look);
  }
}

static int ilog(unsigned int v){
  int ret=0;
  while(v){
    ret++;
    v>>=1;
  }
  return(ret);
}

static int icount(unsigned int v){
  int ret=0;
  while(v){
    ret+=v&1;
    v>>=1;
  }
  return(ret);
}

#if 0
void res0_pack(vorbis_info_residue *vr,oggpack_buffer *opb){
  vorbis_info_residue0 *info=(vorbis_info_residue0 *)vr;
  int j,acc=0;
  oggpack_write(opb,info->begin,24);
  oggpack_write(opb,info->end,24);

  oggpack_write(opb,info->grouping-1,24);  /* residue vectors to group and 
					     code with a partitioned book */
  oggpack_write(opb,info->partitions-1,6); /* possible partition choices */
  oggpack_write(opb,info->groupbook,8);  /* group huffman book */

  /* secondstages is a bitmask; as encoding progresses pass by pass, a
     bitmask of one indicates this partition class has bits to write
     this pass */
  for(j=0;j<info->partitions;j++){
    if(ilog(info->secondstages[j])>3){
      /* yes, this is a minor hack due to not thinking ahead */
      oggpack_write(opb,info->secondstages[j],3); 
      oggpack_write(opb,1,1);
      oggpack_write(opb,info->secondstages[j]>>3,5); 
    }else
      oggpack_write(opb,info->secondstages[j],4); /* trailing zero */
    acc+=icount(info->secondstages[j]);
  }
  for(j=0;j<acc;j++)
    oggpack_write(opb,info->booklist[j],8);

}
#endif

/* vorbis_info is for range checking */
vorbis_info_residue *res0_unpack(vorbis_info *vi,oggpack_buffer *opb){
  int j,acc=0;
  vorbis_info_residue0 *info=_ogg_calloc(1,sizeof(*info));
  codec_setup_info     *ci=vi->codec_setup;

  info->begin=oggpack_read(opb,24);
  info->end=oggpack_read(opb,24);
  info->grouping=oggpack_read(opb,24)+1;
  info->partitions=oggpack_read(opb,6)+1;
  info->groupbook=oggpack_read(opb,8);

  for(j=0;j<info->partitions;j++){
    int cascade=oggpack_read(opb,3);
    if(oggpack_read(opb,1))
      cascade|=(oggpack_read(opb,5)<<3);
    info->secondstages[j]=cascade;

    acc+=icount(cascade);
  }
  for(j=0;j<acc;j++)
    info->booklist[j]=oggpack_read(opb,8);

  if(info->groupbook>=ci->books)goto errout;
  for(j=0;j<acc;j++)
    if(info->booklist[j]>=ci->books)goto errout;

  return(info);
 errout:
  res0_free_info(info);
  return(NULL);
}

vorbis_look_residue *res0_look(vorbis_dsp_state *vd,
			       vorbis_info_residue *vr){
  vorbis_info_residue0 *info=(vorbis_info_residue0 *)vr;
  vorbis_look_residue0 *look=_ogg_calloc(1,sizeof(*look));
  codec_setup_info     *ci=vd->vi->codec_setup;

  int j,k,acc=0;
  int dim;
  int maxstage=0;
  look->info=info;

  look->parts=info->partitions;
  look->fullbooks=ci->fullbooks;
  look->phrasebook=ci->fullbooks+info->groupbook;
  dim=look->phrasebook->dim;

  look->partbooks=_ogg_calloc(look->parts,sizeof(*look->partbooks));

  for(j=0;j<look->parts;j++){
    int stages=ilog(info->secondstages[j]);
    if(stages){
      if(stages>maxstage)maxstage=stages;
      look->partbooks[j]=_ogg_calloc(stages,sizeof(*look->partbooks[j]));
      for(k=0;k<stages;k++)
	if(info->secondstages[j]&(1<<k)){
	  look->partbooks[j][k]=ci->fullbooks+info->booklist[acc++];
#ifdef TRAIN_RES
	  look->training_data[k][j]=calloc(look->partbooks[j][k]->entries,
					   sizeof(***look->training_data));
#endif
	}
    }
  }

  look->partvals=(int)ogg_rint(ogg_pow((float)look->parts,(float)dim));
  look->stages=maxstage;
  look->decodemap=_ogg_malloc(look->partvals*sizeof(*look->decodemap));
  for(j=0;j<look->partvals;j++){
    ogg_int32_t val=j;
    ogg_int32_t mult=look->partvals/look->parts;
    look->decodemap[j]=_ogg_malloc(dim*sizeof(*look->decodemap[j]));
    for(k=0;k<dim;k++){
      ogg_int32_t deco=val/mult;
      val-=deco*mult;
      mult/=look->parts;
      look->decodemap[j][k]=deco;
    }
  }
#ifdef TRAIN_RES
  {
    static int train_seq=0;
    look->train_seq=train_seq++;
  }
#endif
  return(look);
}
#if 0
/* break an abstraction and copy some code for performance purposes */
static int local_book_besterror(codebook *book,float *a){
  int dim=book->dim,i,k,o;
  int best=0;
  encode_aux_threshmatch *tt=book->c->thresh_tree;

  /* find the quant val of each scalar */
  for(k=0,o=dim;k<dim;++k){
    float val=a[--o];
    i=tt->threshvals>>1;

    if(val<tt->quantthresh[i]){      
      if(val<tt->quantthresh[i-1]){
	for(--i;i>0;--i)
	  if(val>=tt->quantthresh[i-1])
	    break;
      }
    }else{
      
      for(++i;i<tt->threshvals-1;++i)
	if(val<tt->quantthresh[i])break;
      
    }
    
    best=(best*tt->quantvals)+tt->quantmap[i];
  }
  /* regular lattices are easy :-) */
  
  if(book->c->lengthlist[best]<=0){
    const static_codebook *c=book->c;
    int i,j;
    float bestf=0.f;
    float *e=book->valuelist;
    best=-1;
    for(i=0;i<book->entries;i++){
      if(c->lengthlist[i]>0){
	float this=0.f;
	for(j=0;j<dim;j++){
	  float val=(e[j]-a[j]);
	  this+=val*val;
	}
	if(best==-1 || this<bestf){
	  bestf=this;
	  best=i;
	}
      }
      e+=dim;
    }
  }

  {
    float *ptr=book->valuelist+best*dim;
    for(i=0;i<dim;i++)
      *a++ -= *ptr++;
  }

  return(best);
}

static int _encodepart(oggpack_buffer *opb,float *vec, int n,
		       codebook *book,ogg_int32_t *acc){
  int i,bits=0;
  int dim=book->dim;
  int step=n/dim;

  for(i=0;i<step;i++){
    int entry=local_book_besterror(book,vec+i*dim);

#ifdef TRAIN_RES
    acc[entry]++;
#endif

    bits+=vorbis_book_encode(book,entry,opb);
  }

  return(bits);
}
#endif
static ogg_int32_t **_01class(vorbis_block *vb,vorbis_look_residue *vl,
		       float **in,int ch){
  ogg_int32_t i,j,k;
  vorbis_look_residue0 *look=(vorbis_look_residue0 *)vl;
  vorbis_info_residue0 *info=look->info;
  vorbis_info           *vi=vb->vd->vi;
  codec_setup_info      *ci=vi->codec_setup;

  /* move all this setup out later */
  int samples_per_partition=info->grouping;
  int possible_partitions=info->partitions;
  int n=info->end-info->begin;
  
  int partvals=n/samples_per_partition;
  ogg_int32_t **partword=_vorbis_block_alloc(vb,ch*sizeof(*partword));
  float scale=100.0f/samples_per_partition;

  /* we find the partition type for each partition of each
     channel.  We'll go back and do the interleaved encoding in a
     bit.  For now, clarity */
 
  for(i=0;i<ch;i++){
    partword[i]=_vorbis_block_alloc(vb,n/samples_per_partition*sizeof(*partword[i]));
    memset(partword[i],0,n/samples_per_partition*sizeof(*partword[i]));
  }
  
  for(i=0;i<partvals;i++){
    int offset=i*samples_per_partition+info->begin;
    for(j=0;j<ch;j++){
      float max=0.;
      float ent=0.;
      for(k=0;k<samples_per_partition;k++){
	if(ogg_fabs(in[j][offset+k])>max)max=ogg_fabs(in[j][offset+k]);
	ent+=ogg_fabs(ogg_rint(in[j][offset+k]));
      }
      ent*=scale;
      
      for(k=0;k<possible_partitions-1;k++)
	if(max<=info->classmetric1[k] &&
	   (info->classmetric2[k]<0 || (int)ent<info->classmetric2[k]))
	  break;
      
      partword[j][i]=k;  
    }
  }

#ifdef TRAIN_RESAUX
  {
    FILE *of;
    char buffer[80];
  
    for(i=0;i<ch;i++){
      sprintf(buffer,"resaux_%d.vqd",look->train_seq);
      of=fopen(buffer,"a");
      for(j=0;j<partvals;j++)
	fprintf(of,"%ld, ",partword[i][j]);
      fprintf(of,"\n");
      fclose(of);
    }
  }
#endif
  look->frames++;

  return(partword);
}

#if 0

/* designed for stereo or other modes where the partition size is an
   integer multiple of the number of channels encoded in the current
   submap */
static ogg_int32_t **_2class(vorbis_block *vb,vorbis_look_residue *vl,float **in,
		      int ch){
  ogg_int32_t i,j,k,l;
  vorbis_look_residue0 *look=(vorbis_look_residue0 *)vl;
  vorbis_info_residue0 *info=look->info;

  /* move all this setup out later */
  int samples_per_partition=info->grouping;
  int possible_partitions=info->partitions;
  int n=info->end-info->begin;

  int partvals=n/samples_per_partition;
  ogg_int32_t **partword=_vorbis_block_alloc(vb,sizeof(*partword));

#ifdef TRAIN_RES
  FILE *of;
  char buffer[80];
#endif