source: git/kernel/fast_mult.cc @ a82c308

/****************************************
*  Computer Algebra System SINGULAR     *
****************************************/
/* $Id$ */
#include "config.h"
#include "mod2.h"
#include <polys/monomials/ring.h>
#include <kernel/fast_mult.h>
#include <polys/kbuckets.h>
#include <kernel/febase.h>

typedef poly fastmultrec(poly f, poly g, ring r);
static const int pass_option=1;
static int mults=0;
int Mults()
{
  return mults;
}
static void degsplit(poly p,int n,poly &p1,poly&p2, int vn, ring r)
{
  poly erg1_i, erg2_i;
  erg1_i=NULL;
  erg2_i=NULL;
  while(p)
  {
    if(p_GetExp(p,vn,r)>=n)
    {
      if (p1==NULL)
      {
        p1=p;
      }
      else
      {
        pNext(erg1_i)=p;
      }
      erg1_i=p;
    }
    else
    {
      if (p2==NULL)
      {
        p2=p;
      }
      else
      {
        pNext(erg2_i)=p;
      }
      erg2_i=p;
    }
    p=pNext(p);
  }
  if(erg2_i)
  {
    pNext(erg2_i)=NULL;
  }
  if (erg1_i)
  {
    pNext(erg1_i)=NULL;
  }

}
static void div_by_x_power_n(poly p, int n, int vn, ring r)
{
  while(p)
  {
    assume(p_GetExp(p,vn,r)>=n);
    int e=p_GetExp(p,vn,r);
    p_SetExp(p,vn,e-n,r);
    p=pNext(p);
  }
}

static poly do_unifastmult(poly f,int df,poly g,int dg, int vn, fastmultrec rec, ring r)
{
  int n=1;
  if ((f==NULL)||(g==NULL)) return NULL;
  //int df=pGetExp(f,vn);//pFDeg(f);
  //  int dg=pGetExp(g,vn);//pFDeg(g);

  int dm;
  if(df>dg)
  {
    dm=df;
  }
  else
  {
    dm=dg;
  }
  while(n<=dm)
    {
      n*=2;
    }
  if(n==1)
  {
    return(pp_Mult_qq(f,g,r));
  }

  int pot=n/2;
  assume(pot*2==n);


  //splitting
  poly f1=NULL;
  poly f0=NULL;//f-(x^(pot)*F1);
  degsplit(p_Copy(f,r),pot,f1,f0,vn,r);
  div_by_x_power_n(f1,pot,vn,r);

  poly g1=NULL;
  poly g0=NULL;
  degsplit(p_Copy(g,r),pot,g1,g0,vn,r);
  div_by_x_power_n(g1,pot,vn,r);

  //p00, p11
  poly p00=rec(f0,g0,r);//unifastmult(f0,g0);
  poly p11=rec(f1,g1,r);

  //construct erg, factor
  poly erg=NULL;
  poly factor=p_ISet(1,r);

  p_SetExp(factor,vn,n,r);
  erg=pp_Mult_mm(p11,factor,r);
  erg=p_Add_q(erg,p_Copy(p00,r),r);





  if((f1!=NULL) &&(f0!=NULL) &&(g0!=NULL) && (g1!=NULL))
  {
    //if(true){
    //eat up f0,f1,g0,g1
    poly s1=p_Add_q(f0,f1,r);
    poly s2=p_Add_q(g0,g1,r);
    poly pbig=rec(s1,s2,r);
    p_Delete(&s1,r);
    p_Delete(&s2,r);


    //eat up pbig
    poly sum=pbig;
    p_SetExp(factor,vn,pot,r);

    //eat up p00
    sum=p_Add_q(sum,p_Neg(p00,r),r);

    //eat up p11
    sum=p_Add_q(sum,p_Neg(p11,r),r);


    sum=p_Mult_mm(sum,factor,r);


    //eat up sum
    erg=p_Add_q(sum,erg,r);
  }
  else
  {
    //eat up f0,f1,g0,g1
    poly s1=rec(f0,g1,r);
    poly s2=rec(g0,f1,r);
    p_SetExp(factor,vn,pot,r);
    poly h=p_Mult_mm(((s1!=NULL)?s1:s2),factor,r);
    p_Delete(&f1,r);
    p_Delete(&f0,r);
    p_Delete(&g0,r);
    p_Delete(&g1,r);
    p_Delete(&p00,r);
    p_Delete(&p11,r);
    erg=p_Add_q(erg,h,r);
  }


  p_Delete(&factor,r);



  return(erg);
}

// poly do_unifastmult_buckets(poly f,poly g){




//   int n=1;
//   if ((f==NULL)||(g==NULL)) return NULL;
//   int df=pGetExp(f,1);//pFDeg(f);
//     int dg=pGetExp(g,1);//pFDeg(g);

//   int dm;
//   if(df>dg){
//     dm=df;
//   }else{
//     dm=dg;
//   }
//   while(n<=dm)
//     {
//       n*=2;
//     }
//   int pseudo_len=0;
//   if(n==1){
//     return ppMult_qq(f,g);

//   }
//   kBucket_pt erg_bucket= kBucketCreate(currRing);
//   kBucketInit(erg_bucket,NULL,0 /*pLength(P.p)*/);

//   int pot=n/2;
//   assume(pot*2==n);


//   //splitting
//   poly f1=NULL;
//   poly f0=NULL;//f-(x^(pot)*F1);
//   degsplit(pCopy(f),pot,f1,f0);
//   div_by_x_power_n(f1,pot);
//   poly g1=NULL;
//   poly g0=NULL;
//   degsplit(pCopy(g),pot,g1,g0);
//   div_by_x_power_n(g1,pot);

//   //p00
//   //p11
//   poly p00=unifastmult(f0,g0);
//   poly p11=unifastmult(f1,g1);




//   //eat up f0,f1,g0,g1
//   poly pbig=unifastmult(pAdd(f0,f1),pAdd(g0,g1));
//   poly factor=pOne();//pCopy(erg_mult);
//   pSetExp(factor,1,n);
//   pseudo_len=0;
//   kBucket_Add_q(erg_bucket,ppMult_mm(p11,factor),&pseudo_len);
//   pseudo_len=0;
//   kBucket_Add_q(erg_bucket,pCopy(p00),&pseudo_len);
//   pSetExp(factor,1,pot);

//   //eat up pbig
//   pseudo_len=0;
//   kBucket_Add_q(erg_bucket,pMult_mm(pbig,factor),&pseudo_len);
//   pseudo_len=0;
//   //eat up p00
//   kBucket_Add_q(erg_bucket,pMult_mm(pNeg(p00),factor),&pseudo_len);
//   pseudo_len=0;
//   //eat up p11
//   kBucket_Add_q(erg_bucket,pMult_mm(pNeg(p11),factor),&pseudo_len);


//   pseudo_len=0;


//   pDelete(&factor);

//   int len=0;
//   poly erg=NULL;
//   kBucketClear(erg_bucket,&erg,&len);
//   kBucketDestroy(&erg_bucket);

//   return erg;
// }

static inline int max(int a, int b)
{
  return (a>b)? a:b;
}
static inline int min(int a, int b)
{
  return (a>b)? b:a;
}
poly unifastmult(poly f,poly g, ring r)
{
  int vn=1;
  if((f==NULL)||(g==NULL)) return NULL;
  int df=p_GetExp(f,vn,r);
  int dg=p_GetExp(g,vn,r);
  if ((df==0)||(dg==0))
    return pp_Mult_qq(f,g,r);
  if (df*dg<100)
    return pp_Mult_qq(f,g,r);
  // if (df*dg>10000)
  //  return
  //    do_unifastmult_buckets(f,g);
  //else
  return do_unifastmult(f,df,g,dg,vn,unifastmult,r);

}

poly multifastmult(poly f, poly g, ring r)
{
  mults++;
  if((f==NULL)||(g==NULL)) return NULL;
  if (pLength(f)*pLength(g)<100)
    return pp_Mult_qq(f,g,r);
  //find vn
  //determine df,dg simultaneously
  int i;
  int can_i=-1;
  int can_df=0;
  int can_dg=0;
  int can_crit=0;
  for(i=1;i<=rVar(r);i++)
  {
    poly p;
    int df=0;
    int dg=0;
    //max min max Strategie
    p=f;
    while(p)
    {
      df=max(df,p_GetExp(p,i,r));
      p=pNext(p);
    }
    if(df>can_crit)
    {
      p=g;
      while(p)
      {
        dg=max(dg,p_GetExp(p,i,r));
        p=pNext(p);
      }
      int crit=min(df,dg);
      if (crit>can_crit)
      {
        can_crit=crit;
        can_i=i;
        can_df=df;
        can_dg=dg;
      }
    }
  }
  if(can_crit==0)
    return pp_Mult_qq(f,g,r);
  else
    {
      poly erg=do_unifastmult(f,can_df,g,can_dg,can_i,multifastmult,r);
      p_Normalize(erg,r);
      return(erg);
    }
}
poly pFastPower(poly f, int n, ring r)
{
  if (n==1) return f;
  if (n==0) return p_ISet(1,r);
  assume(n>=0);
  int i_max=1;
  int pot_max=0;
  while(i_max*2<=n)
  {
    i_max*=2;
    pot_max++;
  }
  int field_size=pot_max+1;
  int* int_pot_array=(int*) omAlloc(field_size*sizeof(int));
  poly* pot_array=(poly*) omAlloc(field_size*sizeof(poly));
  int i;
  int pot=1;
  //initializing int_pot
  for(i=0;i<field_size;i++)
  {
    int_pot_array[i]=pot;
    pot*=2;
  }
  //calculating pot_array
  pot_array[0]=f; //do not delete it
  for(i=1;i<field_size;i++)
  {
    poly p=pot_array[i-1];
    if(rVar(r)==1)
      pot_array[i]=multifastmult(p,p,r);
    else
      pot_array[i]=pp_Mult_qq(p,p,r);
  }



  int work_n=n;
  assume(work_n>=int_pot_array[field_size-1]);
  poly erg=p_ISet(1,r);


  //forward maybe faster, later
  // for(i=field_size-1;i>=0;i--){

//       assume(work_n<2*int_pot_array[i]);
//       if(int_pot_array[i]<=work_n){
//         work_n-=int_pot_array[i];
//         poly prod=multifastmult(erg,pot_array[i],r);
//         pDelete(&erg);
//         erg=prod;
//       }

//       if(i!=0) pDelete(&pot_array[i]);
//   }


  for(i=field_size-1;i>=0;i--)
  {

      assume(work_n<2*int_pot_array[i]);
      if(int_pot_array[i]<=work_n)
      {
        work_n-=int_pot_array[i];
        int_pot_array[i]=1;
      }
      else int_pot_array[i]=0;

  }
  for(i=0;i<field_size;i++)
  {
    if(int_pot_array[i]==1)
    {
      poly prod;
      if(rVar(r)==1)
        prod=multifastmult(erg,pot_array[i],r);
      else
        prod=pp_Mult_qq(erg,pot_array[i],r);
      pDelete(&erg);
      erg=prod;
    }
    if(i!=0) pDelete(&pot_array[i]);
  }
  //free res
  omfree(pot_array);
  omfree(int_pot_array);
  return erg;
}
static omBin lm_bin=NULL;
/*3
* compute for monomials p*q
* destroys p, keeps q
*/
static void p_MonMultMB(poly p, poly q,ring r)
{
  number x, y;
  // int i;

  y = p_GetCoeff(p,r);
  x = n_Mult(y,pGetCoeff(q),r->cf);
  n_Delete(&y,r->cf);
  p_SetCoeff0(p,x,r);
  //for (i=(currRing->N); i!=0; i--)
  //{
  //  pAddExp(p,i, pGetExp(q,i));
  //}
  //p->Order += q->Order;
  p_ExpVectorAdd(p,q,r);
}
/*3
* compute for monomials p*q
* keeps p, q
* uses bin only available in MCPower
*/
static poly p_MonMultCMB(poly p, poly q, ring r)
{
  number x;
  poly res = p_Init(r,lm_bin);

  x = n_Mult(p_GetCoeff(p,r),p_GetCoeff(q,r),r->cf);
  p_SetCoeff0(res,x,r);
  p_ExpVectorSum(res,p, q,r);
  return res;
}
static poly p_MonPowerMB(poly p, int exp, ring r)
{
  int i;

  if(!n_IsOne(p_GetCoeff(p,r),r->cf))
  {
    number x, y;
    y = p_GetCoeff(p,r);
    n_Power(y,exp,&x,r->cf);
    n_Delete(&y,r->cf);
    p_SetCoeff0(p,x,r);
  }
  for (i=rVar(r); i!=0; i--)
  {
    p_MultExp(p,i, exp,r);
  }
  p_Setm(p,r);
  return p;
}
static void buildTermAndAdd(int n,number* facult,poly* f_terms,int* exp,int f_len,kBucket_pt erg_bucket,ring r, number coef, poly & zw, poly tmp, poly** term_pot){

  int i;
  //  poly term=p_Init(r);

   //  number denom=n_Init(1,r);
//   for(i=0;i<f_len;i++){
//     if(exp[i]!=0){
//       number trash=denom;
//       denom=n_Mult(denom,facult[exp[i]],r);
//       n_Delete(&trash,r);
//     }

//   }
//   number coef=n_IntDiv(facult[n],denom,r);   //right function here?
//   n_Delete(&denom,r);
//   poly erg=p_NSet(coef,r);
  poly erg=p_Init(r,lm_bin);
  p_SetCoeff0(erg, coef,r);
  //p_NSet(n_Copy(coef,r),r);
  for(i=0;i<f_len;i++){
    if(exp[i]!=0){
      ///poly term=p_Copy(f_terms[i],r);
      poly term=term_pot[i][exp[i]];
        //tmp;
        //p_ExpVectorCopy(term,f_terms[i],r);
        //p_SetCoeff(term, n_Copy(p_GetCoeff(f_terms[i],r),r),r);

      //term->next=NULL;

      //p_MonPowerMB(term, exp[i],r);
      p_MonMultMB(erg,term,r);
      //p_Delete(&term,r);
    }

  }
  zw=erg;
}



static void MC_iterate(poly f, int n, ring r, int f_len,number* facult, int* exp,poly* f_terms,kBucket_pt erg_bucket,int pos,int sum, number coef, poly & zw, poly tmp, poly** term_pot){
  int i;

  if (pos<f_len-1)
  {
    poly zw_l=NULL;
    number new_coef;
    for(i=0;i<=n-sum;i++)
    {
      exp[pos]=i;
      if(i==0)
      {
        new_coef=n_Copy(coef,r->cf);
      }
      else
      {
        number old=new_coef;
        number old_rest=n_Init(n-sum-(i-1),r->cf);
        new_coef=n_Mult(new_coef,old_rest,r->cf);
        n_Delete(&old_rest,r->cf);
        n_Delete(&old,r->cf);
        number i_number=n_Init(i,r->cf);
        old=new_coef;
        new_coef=n_IntDiv(new_coef,i_number,r->cf);
        n_Delete(&old,r->cf);
        n_Delete(&i_number,r->cf);
      }
      //new_coef is
      //(n                          )
      //(exp[0]..exp[pos] 0 0 0 rest)
      poly zw_real=NULL;
      MC_iterate(f, n, r, f_len,facult, exp,f_terms,erg_bucket,pos+1,sum+i,new_coef,zw_real,tmp,term_pot);
      if (pos==f_len-2)
      {
        //get first small polys

        zw_real->next=zw_l;
        zw_l=zw_real;
      }
      //n_Delete(& new_coef,r->cf);
    }
    n_Delete(&new_coef,r->cf);
    if (pos==f_len-2)
    {
      int len=n-sum+1;
      kBucket_Add_q(erg_bucket,zw_l,&len);
    }
    return;
  }
  if(pos==f_len-1)
  {
    i=n-sum;
    exp[pos]=i;
    number new_coef=n_Copy(coef,r->cf);//n_IntDiv(coef,facult[i],r); //really consumed???????
    buildTermAndAdd(n,facult,f_terms,exp,f_len,erg_bucket,r, new_coef,zw, tmp,term_pot);
    // n_Delete(& new_coef,r);
  }
  assume(pos<=f_len-1);
}
poly pFastPowerMC(poly f, int n, ring r)
{
  //only char=0

  if(rChar(r)!=0)
    Werror("Char not 0, pFastPowerMC not implemented for this case");
  if (n<=1)
    Werror("not implemented for so small n, recursion fails");//should be length(f)
   if (pLength(f)<=1)
    Werror("not implemented for so small lenght of f, recursion fails");
  //  number null_number=n_Init(0,r);
  number* facult=(number*) omAlloc((n+1)*sizeof(number));
  facult[0]=n_Init(1,r->cf);
  int i;
  for(i=1;i<=n;i++)
  {
    number this_n=n_Init(i,r->cf);
    facult[i]=n_Mult(this_n,facult[i-1],r->cf);
    n_Delete(&this_n,r->cf);
  }

  lm_bin=omGetSpecBin(POLYSIZE + (r->ExpL_Size)*sizeof(long));

  kBucket_pt erg_bucket= kBucketCreate(currRing);
  kBucketInit(erg_bucket,NULL,0);
  const int f_len=pLength(f);
  int* exp=(int*)omAlloc(f_len*sizeof(int));
  //poly f_terms[f_len];
  poly* f_terms=(poly*)omAlloc(f_len*sizeof(poly));
  poly** term_potences=(poly**) omAlloc(f_len*sizeof(poly*));

  poly f_iter=f;
  for(i=0;i<f_len;i++)
  {
    f_terms[i]=f_iter;
    f_iter=pNext(f_iter);
  }
  for(i=0;i<f_len;i++)
  {
    term_potences[i]=(poly*)omAlloc((n+1)*sizeof(poly));
    term_potences[i][0]=p_ISet(1,r);
    int j;
    for(j=1;j<=n;j++){
      term_potences[i][j]=p_MonMultCMB(f_terms[i],term_potences[i][j-1],r);
    }
  }
  assume(f_iter==NULL);
  poly zw=NULL;
  poly tmp=p_Init(r);
  number one=n_Init(1,r->cf);
  MC_iterate(f,n,r,f_len,&facult[0], &exp[0], &f_terms[0],erg_bucket,0,0,one/*facult[n]*/,zw,tmp, term_potences);


  n_Delete(&one,r->cf);



  //free res

  //free facult
  for(i=0;i<=n;i++)
  {
    n_Delete(&facult[i],r->cf);
  }
  int i2;
  for (i=0;i<f_len;i++)
  {
    for(i2=0;i2<=n;i2++)
    {
      p_Delete(&term_potences[i][i2],r);
    }
    omfree(term_potences[i]);

  }
  omfree(term_potences);
  p_Delete(&tmp,r);
  omfree(exp);
  omfree(facult);
  omfree(f_terms);
  int len=0;
  poly erg;
  kBucketClear(erg_bucket,&erg, &len);
  kBucketDestroy(&erg_bucket);
  omUnGetSpecBin(&lm_bin);
  return erg;
}