xref: /dports/math/giacxcas/giac-1.6.0/src/cocoa.cc

/* -*- mode:C++ ; compile-command: "g++ -I.. -I../include -I.. -g -c -fno-strict-aliasing -DGIAC_GENERIC_CONSTANTS -DHAVE_CONFIG_H -DIN_GIAC -Wall cocoa.cc" -*- */
// Thanks to Zoltan Kovacs for motivating this work, in order to improve geogebra theorem proving
// Special thanks to Anna M. Bigatti from CoCoA team for insightfull discussions on how to choose an order for elimination
#include "giacPCH.h"

#ifndef WIN32
#define COCOA9950
#endif

#ifdef HAVE_LIBPTHREAD
#endif

#if defined(USE_GMP_REPLACEMENTS) || defined(GIAC_VECTOR)
#undef HAVE_LIBCOCOA
#endif
#ifdef HAVE_LIBCOCOA
#ifdef COCOA9950
#include <CoCoA/RingZZ.H>
#include <CoCoA/RingQQ.H>
#else
#include <CoCoA/ZZ.H>
#include <CoCoA/RingQ.H>
#endif
#include <CoCoA/FractionField.H>
#include <CoCoA/GlobalManager.H>
#include <CoCoA/SparsePolyRing.H>
#include <CoCoA/TmpF5.H>
//#include <CoCoA/io.H>
#include <CoCoA/VectorOps.H>
#include <CoCoA/obsolescent.H>
#include <CoCoA/SparsePolyOps-RingElem.H>
#include <CoCoA/SparsePolyIter.H>
#include <CoCoA/BigInt.H>
//
#include <CoCoA/symbol.H>
#include "TmpFGLM.H"
#endif
/*
 *  Copyright (C) 2007,2014 B. Parisse, Institut Fourier, 38402 St Martin d'Heres
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 3 of the License, or
 *  (at your option) any later version.
 *
 *  This program 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 General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program. If not, see <http://www.gnu.org/licenses/>.
 */
using namespace std;

#include <iostream>
#if !defined FXCG && !defined KHICAS
#include <iomanip>
#endif
#include "cocoa.h"
#include "gausspol.h"
#include "identificateur.h"
#include "giacintl.h"
#include "index.h"
#include "modpoly.h"
#ifdef HAVE_SYS_RESOURCE_H
#include <sys/resource.h>
#endif

#if defined(USE_GMP_REPLACEMENTS) || defined(GIAC_VECTOR)
#undef HAVE_LIBCOCOA
#endif

#if defined VISUALC && defined x86_64
#undef x86_64
#endif

#ifndef NO_NAMESPACE_GIAC
namespace giac {
#endif // ndef NO_NAMESPACE_GIAC

  //  vecteur trim(const vecteur & p,environment * env);

#ifdef HAVE_LIBCOCOA
  struct order_vectpoly {
    int order;
    vectpoly v;
  };

  struct Qx_I {
    CoCoA::SparsePolyRing * Qxptr;
    CoCoA::ideal * idealptr;
    CoCoA::PPOrdering * cocoa_order;
    Qx_I(): Qxptr(0), idealptr(0),cocoa_order(0){}
  };

  static bool operator < (const order_vectpoly & ov1,const order_vectpoly & ov2){
    if (ov1.order!=ov2.order)
      return ov1.order<ov2.order;
    unsigned ov1s=ov1.v.size(),ov2s=ov2.v.size();
    if (ov1s!=ov2s)
      return ov1s<ov2s;
    for (unsigned i=0;i<ov1s;++i){
      if (ov1.v[i].dim!=ov2.v[i].dim)
	return ov1.v[i].dim<ov2.v[i].dim;
      polynome p = ov1.v[i]-ov2.v[i];
      if (p.coord.empty())
	continue;
      return p.coord.front().value.islesscomplexthan(0);
    }
    return false;
  }

  static CoCoA::GlobalManager CoCoAFoundations;

  // cache here a list of already known ideals
  static std::map<order_vectpoly,Qx_I> cocoa_idealptr_map;

#ifdef COCOA9950
  static CoCoA::BigInt gen2ZZ(const gen & g){
    switch (g.type){
    case _INT_:
      return CoCoA::BigInt(g.val);
    case _ZINT:
#ifdef COCOA9950
      return CoCoA::BigIntFromMPZ(*g._ZINTptr);
      //return CoCoA::BigInt(*g._ZINTptr);
#else
      return CoCoA::BigInt(CoCoA::CopyFromMPZ,*g._ZINTptr);
#endif
    default:
      setsizeerr(gettext("Invalid giac gen -> CoCoA ZZ conversion")+g.print());
      return CoCoA::BigInt(0);
    }
  }

  static gen ZZ2gen(const CoCoA::RingElem & z){
    CoCoA::BigInt n,d;
    if (CoCoA::IsInteger(n, z))
      return gen(CoCoA::mpzref(n));
    CoCoA::RingElem znum=CoCoA::num(z),zden=CoCoA::den(z);
    if (CoCoA::IsInteger(n, znum) && CoCoA::IsInteger(d, zden))
      return gen(CoCoA::mpzref(n))/gen(CoCoA::mpzref(d));
    setsizeerr(gettext("Unable to convert CoCoA data"));
    return undef;
  }
#else
  static CoCoA::ZZ gen2ZZ(const gen & g){
    switch (g.type){
    case _INT_:
      return CoCoA::ZZ(g.val);
    case _ZINT:
      return CoCoA::ZZ(CoCoA::CopyFromMPZ,*g._ZINTptr);
    default:
      setsizeerr(gettext("Invalid giac gen -> CoCoA ZZ conversion")+g.print());
      return CoCoA::ZZ(0);
    }
  }

  static gen ZZ2gen(const CoCoA::RingElem & z){
    CoCoA::ZZ n,d;
    if (CoCoA::IsInteger(n, z))
      return gen(CoCoA::mpzref(n));
    CoCoA::RingElem znum=CoCoA::num(z),zden=CoCoA::den(z);
    if (CoCoA::IsInteger(n, znum) && CoCoA::IsInteger(d, zden))
      return gen(CoCoA::mpzref(n))/gen(CoCoA::mpzref(d));
    setsizeerr(gettext("Unable to convert CoCoA data"));
    return undef;
  }
#endif

  static CoCoA::RingElem polynome2ringelem(const polynome & p,const std::vector<CoCoA::RingElem> & x){
    if (unsigned(p.dim)>x.size())
      setdimerr();
    CoCoA::RingElem res(x[0]-x[0]); // how do you construct 0 in CoCoA?
    vector<monomial<gen> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    for (;it!=itend;++it){
      CoCoA::RingElem tmp(gen2ZZ(it->value)*power(x[0],0));
      index_t::const_iterator jt=it->index.begin(),jtend=it->index.end();
      for (int i=0;jt!=jtend;++jt,++i)
	tmp *= power(x[i],*jt);
      res += tmp;
    }
    return res;
  }

  static polynome ringelem2polynome(const CoCoA::RingElem & f,const gen & order){
    CoCoA::SparsePolyIter it=CoCoA::BeginIter(f);
    unsigned dim=CoCoA::IsEnded(it)?0:CoCoA::NumIndets(CoCoA::owner(CoCoA::PP(it)));
    polynome res(dim);
    vector<long> expv;
    index_t index(dim);
    for (;!CoCoA::IsEnded(it);++it){
      const CoCoA::RingElem & z=CoCoA::coeff(it);
      gen coeff=ZZ2gen(z);
      const CoCoA::PPMonoidElem & pp=CoCoA::PP(it);
      CoCoA::exponents(expv,pp);
      for (unsigned i=0;i<dim;++i)
	index[i]=expv[i];
      res.coord.push_back(monomial<gen>(coeff,index));
    }
    change_monomial_order(res,order); // res.tsort();
    return res;
  }

  static void vector_polynome2vector_ringelem(const vectpoly & v,const CoCoA::SparsePolyRing & Qx,vector<CoCoA::RingElem> & g){
    const vector<CoCoA::RingElem> & x = CoCoA::indets(Qx);
    g.reserve(v.size());
    vectpoly::const_iterator it=v.begin(),itend=v.end();
    for (;it!=itend;++it){
      g.push_back(polynome2ringelem(*it,x));
    }
  }
#if 0
  static vector<CoCoA::RingElem> vector_polynome2vector_ringelem(const vectpoly & v,const CoCoA::SparsePolyRing & Qx){
    vector<CoCoA::RingElem> g;
    vector_polynome2vector_ringelem(v,Qx,g);
    return g;
  }
  static vectpoly vector_ringelem2vector_polynome(const vector<CoCoA::RingElem> & g,const gen & order){
    vectpoly res;
    vector_ringelem2vector_polynome(g,res,order);
    return res;
  }
#endif
  static void vector_ringelem2vector_polynome(const vector<CoCoA::RingElem> & g, vectpoly & res,const gen & order){
    vector<CoCoA::RingElem>::const_iterator it=g.begin(),itend=g.end();
    res.reserve(itend-it);
    for (;it!=itend;++it)
      res.push_back(ringelem2polynome(*it,order));
    sort(res.begin(),res.end(),tensor_is_strictly_greater<gen>);
    reverse(res.begin(),res.end());
  }

  static Qx_I get_or_make_idealptr(const vectpoly & v,const gen & order){
    if (order.type!=_INT_ || v.empty())
      settypeerr();
    order_vectpoly ov;
    ov.v=v;
    ov.order=order.val;
    std::map<order_vectpoly,Qx_I>::const_iterator it=cocoa_idealptr_map.find(ov),itend=cocoa_idealptr_map.end();
    if (it!=itend)
      return it->second;
    int d=v[0].dim;
    Qx_I qx_i;
    if (order.type==_INT_ && order.val!=0){
      switch (order.val){
      case _PLEX_ORDER:
	qx_i.cocoa_order = new CoCoA::PPOrdering(CoCoA::NewLexOrdering(d));
	break;
      case _TDEG_ORDER:
	qx_i.cocoa_order = new CoCoA::PPOrdering(CoCoA::NewStdDegLexOrdering(d));
	break;
      default:
	qx_i.cocoa_order = new CoCoA::PPOrdering(CoCoA::NewStdDegRevLexOrdering(d));
      }
      qx_i.Qxptr = new CoCoA::SparsePolyRing(CoCoA::NewPolyRing(
#ifdef COCOA9950
								CoCoA::RingQQ(),
#else
								CoCoA::RingQ(),
#endif
								CoCoA::SymbolRange("x",0,d-1),*qx_i.cocoa_order));
    }
    else
      qx_i.Qxptr = new CoCoA::SparsePolyRing(CoCoA::NewPolyRing(
#ifdef COCOA9950
								CoCoA::RingQQ(),
#else
								CoCoA::RingQ(),
#endif
								CoCoA::SymbolRange("x",0,d-1)));
    vector<CoCoA::RingElem> g;
    vector_polynome2vector_ringelem(v,*qx_i.Qxptr,g);
    qx_i.idealptr=new CoCoA::ideal(*qx_i.Qxptr,g);
    cocoa_idealptr_map[ov]=qx_i;
    // if (cocoa_order)
    //  delete cocoa_order;
    return qx_i;
  }

  // add a dimension so that p is homogeneous of degree d
  static void homogeneize(polynome & p,int deg){
    vector<monomial<gen> >::iterator it=p.coord.begin(),itend=p.coord.end();
    int n;
    for (;it!=itend;++it){
      index_t i=it->index.iref();
      n=total_degree(i);
      i.push_back(deg-n);
      it->index=i;
    }
    ++p.dim;
  }

  static void homogeneize(vectpoly & v){
    vectpoly::iterator it=v.begin(),itend=v.end();
    int d=0;
    for (;it!=itend;++it){
      d=giacmax(d,total_degree(*it));
    }
    for (it=v.begin();it!=itend;++it){
      homogeneize(*it,d);
    }
  }

  static void unhomogeneize(polynome & p){
    vector<monomial<gen> >::iterator it=p.coord.begin(),itend=p.coord.end();
    for (;it!=itend;++it){
      index_t i=it->index.iref();
      i.pop_back();
      it->index=i;
    }
    --p.dim;
  }

  static void unhomogeneize(vectpoly & v){
    vectpoly::iterator it=v.begin(),itend=v.end();
    for (;it!=itend;++it){
      unhomogeneize(*it);
    }
  }

  bool f5(vectpoly & v,const gen & order){
    homogeneize(v);
    CoCoA::SparsePolyRing Qx = CoCoA::NewPolyRing(
#ifdef COCOA9950
								CoCoA::RingQQ(),
#else
								CoCoA::RingQ(),
#endif
						  CoCoA::SymbolRange("x",0,v[0].dim-1));
    vector<CoCoA::RingElem> g;
    vector_polynome2vector_ringelem(v,Qx,g);
    CoCoA::ideal I(Qx,g);
    vector<CoCoA::RingElem> gb;
    CoCoA::F5(gb,I);
    CoCoA::operator<<(cout,gb);
    cout << '\n';
    v.clear();
    vector_ringelem2vector_polynome(gb,v,order);
    unhomogeneize(v);
    return true;
  }

  // order may be 0 (use cocoa default and convert to lex for 0-dim ideals)
  // or _PLEX_ORDER (lexicographic) or _TDEG_ORDER (total degree)
  // or _REVLEX_ORDER (total degree then reverse of lexicographic)
  bool cocoa_gbasis(vectpoly & v,const gen & order){
    Qx_I qx_i = get_or_make_idealptr(v,order);
    int d=v[0].dim;
    vector<CoCoA::RingElem> gb=TidyGens(*qx_i.idealptr);
    // CoCoA::operator<<(cout,gb);
    // cout << '\n';
    // 0-dim ideals convert to lexicographic order using CoCoA FGLM routine
    // otherwise leaves revlex order
    vector<CoCoA::RingElem> NewGBasis;
    if (order.type==_INT_ && order.val!=0)
      NewGBasis=gb;
    else {
      CoCoA::PPOrdering NewOrdering = CoCoA::NewLexOrdering(d);
      try {
	CoCoADortmund::FGLMBasisConversion(NewGBasis, gb, NewOrdering);
      } catch (...){
	v.clear();
	vector_ringelem2vector_polynome(gb,v,order);
	return false;
      }
      // CoCoA::operator<<(cout,NewGBasis);
      // cout << '\n';
    }
    v.clear();
    vector_ringelem2vector_polynome(NewGBasis,v,order);
    // reverse(v.begin(),v.end());
    // unhomogeneize(v);
    return true;
  }

  vecteur cocoa_in_ideal(const vectpoly & r,const vectpoly & v,const gen & order){
    Qx_I qx_i = get_or_make_idealptr(v,order);
    vector<CoCoA::RingElem> cocoa_r;
    vector_polynome2vector_ringelem(r,*qx_i.Qxptr,cocoa_r);
    int s=cocoa_r.size();
    gen tmp(-1);
    tmp.subtype=_INT_BOOLEAN;
    vecteur res(s,tmp);
    for (int i=0;i<s;++i)
      res[i].val=CoCoA::IsElem(cocoa_r[i],*qx_i.idealptr);
    return res;
  }

  bool cocoa_greduce(const vectpoly & r,const vectpoly & v,const gen & order,vectpoly & res){
    Qx_I qx_i = get_or_make_idealptr(v,order);
    vector<CoCoA::RingElem> cocoa_r;
    vector_polynome2vector_ringelem(r,*qx_i.Qxptr,cocoa_r);
    int s=cocoa_r.size();
    polynome tmp;
    for (int i=0;i<s;++i){
      CoCoA::RingElem tmpc=CoCoA::NF(cocoa_r[i],*qx_i.idealptr);
      tmp=ringelem2polynome(tmpc,order);
      res.push_back(tmp);
    }
    return true;
  }

#else // HAVE_LIBCOCOA

  bool f5(vectpoly & v,const gen & ordre){
    return false;
  }

  bool cocoa_gbasis(vectpoly & v,const gen & ordre){
    return false;
  }

  vecteur cocoa_in_ideal(const vectpoly & r,const vectpoly & v,const gen & ordre){
    return vecteur(r.size(),-1);
  }

  bool cocoa_greduce(const vectpoly & r,const vectpoly & v,const gen & order,vectpoly & res){
    return false;
  }

#endif  // HAVE_LIBCOCOA

  struct paire {
    unsigned first;
    unsigned second;
    bool live; // set to false to defer the pair for reduction at end
    paire(unsigned f,unsigned s):first(f),second(s),live(true){}
    paire(const paire & p):first(p.first),second(p.second),live(p.live) {}
    paire():first(-1),second(-1) {}
  };

  ostream & operator << (ostream & os,const paire & p){
    return os << "<" << p.first << "," << p.second << ">";
  }

  inline bool operator == (const paire & a,const paire &b){
    return a.first==b.first && a.second==b.second;
  }

#ifdef WORDS_BIGENDIAN // autoconf macro defines this (thanks to Julien Puydt for pointing this and checking for s390x architecture)
#define BIGENDIAN
#endif

#if !defined CAS38_DISABLED && !defined FXCG
  //#define GBASIS_SELECT_TOTAL_DEGREE
#if GROEBNER_VARS!=15 && !defined BIGENDIAN // double revlex ordering is not compatible with indices swapping
#define GBASIS_SWAP
#endif
  // minimal numbers of pair to reduce simultaneously with f4buchberger
#ifdef __APPLE__
  #define GBASISF4_BUCHBERGER 0 // temp. disabled on mac
#else
  #define GBASISF4_BUCHBERGER 4
#endif

#define GBASIS_POSTF4BUCHBERGER 0 // 0 means final simplification at the end, 1 at each loop

  // #define GIAC_GBASIS_REDUCTOR_MAXSIZE 10 // max size for keeping a reductor even if it should be removed from gbasis

  // if GIAC_SHORTSHIFTTYPE is defined, sparse matrix is using shift index
  // coded on 2 bytes -> FIXME segfault for cyclic9

#define GIAC_SHORTSHIFTTYPE 16
  //#define GIAC_GBASIS_DELAYPAIRS

  void swap_indices(short * tab){
    swap(tab[1],tab[3]);
    swap(tab[4],tab[7]);
    swap(tab[5],tab[6]);
    swap(tab[8],tab[11]);
    swap(tab[9],tab[10]);
#if GROEBNER_VARS>11
    swap(tab[12],tab[15]);
    swap(tab[13],tab[14]);
#endif
  }

  template<class T>
  void swap_indices14(T * tab){
    swap(tab[2],tab[7]);
    swap(tab[3],tab[6]);
    swap(tab[4],tab[5]);
    swap(tab[8],tab[15]);
    swap(tab[9],tab[14]);
    swap(tab[10],tab[13]);
    swap(tab[11],tab[12]);
  }

  void swap_indices11(short * tab){
    swap(tab[1],tab[3]);
    swap(tab[4],tab[7]);
    swap(tab[5],tab[6]);
    swap(tab[8],tab[11]);
    swap(tab[9],tab[10]);
  }

  void swap_indices15_revlex(short * tab){
    swap(tab[1],tab[3]);
    swap(tab[4],tab[7]);
    swap(tab[5],tab[6]);
    swap(tab[8],tab[11]);
    swap(tab[9],tab[10]);
    swap(tab[12],tab[15]);
    swap(tab[13],tab[14]);
  }

  void swap_indices15_3(short * tab){
    swap(tab[1],tab[3]);
    swap(tab[5],tab[7]);
    swap(tab[8],tab[11]);
    swap(tab[9],tab[10]);
    swap(tab[12],tab[15]);
    swap(tab[13],tab[14]);
  }

  void swap_indices15_7(short * tab){
    swap(tab[1],tab[3]);
    swap(tab[4],tab[7]);
    swap(tab[5],tab[6]);
    swap(tab[9],tab[11]);
    swap(tab[12],tab[15]);
    swap(tab[13],tab[14]);
  }

  void swap_indices15_11(short * tab){
    swap(tab[1],tab[3]);
    swap(tab[4],tab[7]);
    swap(tab[5],tab[6]);
    swap(tab[8],tab[11]);
    swap(tab[9],tab[10]);
    swap(tab[13],tab[15]);
  }

  void swap_indices15(short * tab,int o){
    if (o==_REVLEX_ORDER){
      swap_indices15_revlex(tab);
      return;
    }
    if (o==_3VAR_ORDER){
      swap_indices15_3(tab);
      return;
    }
    if (o==_7VAR_ORDER){
      swap_indices15_7(tab);
      return;
    }
    if (o==_11VAR_ORDER){
      swap_indices15_11(tab);
      return;
    }
  }

  // #define GIAC_CHARDEGTYPE should be in solve.h
#if defined BIGENDIAN && defined GIAC_CHARDEGTYPE
#undef GIAC_CHARDEGTYPE
#endif

#ifdef GIAC_CHARDEGTYPE
  typedef unsigned char degtype; // type for degree for large number of variables
  #define degratio 8
  #define degratiom1 7
#else
  typedef short degtype; // type for degree for large number of variables
  #define degratio 4
  #define degratiom1 3
#endif

#ifndef GIAC_64VARS
#define GBASIS_NO_OUTPUT
#endif

#define GIAC_RDEG
  // #define GIAC_HASH
#if defined GIAC_HASH && defined HASH_MAP_NAMESPACE
#define GIAC_RHASH
#endif

#ifdef GIAC_RHASH
  class tdeg_t64;
  class tdeg_t64_hash_function_object {
  public:
    size_t operator () (const tdeg_t64 & ) const; // defined at the end of this file
    tdeg_t64_hash_function_object() {};
  };

  typedef HASH_MAP_NAMESPACE::hash_map< tdeg_t64,int,tdeg_t64_hash_function_object > tdeg_t64_hash_t ;
#endif

#ifndef VISUALC
  #define GIAC_ELIM
#endif
  short hash64tab[]={1933,1949,1951,1973,1979,1987,1993,1997,1999,2003,2011,2017,2027,2029,2039,2053,2063,2069,2081,2083,2087,2089,2099,2111,2113,2129,2131,2137,2141,2143,2153,2161,2179,2203,2207,2213,2221,2237,2239,2243,2251,2267,2269,2273,2281,2287,2293,2297,2309,2311,2333,2339,2341,2347,2351,2357,2371,2377,2381,2383,2389,2393,2399,2411};

  // storing indices in reverse order to have tdeg_t_greater access them
  // in increasing order seems slower (cocoa.cc.64)
  struct tdeg_t64 {
    bool vars64() const {return true;}
#ifdef GIAC_RHASH
    int hash_index(void * ptr_) const {
      if (!ptr_) return -1;
      tdeg_t64_hash_t & h=*(tdeg_t64_hash_t *) ptr_;
      tdeg_t64_hash_t::const_iterator it=h.find(*this),itend=h.end();
      if (it==itend)
	return -1;
      return it->second;
    }
    bool add_to_hash(void *ptr_,int no) const {
      if (!ptr_) return false;
      tdeg_t64_hash_t & h=*(tdeg_t64_hash_t *) ptr_;
      h[*this]=no;
      return true;
    }
#else
  int hash_index(void * ptr_) const { return -1; }
  bool add_to_hash(void *ptr_,int no) const { return false; }
#endif
    void dbgprint() const;
    // data
#ifdef GIAC_64VARS
    union {
      short tab[GROEBNER_VARS+1];
      struct {
	short tdeg; // actually it's twice the total degree+1
	short tdeg2;
	order_t order_;
	longlong * ui;
#ifdef GIAC_HASH
	longlong hash;
#endif
#ifdef GIAC_ELIM
	ulonglong elim; // used for elimination order (modifies revlex/revlex)
#endif
      };
    };
    //int front() const { if (tdeg % 2) return (*(ui+1)) & 0xffff; else return order_.o==_PLEX_ORDER?tab[0]:tab[1];}
    tdeg_t64(const tdeg_t64 & a){
      if (a.tab[0]%2){
	tdeg=a.tdeg;
	tdeg2=a.tdeg2;
	order_=a.order_;
	ui=a.ui;
#ifdef GIAC_HASH
	hash=a.hash;
#endif
#ifdef GIAC_ELIM
	elim=a.elim;
#endif
	++(*ui);
      }
      else {
	longlong * ptr = (longlong *) tab;
	longlong * aptr = (longlong *) a.tab;
	ptr[0]=aptr[0];
	ptr[1]=aptr[1];
	ptr[2]=aptr[2];
	ptr[3]=aptr[3];
      }
    }
#ifdef GIAC_ELIM
    void compute_elim(longlong * ptr,longlong * ptrend){
      elim=0;
      if (order_.o==_PLEX_ORDER) // don't use elim for plex
	return;
      if (tdeg>31){
	elim=0x1fffffffffffffffULL;
	return;
      }
      bool tdegcare=false;
      if (ptr<ptrend-3)
	ptr=ptrend-3;
      for (--ptrend;ptr<=ptrend;--ptrend){
	longlong x=*ptrend;
	elim <<= 20;
	elim += (x&0xffff)+(((x>>16)&0xffff)<<5)+(((x>>32)&0xffff)<<10)+(((x>>48)&0xffff)<<15);
      }
    }
#endif
    void compute_degs(){
      if (tab[0]%2){
	longlong * ptr=ui+1;
	tdeg=0;
	int firstblock=order_.o;
	if (firstblock!=_3VAR_ORDER && firstblock<_7VAR_ORDER)
	  firstblock=order_.dim;
	longlong * ptrend=ui+1+(firstblock+degratiom1)/degratio;
#ifdef GIAC_HASH
	hash=0;
#endif
#ifdef GIAC_ELIM
	compute_elim(ptr,ptrend);
#endif
	int i=0;
	for (;ptr!=ptrend;i+=4,++ptr){
	  longlong x=*ptr;
#ifdef GIAC_CHARDEGTYPE
	  tdeg += ((x+(x>>8)+(x>>16)+(x>>24)+(x>>32)+(x>>40)+(x>>48)+(x>>56))&0xff);
#else
	  tdeg += ((x+(x>>16)+(x>>32)+(x>>48))&0xffff);
#endif
#ifdef GIAC_HASH
	  hash += (x&0xffff)*hash64tab[i]+((x>>16)&0xffff)*hash64tab[i+1]+((x>>32)&0xffff)*hash64tab[i+2]+(x>>48)*hash64tab[i+3];
#endif
	}
#ifdef GIAC_ELIM
	if (tdeg>=16)
	  elim=0x1fffffffffffffffULL;
#endif
	tdeg=2*tdeg+1;
	tdeg2=0;
	ptrend=ui+1+(order_.dim+degratiom1)/degratio;
	for (;ptr!=ptrend;i+=4,++ptr){
	  longlong x=*ptr;
#ifdef GIAC_CHARDEGTYPE
	  tdeg2 += ((x+(x>>8)+(x>>16)+(x>>24)+(x>>32)+(x>>40)+(x>>48)+(x>>56))&0xff);
#else
	  tdeg2 += ((x+(x>>16)+(x>>32)+(x>>48))&0xffff);
#endif
#ifdef GIAC_HASH
	  hash += (x&0xffff)*hash64tab[i]+((x>>16)&0xffff)*hash64tab[i+1]+((x>>32)&0xfff)*hash64tab[i+2]+(x>>48)*hash64tab[i+3];
#endif
	}
      }
    }
    ~tdeg_t64(){
      if (tab[0]%2){
	--(*ui);
	if (*ui==0)
	  free(ui);
      }
    }
    tdeg_t64 & operator = (const tdeg_t64 & a){
      if (tab[0] % 2){
	--(*ui);
	if (*ui==0)
	  free(ui);
	if (a.tab[0] % 2){
	  tdeg=a.tdeg;
	  tdeg2=a.tdeg2;
	  order_=a.order_;
	  ui=a.ui;
#ifdef GIAC_HASH
	  hash=a.hash;
#endif
#ifdef GIAC_ELIM
	  elim=a.elim;
#endif
	  ++(*ui);
	  return *this;
	}
      }
      else {
	if (a.tab[0]%2){
	  ++(*a.ui);
	}
      }
      longlong * ptr = (longlong *) tab;
      longlong * aptr = (longlong *) a.tab;
      ptr[0]=aptr[0];
      ptr[1]=aptr[1];
      ptr[2]=aptr[2];
      ptr[3]=aptr[3];
      return *this;
    }
#else
    short tab[GROEBNER_VARS+1];
    int front(){ return tab[1];}
#endif
    // methods
    inline unsigned selection_degree(order_t order) const {
#ifdef GBASIS_SELECT_TOTAL_DEGREE
      return total_degree(order);
#endif
#ifdef GIAC_64VARS
      if (tab[0]%2)
	return tdeg/2;
#endif
      return tdeg;
    }
    inline unsigned total_degree(order_t order) const {
#ifdef GIAC_64VARS
      if (tab[0]%2)
	return tdeg/2+tdeg2;
#endif
      // works only for revlex and tdeg
#if 0
      if (order==_REVLEX_ORDER || order==_TDEG_ORDER)
	return tab[0];
      if (order==_3VAR_ORDER)
	return (tab[0] << 16)+tab[4];
      if (order==_7VAR_ORDER)
	return (tab[0] << 16) +tab[8];
      if (order==_11VAR_ORDER)
	return (tab[0] << 16) +tab[12];
#endif
      return tab[0];
    }
    // void set_total_degree(unsigned d) { tab[0]=d;}
    tdeg_t64() {
      longlong * ptr = (longlong *) tab;
      ptr[2]=ptr[1]=ptr[0]=0;
#if GROEBNER_VARS>11
      ptr[3]=0;
#endif
    }
    tdeg_t64(int i){
      longlong * ptr = (longlong *) tab;
      ptr[2]=ptr[1]=ptr[0]=0;
#if GROEBNER_VARS>11
      ptr[3]=0;
#endif
    }
    void get_tab(short * ptr,order_t order) const {
#ifdef GIAC_64VARS
      if (tab[0]%2){ // copy only 16 first
	degtype * ptr_=(degtype *)(ui+1);
	for (unsigned i=0;i<=GROEBNER_VARS;++i)
	  ptr[i]=ptr_[i];
	return;
      }
#endif
      for (unsigned i=0;i<=GROEBNER_VARS;++i)
	ptr[i]=tab[i];
#ifdef GIAC_64VARS
      ptr[0]/=2;
#endif
#ifdef GBASIS_SWAP
      swap_indices(ptr);
#endif
    }
    tdeg_t64(const index_m & lm,order_t order){
#ifdef GIAC_64VARS
      if (lm.size()>GROEBNER_VARS){
	ui=(longlong *)malloc((1+(lm.size()+degratiom1)/degratio)*sizeof(longlong));
	longlong* ptr=ui;
	*ptr=1; ++ ptr;
#ifdef GIAC_CHARDEGTYPE
	for (int i=0;i<lm.size();){
	  unsigned char tableau[8]={0,0,0,0,0,0,0,0};
	  tableau[0]=lm[i];
	  ++i;
	  if (i<lm.size())
	    tableau[1] = lm[i];
	  ++i;
	  if (i<lm.size())
	    tableau[2]= lm[i];
	  ++i;
	  if (i<lm.size())
	    tableau[3]= lm[i];
	  ++i;
	  if (i<lm.size())
	    tableau[4]= lm[i];
	  ++i;
	  if (i<lm.size())
	    tableau[5]= lm[i];
	  ++i;
	  if (i<lm.size())
	    tableau[6]= lm[i];
	  ++i;
	  if (i<lm.size())
	    tableau[7]= lm[i];
	  ++i;
	  *ptr = * (longlong *)tableau ;
	  ++ptr;
	}
#else
	for (int i=0;i<lm.size();){
	  longlong x=lm[i];
	  ++i;
	  x <<= 16;
	  if (i<lm.size())
	    x += lm[i];
	  ++i;
	  x <<= 16;
	  if (i<lm.size())
	    x += lm[i];
	  ++i;
	  x <<= 16;
	  if (i<lm.size())
	    x += lm[i];
#ifndef BIGENDIAN
	  x = (x>>48) | (((x>>32)&0xffff)<<16) | (((x>>16)&0xffff)<<32) | ((x&0xffff)<<48);
#endif
	  *ptr = x;
	  ++ptr;
	  ++i;
	}
#endif // GIAC_CHARDEGTYPE
	if (order.o==_3VAR_ORDER || order.o>=_7VAR_ORDER){
	  tdeg=2*nvar_total_degree(lm,order.o)+1;
	  tdeg2=sum_degree_from(lm,order.o);
	}
	else {
	  tdeg=short(2*lm.total_degree()+1);
	  tdeg2=0;
	}
	order_=order;
#if 1 // def GIAC_HASH
	compute_degs(); // for hash
#else
#ifdef GIAC_ELIM
	ptr=ui+1;
	int firstblock=order_.o;
	if (firstblock!=_3VAR_ORDER && firstblock<_7VAR_ORDER)
	  firstblock=order_.dim;
	compute_elim(ptr,ptr+(firstblock+degratiom1)/degratio);
#endif
#endif
	return;
      }
#endif // GIAC_64VARS
      longlong * ptr_ = (longlong *) tab;
      ptr_[2]=ptr_[1]=ptr_[0]=0;
      short * ptr=tab;
#if GROEBNER_VARS>11
      ptr_[3]=0;
#endif
      // tab[GROEBNER_VARS]=order;
#if GROEBNER_VARS==15
      if (order.o==_3VAR_ORDER){
#ifdef GIAC_64VARS
	ptr[0]=2*(lm[0]+lm[1]+lm[2]);
#else
	ptr[0]=lm[0]+lm[1]+lm[2];
#endif
	ptr[1]=lm[2];
	ptr[2]=lm[1];
	ptr[3]=lm[0];
	ptr +=5;
	short t=0;
	vector<deg_t>::const_iterator it=lm.begin()+3,itend=lm.end();
	for (--itend,--it;it!=itend;++ptr,--itend){
	  t += *itend;
	  *ptr=*itend;
	}
	tab[4]=t;
	return;
      }
      if (order.o==_7VAR_ORDER){
#ifdef GIAC_64VARS
	ptr[0]=2*(lm[0]+lm[1]+lm[2]+lm[3]+lm[4]+lm[5]+lm[6]);
#else
	ptr[0]=lm[0]+lm[1]+lm[2]+lm[3]+lm[4]+lm[5]+lm[6];
#endif
	ptr[1]=lm[6];
	ptr[2]=lm[5];
	ptr[3]=lm[4];
	ptr[4]=lm[3];
	ptr[5]=lm[2];
	ptr[6]=lm[1];
	ptr[7]=lm[0];
	ptr +=9;
	short t=0;
	vector<deg_t>::const_iterator it=lm.begin()+7,itend=lm.end();
	for (--itend,--it;it!=itend;++ptr,--itend){
	  t += *itend;
	  *ptr=*itend;
	}
	tab[8]=t;
	return;
      }
      if (order.o==_11VAR_ORDER){
#ifdef GIAC_64VARS
	ptr[0]=2*(lm[0]+lm[1]+lm[2]+lm[3]+lm[4]+lm[5]+lm[6]+lm[7]+lm[8]+lm[9]+lm[10]);
#else
	ptr[0]=lm[0]+lm[1]+lm[2]+lm[3]+lm[4]+lm[5]+lm[6]+lm[7]+lm[8]+lm[9]+lm[10];
#endif
	ptr[1]=lm[10];
	ptr[2]=lm[9];
	ptr[3]=lm[8];
	ptr[4]=lm[7];
	ptr[5]=lm[6];
	ptr[6]=lm[5];
	ptr[7]=lm[4];
	ptr[8]=lm[3];
	ptr[9]=lm[2];
	ptr[10]=lm[1];
	ptr[11]=lm[0];
	ptr += 13;
	short t=0;
	vector<deg_t>::const_iterator it=lm.begin()+11,itend=lm.end();
	for (--itend,--it;it!=itend;++ptr,--itend){
	  t += *itend;
	  *ptr=*itend;
	}
	tab[12]=t;
	return;
      }
#endif
      vector<deg_t>::const_iterator it=lm.begin(),itend=lm.end();
      if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
	*ptr=sum_degree(lm);
	++ptr;
      }
      if (order.o==_REVLEX_ORDER){
	for (--itend,--it;it!=itend;++ptr,--itend)
	  *ptr=*itend;
      }
      else {
	for (;it!=itend;++ptr,++it)
	  *ptr=*it;
      }
#ifdef GBASIS_SWAP
      swap_indices(tab);
#endif
#ifdef GIAC_64VARS
      *tab *=2;
#endif
    }
  };

  typedef map<tdeg_t64,unsigned> annuaire;

#ifdef NSPIRE
  template<class T>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os,const tdeg_t64 & x){
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
      os << "[";
      const longlong * ptr=x.ui+1,*ptrend=ptr+(x.order_.dim+degratiom1)/degratio;
      for (;ptr!=ptrend;++ptr){
	longlong x=*ptr;
#ifdef BIGENDIAN
	os << ((x>>48) &0xffff)<< "," << ((x>>32) & 0xffff) << "," << ((x>>16) & 0xffff) << "," << ((x) & 0xffff) << ",";
#else
	os << ((x) &0xffff)<< "," << ((x>>16) & 0xffff) << "," << ((x>>32) & 0xffff) << "," << ((x>>48) & 0xffff) << ",";
#endif
      }
      return os << "]";
    }
#endif
    os << "[";
    for (unsigned i=0; i<=GROEBNER_VARS;++i){
      os << x.tab[i] << ",";
    }
    return os << "]";
  }
#else
  ostream & operator << (ostream & os,const tdeg_t64 & x){
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
      // debugging
      tdeg_t64 xsave(x); xsave.compute_degs();
      if (xsave.tdeg!=x.tdeg || xsave.tdeg2!=x.tdeg2)
	os << "degree error " ;
      os << "[";
      const longlong * ptr=x.ui+1,*ptrend=ptr+(x.order_.dim+degratiom1)/degratio;
      for (;ptr!=ptrend;++ptr){
	longlong x=*ptr;
#ifdef BIGENDIAN
	os << ((x>>48) &0xffff)<< "," << ((x>>32) & 0xffff) << "," << ((x>>16) & 0xffff) << "," << ((x) & 0xffff) << ",";
#else
	os << ((x) &0xffff)<< "," << ((x>>16) & 0xffff) << "," << ((x>>32) & 0xffff) << "," << ((x>>48) & 0xffff) << ",";
#endif
      }
      return os << "]";
    }
#endif
    os << "[";
    for (unsigned i=0; i<=GROEBNER_VARS;++i){
      os << x.tab[i] << ",";
    }
    return os << "]";
  }
#endif
  void tdeg_t64::dbgprint() const { COUT << * this << '\n'; }
  tdeg_t64 operator + (const tdeg_t64 & x,const tdeg_t64 & y);
  tdeg_t64 & operator += (tdeg_t64 & x,const tdeg_t64 & y){
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(y.tab[0]%2)){
	y.dbgprint();
	COUT << "erreur" << '\n';
      }
#endif
      return x=x+y;
    }
#endif
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    xtab[0]+=ytab[0];
    xtab[1]+=ytab[1];
    xtab[2]+=ytab[2];
#if GROEBNER_VARS>11
    xtab[3]+=ytab[3];
#endif
#else
    for (unsigned i=0;i<=GROEBNER_VARS;++i)
      x.tab[i]+=y.tab[i];
#endif
    return x;
  }

  inline tdeg_t64 dynamic_plus(const tdeg_t64 & x,const tdeg_t64 & y){
    tdeg_t64 res;
    res.order_=x.order_;
    res.ui=(longlong *)malloc((1+(x.order_.dim+degratiom1)/degratio)*sizeof(longlong));
    res.ui[0]=1;
    const longlong * xptr=x.ui+1,*xend=xptr+(x.order_.dim+degratiom1)/degratio,*yptr=y.ui+1;
    longlong * resptr=res.ui+1;
    for (;xptr!=xend;++resptr,++yptr,++xptr)
      *resptr=*xptr+*yptr;
#if 1
    res.tdeg=1+2*(x.tdeg/2+y.tdeg/2);
    res.tdeg2=x.tdeg2+y.tdeg2;
#ifdef GIAC_HASH
    res.hash=x.hash+y.hash;
#endif
#ifdef GIAC_ELIM
    if (res.tdeg>=33)
      res.elim=0x1fffffffffffffffULL;
    else
      res.elim=x.elim+y.elim;
#endif
#else
    res.tdeg=1;
    res.compute_degs();
#endif
    return res;
  }

  tdeg_t64 operator + (const tdeg_t64 & x,const tdeg_t64 & y){
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(y.tab[0]%2))
	COUT << "erreur" << '\n';
#endif
      return dynamic_plus(x,y);
    }
#endif
    tdeg_t64 res(x);
    return res += y;
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]+ytab[0];
    ztab[1]=xtab[1]+ytab[1];
    ztab[2]=xtab[2]+ytab[2];
#if GROEBNER_VARS>11
    ztab[3]=xtab[3]+ytab[3];
#endif
#else
    for (unsigned i=0;i<=GROEBNER_VARS;++i)
      res.tab[i]=x.tab[i]+y.tab[i];
#endif
    return res;
  }
  inline void add(const tdeg_t64 & x,const tdeg_t64 & y,tdeg_t64 & res,int dim){
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(y.tab[0]%2))
	COUT << "erreur" << '\n';
#endif
      if (res.tab[0]%2 && res.ui[0]==1){
	const longlong * xptr=x.ui+1,*xend=xptr+(x.order_.dim+degratiom1)/degratio,*yptr=y.ui+1;
	longlong * resptr=res.ui+1;
#ifndef GIAC_CHARDEGTYPE
	*resptr=*xptr+*yptr;++resptr,++yptr,++xptr;
	*resptr=*xptr+*yptr;++resptr,++yptr,++xptr;
	*resptr=*xptr+*yptr;++resptr,++yptr,++xptr;
	*resptr=*xptr+*yptr;++resptr,++yptr,++xptr;
#endif
	for (;xptr!=xend;++resptr,++yptr,++xptr)
	  *resptr=*xptr+*yptr;
#if 1
	res.tdeg=1+2*(x.tdeg/2+y.tdeg/2);
	res.tdeg2=x.tdeg2+y.tdeg2;
#ifdef GIAC_HASH
	res.hash=x.hash+y.hash;
#endif
#ifdef GIAC_ELIM
	if (res.tdeg>=33)
	  res.elim=0x1fffffffffffffffULL;
	else
	  res.elim=x.elim+y.elim;
#endif
#else
	res.tdeg=1;
	res.compute_degs();
#endif
      }
      else
	res=dynamic_plus(x,y);
      return;
    }
#endif
#if 0 // def GIAC_64VARS
    if (x.tab[0]%2){
      res = x;
      res += y;
      return;
    }
#endif
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]+ytab[0];
    ztab[1]=xtab[1]+ytab[1];
    ztab[2]=xtab[2]+ytab[2];
#if GROEBNER_VARS>11
    ztab[3]=xtab[3]+ytab[3];
#endif
#else
    for (unsigned i=0;i<=dim;++i)
      res.tab[i]=x.tab[i]+y.tab[i];
#endif
  }

  tdeg_t64 operator - (const tdeg_t64 & x,const tdeg_t64 & y){
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(y.tab[0]%2))
	COUT << "erreur" << '\n';
#endif
      tdeg_t64 res;
      res.order_=x.order_;
      res.ui=(longlong *)malloc((1+(x.order_.dim+degratiom1)/degratio)*sizeof(longlong));
      res.ui[0]=1;
      const longlong * xptr=x.ui+1,*xend=xptr+(x.order_.dim+degratiom1)/degratio,*yptr=y.ui+1;
      longlong * resptr=res.ui+1;
      for (;xptr!=xend;++resptr,++yptr,++xptr)
	*resptr=*xptr-*yptr;
      res.tdeg=1;
      res.compute_degs();
      return res;
    }
#endif
    tdeg_t64 res;
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]-ytab[0];
    ztab[1]=xtab[1]-ytab[1];
    ztab[2]=xtab[2]-ytab[2];
#if GROEBNER_VARS>11
    ztab[3]=xtab[3]-ytab[3];
#endif
#else
    for (unsigned i=0;i<=GROEBNER_VARS;++i)
      res.tab[i]=x.tab[i]-y.tab[i];
#endif
    return res;
  }
  inline bool operator == (const tdeg_t64 & x,const tdeg_t64 & y){
    longlong X=((longlong *) x.tab)[0];
    if (X!= ((longlong *) y.tab)[0])
      return false;
#ifdef GIAC_HASH
    if (x.hash!=y.hash)
      return false;
#endif
#ifdef GIAC_ELIM
    if (x.elim!=y.elim) return false;
#endif
#ifdef GIAC_64VARS
    if (
#ifdef BIGENDIAN
	x.tab[0]%2
#else
	X%2
#endif
	){
      //if (x.ui==y.ui) return true;
      const longlong * xptr=x.ui+1,*xend=xptr+(x.order_.dim+degratiom1)/degratio,*yptr=y.ui+1;
#ifndef GIAC_CHARTABDEG
      // dimension+3 is at least 16 otherwise the alternative code would be called
      if (*xptr!=*yptr)
	return false;
      ++yptr,++xptr;
      if (*xptr!=*yptr)
	return false;
      ++yptr,++xptr;
      if (*xptr!=*yptr)
	return false;
      ++yptr,++xptr;
      if (*xptr!=*yptr)
	return false;
      ++yptr,++xptr;
#endif
      for (;xptr!=xend;++yptr,++xptr){
	if (*xptr!=*yptr)
	  return false;
      }
      return true;
    }
#endif
    return ((longlong *) x.tab)[1] == ((longlong *) y.tab)[1] &&
      ((longlong *) x.tab)[2] == ((longlong *) y.tab)[2]
#if GROEBNER_VARS>11
      &&  ((longlong *) x.tab)[3] == ((longlong *) y.tab)[3]
#endif
    ;
  }
  inline bool operator != (const tdeg_t64 & x,const tdeg_t64 & y){
    return !(x==y);
  }

  static inline int tdeg_t64_revlex_greater (const tdeg_t64 & x,const tdeg_t64 & y){
#ifdef GBASIS_SWAP
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    if (xtab[0]!=ytab[0]) // tdeg test already donne by caller
      return xtab[0]<=ytab[0]?1:0;
    if (xtab[1]!=ytab[1])
      return xtab[1]<=ytab[1]?1:0;
    if (xtab[2]!=ytab[2])
      return xtab[2]<=ytab[2]?1:0;
#if GROEBNER_VARS>11
    return xtab[3]<=ytab[3]?1:0;
#endif
    return 2;
#else // GBASIS_SWAP
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>=y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]<=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
    }
#if GROEBNER_VARS>11
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]<=y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
    }
#endif
    return 2;
#endif // GBASIS_SWAP
  }

  // inline bool operator <  (const tdeg_t64 & x,const tdeg_t64 & y){ return !tdeg_t64_revlex_greater(x,y); }
  // inline bool operator >  (const tdeg_t64 & x,const tdeg_t64 & y){ return !tdeg_t64_revlex_greater(y,x); }
  // inline bool operator <= (const tdeg_t64 & x,const tdeg_t64 & y){ return tdeg_t64_revlex_greater(y,x); }
  // inline bool operator >=  (const tdeg_t64 & x,const tdeg_t64 & y){ return tdeg_t64_revlex_greater(x,y); }

#if GROEBNER_VARS==15

  int tdeg_t64_3var_greater (const tdeg_t64 & x,const tdeg_t64 & y){
    if (x.tab[0]!=y.tab[0])
      return x.tab[0]>=y.tab[0]?1:0;
    if (x.tab[4]!=y.tab[4])
      return x.tab[4]>=y.tab[4]?1:0;
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]<=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
    }
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]<=y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
    }
    return 2;
  }

  int tdeg_t64_7var_greater (const tdeg_t64 & x,const tdeg_t64 & y){
    if (x.tab[0]!=y.tab[0])
      return x.tab[0]>=y.tab[0]?1:0;
    if (x.tab[8]!=y.tab[8])
      return x.tab[8]>=y.tab[8]?1:0;
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
    }
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]<=y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
    }
    return 2;
  }

  int tdeg_t64_11var_greater (const tdeg_t64 & x,const tdeg_t64 & y){
    if (x.tab[0]!=y.tab[0])
      return x.tab[0]>=y.tab[0]?1:0;
    if (x.tab[12]!=y.tab[12])
      return x.tab[12]>=y.tab[12]?1:0;
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]<=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
    }
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
    }
    return 2;
  }
#endif // GROEBNER_VARS==15

  int tdeg_t64_lex_greater (const tdeg_t64 & x,const tdeg_t64 & y){
#ifdef GBASIS_SWAP
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    ulonglong X=*xtab, Y=*ytab;
    if (X!=Y){
      if ( (X & 0xffff) != (Y &0xffff))
	return (X&0xffff)>=(Y&0xffff)?1:0;
      return X>=Y?1:0;
    }
    if (xtab[1]!=ytab[1])
      return xtab[1]>=ytab[1]?1:0;
    if (xtab[2]!=ytab[2])
      return xtab[2]>=ytab[2]?1:0;
#if GROEBNER_VARS>11
    return xtab[3]>=ytab[3]?1:0;
#endif
    return 2;
#else
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]>y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]>y.tab[2]?1:0;
      return x.tab[3]>y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]>y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]>y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]>y.tab[6]?1:0;
      return x.tab[7]>y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]>y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]>y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]>y.tab[10]?1:0;
      return x.tab[11]>=y.tab[11]?1:0;
    }
#if GROEBNER_VARS>11
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]>y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]>y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]>y.tab[14]?1:0;
      return x.tab[15]>=y.tab[15]?1:0;
    }
#endif
    return 2;
#endif
  }

  inline int tdeg_t_greater_dyn(const tdeg_t64 & x,const tdeg_t64 & y,order_t order){
    const longlong * it1=x.ui,* it2=y.ui,*it1beg;
    longlong a=0;
    switch (order.o){
    case _REVLEX_ORDER:
      it1beg=x.ui;
      it1=x.ui+(x.order_.dim+degratiom1)/degratio;
      it2=y.ui+(y.order_.dim+degratiom1)/degratio;
#if 0 // def GIAC_ELIM
      --it1; --it2; // first test already done with elim field
#endif
      for (;it1!=it1beg;--it2,--it1){
	a=*it1-*it2;
	if (a)
	  return a<=0?1:0;
      }
      return 2;
    case _64VAR_ORDER:
      a=it1[16]-it2[16];
      if (a)
	return a<=0?1:0;
      a=it1[15]-it2[15];
      if (a)
	return a<=0?1:0;
      a=it1[14]-it2[14];
      if (a)
	return a<=0?1:0;
      a=it1[13]-it2[13];
      if (a)
	return a<=0?1:0;
    case _48VAR_ORDER:
      a=it1[12]-it2[12];
      if (a)
	return a<=0?1:0;
      a=it1[11]-it2[11];
      if (a)
	return a<=0?1:0;
      a=it1[10]-it2[10];
      if (a)
	return a<=0?1:0;
      a=it1[9]-it2[9];
      if (a)
	return a<=0?1:0;
    case _32VAR_ORDER:
      a=it1[8]-it2[8];
      if (a)
	return a<=0?1:0;
      a=it1[7]-it2[7];
      if (a)
	return a<=0?1:0;
      a=it1[6]-it2[6];
      if (a)
	return a<=0?1:0;
      a=it1[5]-it2[5];
      if (a)
	return a<=0?1:0;
    case _16VAR_ORDER:
      a=it1[4]-it2[4];
      if (a)
	return a<=0?1:0;
    case _11VAR_ORDER:
      a=it1[3]-it2[3];
      if (a)
	return a<=0?1:0;
    case _7VAR_ORDER:
      a=it1[2]-it2[2];
      if (a)
	return a<=0?1:0;
    case _3VAR_ORDER:
      a=it1[1]-it2[1];
      if (a)
	return a<=0?1:0;
      if (x.tdeg2!=y.tdeg2)
	return x.tdeg2>y.tdeg2?1:0;
      it1beg=it1+(x.order_.o+degratiom1)/degratio;
      it1 += (x.order_.dim+degratiom1)/degratio;;
      it2 += (x.order_.dim+degratiom1)/degratio;;
      for (;;){
	a=*it1-*it2;
	if (a)
	  return a<=0?1:0;
	--it2;--it1;
	if (it1<=it1beg)
	  return 2;
      }
    case _TDEG_ORDER: case _PLEX_ORDER: {
      const degtype * it1=(degtype *)(x.ui+1),*it1end=it1+x.order_.dim,*it2=(degtype *)(y.ui+1);
      for (;it1!=it1end;++it2,++it1){
	if (*it1!=*it2)
	  return *it1>=*it2?1:0;
      }
      return 2;
    }
    } // end swicth
    return -1;
  }

  inline int tdeg_t_greater(const tdeg_t64 & x,const tdeg_t64 & y,order_t order){
    short X=x.tab[0];
    if (X!=y.tab[0]) return X>y.tab[0]?1:0; // since tdeg is tab[0] for plex
#ifdef GIAC_64VARS
    if (X%2){
      if (x.tdeg2!=y.tdeg2) return x.tdeg2>y.tdeg2?1:0;
#ifdef GIAC_ELIM
      if ( x.elim!=y.elim) return x.elim<y.elim?1:0;
#endif
      //if (x.ui==y.ui) return 2;
#if 1 && !defined BIGENDIAN && !defined GIAC_CHARDEGTYPE
      return tdeg_t_greater_dyn(x,y,order);
#else // BIGENDIAN, GIAC_CHARDEGTYPE
      if (order.o>=_7VAR_ORDER || order.o==_3VAR_ORDER){
	int n=(order.o+degratiom1)/degratio;
	const longlong * it1beg=x.ui,*it1=x.ui+n,*it2=y.ui+n;
	longlong a=0,b=0;
#ifdef BIGENDIAN
	for (;it1!=it1beg;--it2,--it1){
	  a=*it1;
	  b=*it2;
	  if (a!=b)
	    break;
	}
	if (a!=b){
	  if ( ((a)&0xffff) != ((b)&0xffff) )
	    return ((a)&0xffff) <= ((b)&0xffff)?1:0;
	  if ( ((a>>16)&0xffff) != ((b>>16)&0xffff) )
	    return ((a>>16)&0xffff) <= ((b>>16)&0xffff)?1:0;
	  if ( ((a>>32)&0xffff) != ((b>>32)&0xffff) )
	    return ((a>>32)&0xffff) <= ((b>>32)&0xffff)?1:0;
	  return a <= b?1:0;
	}
#else
	for (;it1!=it1beg;--it2,--it1){
	  a=*it1-*it2;
	  if (a)
	    return a<=0?1:0;
	}
#endif
	// if (x.tdeg2!=y.tdeg2) return x.tdeg2>=y.tdeg2;
	it1beg=x.ui+n;
	n=(x.order_.dim+degratiom1)/degratio;
	it1=x.ui+n;
	it2=y.ui+n;
#ifdef BIGENDIAN
	for (a=0,b=0;it1!=it1beg;--it2,--it1){
	  a=*it1; b=*it2;
	  if (a!=b)
	    break;
	}
	if (a!=b){
	  if ( ((a)&0xffff) != ((b)&0xffff) )
	    return ((a)&0xffff) <= ((b)&0xffff)?1:0;
	  if ( ((a>>16)&0xffff) != ((b>>16)&0xffff) )
	    return ((a>>16)&0xffff) <= ((b>>16)&0xffff) ?1:0;
	  if ( ((a>>32)&0xffff) != ((b>>32)&0xffff) )
	    return ((a>>32)&0xffff) <= ((b>>32)&0xffff) ?1:0;
	  return a <= b?1:0;
	}
#else
	for (;it1!=it1beg;--it2,--it1){
	  a=*it1-*it2;
	  if (a)
	    return a<=0?1:0;
	}
#endif
	return 2;
      }
      if (order.o==_REVLEX_ORDER){
	//if (x.tdeg!=y.tdeg) return x.tdeg>y.tdeg?1:0;
	const longlong * it1beg=x.ui,*it1=x.ui+(x.order_.dim+degratiom1)/degratio,*it2=y.ui+(y.order_.dim+degratiom1)/degratio;
	longlong a=0,b=0;
#ifdef BIGENDIAN
	for (;it1!=it1beg;--it2,--it1){
	  a=*it1; b=*it2;
	  if (a!=b)
	    break;
	}
	if (a!=b){
	  if ( ((a)&0xffff) != ((b)&0xffff) )
	    return ((a)&0xffff) <= ((b)&0xffff)?1:0;
	  if ( ((a>>16)&0xffff) != ((b>>16)&0xffff) )
	    return ((a>>16)&0xffff) <= ((b>>16)&0xffff)?1:0;
	  if ( ((a>>32)&0xffff) != ((b>>32)&0xffff) )
	    return ((a>>32)&0xffff) <= ((b>>32)&0xffff)?1:0;
	  return a <= b?1:0;
	}
#else
	for (;it1!=it1beg;--it2,--it1){
	  a=*it1-*it2;
	  if (a)
	    return a<=0?1:0;
	}
#endif
	return 2;
      }
      // plex and tdeg share the same code since total degree already checked
      const degtype * it1=(degtype *)(x.ui+1),*it1end=it1+x.order_.dim,*it2=(degtype *)(y.ui+1);
      for (;it1!=it1end;++it2,++it1){
	if (*it1!=*it2)
	  return *it1>=*it2?1:0;
      }
      return 2;
#endif // BIGENDIAN, GIAC_CHARDEGTYPE
    } // end if X%2
#endif // GIAC_64VARS
    if (order.o==_REVLEX_ORDER)
      return tdeg_t64_revlex_greater(x,y);
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER)
      return tdeg_t64_3var_greater(x,y);
    if (order.o==_7VAR_ORDER)
      return tdeg_t64_7var_greater(x,y);
    if (order.o==_11VAR_ORDER)
      return tdeg_t64_11var_greater(x,y);
#endif
    return tdeg_t64_lex_greater(x,y);
  }
  inline bool tdeg_t_strictly_greater (const tdeg_t64 & x,const tdeg_t64 & y,order_t order){
    return !tdeg_t_greater(y,x,order); // total order
  }

  inline bool tdeg_t_all_greater(const tdeg_t64 & x,const tdeg_t64 & y,order_t order){
    if ((*((ulonglong*)&x)-*((ulonglong*)&y)) & 0x8000800080008000ULL)
      return false;
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
#ifdef GIAC_ELIM
      //bool debug=false;
      if ( !( (x.elim | y.elim) & 0x1000000000000000ULL) &&
#if 0 // ndef BESTA_OS // Keil compiler, for some reason does not like the 0b syntax!
	   ( (x.elim-y.elim) & 0b1111100001000010000100001000010000100001000010000100001000010000ULL) )
#else
	( (x.elim-y.elim) & 0xf842108421084210ULL) )
#endif
      {
	//debug=true;
	return false;
      }
#endif
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(y.tab[0]%2))
	COUT << "erreur" << '\n';
#endif
#ifdef GIAC_HASH
      if (x.hash<y.hash)
	return false;
#endif
#if 0
      const degtype * it1=(degtype *)(x.ui+1),*it1end=it1+x.order_.dim,*it2=(degtype *)(y.ui+1);
      for (;it1!=it1end;++it2,++it1){
	if (*it1<*it2)
	  return false;
      }
#else
      const longlong * it1=x.ui+1,*it1end=it1+(x.order_.dim+degratiom1)/degratio,*it2=y.ui+1;
#ifndef GIAC_CHARDEGTYPE
      if ((*it1-*it2) & 0x8000800080008000ULL)
	return false;
      ++it2; ++it1;
      if ((*it1-*it2) & 0x8000800080008000ULL)
	return false;
      ++it2; ++it1;
      if ((*it1-*it2) & 0x8000800080008000ULL)
	return false;
      ++it2; ++it1;
      if ((*it1-*it2) & 0x8000800080008000ULL)
	return false;
      ++it2; ++it1;
#endif
      for (;it1!=it1end;++it2,++it1){
#ifdef GIAC_CHARDEGTYPE
	if ((*it1-*it2) & 0x8080808080808080ULL)
	  return false;
#else
	if ((*it1-*it2) & 0x8000800080008000ULL)
	  return false;
#endif
      }
#endif
      // if (debug) CERR << x << " " << y << '\n' << x.elim << " " << y.elim << " " << x.tdeg << " " << y.tdeg << '\n';
      return true;
    }
#endif
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
  // if ((xtab[0]-ytab[0]) & 0x8000800080008000ULL) return false;
    if ((xtab[1]-ytab[1]) & 0x8000800080008000ULL)
      return false;
    if ((xtab[2]-ytab[2]) & 0x8000800080008000ULL)
      return false;
#if GROEBNER_VARS>11
    if ((xtab[3]-ytab[3]) & 0x8000800080008000ULL)
      return false;
#endif
    return true;
  }

  const longlong mask2=0x8080808080808080ULL;
#ifdef GIAC_CHARDEGTYPE
    const longlong mask=0x8080808080808080ULL;
#else
    const longlong mask=0x8000800080008000ULL;
#endif

  // 1 (all greater), 0 (unknown), -1 (all smaller)
  int tdeg_t_compare_all(const tdeg_t64 & x,const tdeg_t64 & y,order_t order){
#ifdef GIAC_64VARS
    if (x.tab[0]%2){
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(y.tab[0]%2))
	COUT << "erreur" << '\n';
#endif
      if ( (x.tdeg<y.tdeg) ^ (x.tdeg2<y.tdeg2))
	return 0;
      const longlong * it1=x.ui+1,*it1end=it1+(x.order_.dim+degratiom1)/degratio,*it2=y.ui+1;
      int res=0;
      for (;it1!=it1end;++it2,++it1){
	longlong tmp=*it1-*it2;
	if (tmp & mask){
	  if (res==1 || ((-tmp) & mask)) return 0;
	  res=-1;
	}
	else {
	  if (res==-1) return 0; else res=1;
	}
      }
      return res;
    }
#endif // GIAC_64VARS
    int res=0;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    longlong tmp=xtab[0]-ytab[0];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    tmp=xtab[1]-ytab[1];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    tmp=xtab[2]-ytab[2];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
#if GROEBNER_VARS>11
    tmp=xtab[3]-ytab[3];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
#endif
    return res;
  }

  void index_lcm(const tdeg_t64 & x,const tdeg_t64 & y,tdeg_t64 & z,order_t order){
#ifdef GIAC_64VARS
    if (x.tdeg%2){
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(y.tab[0]%2))
	COUT << "erreur" << '\n';
#endif
      z=tdeg_t64();
      z.tdeg=1;
      z.order_=x.order_;
      z.ui=(longlong *)malloc((1+(x.order_.dim+degratiom1)/degratio)*sizeof(longlong));
      z.ui[0]=1;
      const degtype * xptr=(degtype *)(x.ui+1),*xend=xptr+degratio*((x.order_.dim+degratiom1)/degratio),*yptr=(degtype *)(y.ui+1);
      degtype * resptr=(degtype *)(z.ui+1);
      for (;xptr!=xend;++resptr,++yptr,++xptr)
	*resptr=*xptr>*yptr?*xptr:*yptr;
      z.tdeg=1;
      z.compute_degs();
      return ;
    }
#endif
    int t=0;
    const short * xtab=&x.tab[1],*ytab=&y.tab[1];
    short *ztab=&z.tab[1];
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 1
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 2
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 3
    ++xtab; ++ytab; ++ztab;
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER){
#ifdef GIAC_64VARS
      z.tab[0]=2*t;
#else
      z.tab[0]=t;
#endif
      t=0;
      ++xtab;++ytab;++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 5
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 6
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 7
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 8
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 12
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
      z.tab[4]=t; // 4
      return;
    }
#endif
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 4
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 5
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 6
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 7
    ++xtab; ++ytab; ++ztab;
#if GROEBNER_VARS==15
    if (order.o==_7VAR_ORDER){
#ifdef GIAC_64VARS
      z.tab[0]=2*t;
#else
      z.tab[0]=t;
#endif
      t=0;
      ++xtab;++ytab;++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 12
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
      z.tab[8]=t; // 8
      return;
    }
#endif
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 8
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
#if GROEBNER_VARS>11
    ++xtab; ++ytab; ++ztab;
#if GROEBNER_VARS==15
    if (order.o==_11VAR_ORDER){
#ifdef GIAC_64VARS
      z.tab[0]=2*t;
#else
      z.tab[0]=t;
#endif
      t=0;
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
      z.tab[12]=t; // 12
      return;
    }
#endif
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 12
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
#endif
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
#ifdef GIAC_64VARS
      z.tab[0]=2*t;
#else
      z.tab[0]=t;
#endif
    }
    else {
#ifdef GIAC_64VARS
      z.tab[0]=2*((x.tab[0]>y.tab[0])?x.tab[0]:y.tab[0]);
#else
      z.tab[0]=(x.tab[0]>y.tab[0])?x.tab[0]:y.tab[0];
#endif
    }
  }

  void index_lcm_overwrite(const tdeg_t64 & x,const tdeg_t64 & y,tdeg_t64 & z,order_t order){
    if (z.tdeg%2==0){
      index_lcm(x,y,z,order);
      return;
    }
    const degtype * xptr=(degtype *)(x.ui+1),*xend=xptr+degratio*((x.order_.dim+degratiom1)/degratio),*yptr=(degtype *)(y.ui+1);
    degtype * resptr=(degtype *)(z.ui+1);
    for (;xptr!=xend;++resptr,++yptr,++xptr)
      *resptr=*xptr>*yptr?*xptr:*yptr;
    z.tdeg=1;
    z.compute_degs();
  }

  void get_index(const tdeg_t64 & x_,index_t & idx,order_t order,int dim){
#ifdef GIAC_64VARS
    if (x_.tab[0]%2){
      idx.resize(dim);
      const degtype * ptr=(degtype *)(x_.ui+1),*ptrend=ptr+x_.order_.dim;
      index_t::iterator target=idx.begin();
      for (;ptr!=ptrend;++target,++ptr)
	*target=*ptr;
      return;
    }
#endif
    idx.resize(dim);
#ifdef GBASIS_SWAP
    tdeg_t64 x(x_);
    swap_indices(x.tab);
#else
    const tdeg_t64 & x= x_;
#endif
    const short * ptr=x.tab;
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER){
      ++ptr;
      for (int i=1;i<=3;++ptr,++i)
	idx[3-i]=*ptr;
      ++ptr;
      for (int i=1;i<=dim-3;++ptr,++i)
	idx[dim-i]=*ptr;
      return;
    }
    if (order.o==_7VAR_ORDER){
      ++ptr;
      for (int i=1;i<=7;++ptr,++i)
	idx[7-i]=*ptr;
      ++ptr;
      for (int i=1;i<=dim-7;++ptr,++i)
	idx[dim-i]=*ptr;
      return;
    }
    if (order.o==_11VAR_ORDER){
      ++ptr;
      for (int i=1;i<=11;++ptr,++i)
	idx[11-i]=*ptr;
      ++ptr;
      for (int i=1;i<=dim-11;++ptr,++i)
	idx[dim-i]=*ptr;
      return;
    }
#endif
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER)
      ++ptr;
    if (order.o==_REVLEX_ORDER){
      for (int i=1;i<=dim;++ptr,++i)
	idx[dim-i]=*ptr;
    }
    else {
      for (int i=0;i<dim;++ptr,++i)
	idx[i]=*ptr;
#ifdef GIAC_64VARS
      idx[0]/=2;
#endif
    }
  }

  bool disjoint(const tdeg_t64 & a,const tdeg_t64 & b,order_t order,short dim){
#ifdef GIAC_64VARS
    if (a.tab[0]%2){
#ifdef GIAC_DEBUG_TDEG_T64
      if (!(b.tab[0]%2))
	COUT << "erreur" << '\n';
#endif
      const degtype * xptr=(degtype *)(a.ui+1),*xend=xptr+dim,*yptr=(degtype *)(b.ui+1);
      for (;xptr!=xend;++yptr,++xptr){
	if (*xptr && *yptr)
	  return false;
      }
      return true;
    }
#endif
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER){
      if ( (a.tab[1] && b.tab[1]) ||
	   (a.tab[2] && b.tab[2]) ||
	   (a.tab[3] && b.tab[3]) ||
	   (a.tab[5] && b.tab[5]) ||
	   (a.tab[6] && b.tab[6]) ||
	   (a.tab[7] && b.tab[7]) ||
	   (a.tab[8] && b.tab[8]) ||
	   (a.tab[9] && b.tab[9]) ||
	   (a.tab[10] && b.tab[10]) ||
	   (a.tab[11] && b.tab[11]) ||
	   (a.tab[12] && b.tab[12]) ||
	   (a.tab[13] && b.tab[13]) ||
	   (a.tab[14] && b.tab[14]) ||
	   (a.tab[15] && b.tab[15]) )
	return false;
      return true;
    }
    if (order.o==_7VAR_ORDER){
      if ( (a.tab[1] && b.tab[1]) ||
	   (a.tab[2] && b.tab[2]) ||
	   (a.tab[3] && b.tab[3]) ||
	   (a.tab[4] && b.tab[4]) ||
	   (a.tab[5] && b.tab[5]) ||
	   (a.tab[6] && b.tab[6]) ||
	   (a.tab[7] && b.tab[7]) ||
	   (a.tab[9] && b.tab[9]) ||
	   (a.tab[10] && b.tab[10]) ||
	   (a.tab[11] && b.tab[11]) ||
	   (a.tab[12] && b.tab[12]) ||
	   (a.tab[13] && b.tab[13]) ||
	   (a.tab[14] && b.tab[14]) ||
	   (a.tab[15] && b.tab[15]) )
	return false;
      return true;
    }
    if (order.o==_11VAR_ORDER){
      if ( (a.tab[1] && b.tab[1]) ||
	   (a.tab[2] && b.tab[2]) ||
	   (a.tab[3] && b.tab[3]) ||
	   (a.tab[4] && b.tab[4]) ||
	   (a.tab[5] && b.tab[5]) ||
	   (a.tab[6] && b.tab[6]) ||
	   (a.tab[7] && b.tab[7]) ||
	   (a.tab[8] && b.tab[8]) ||
	   (a.tab[9] && b.tab[9]) ||
	   (a.tab[10] && b.tab[10]) ||
	   (a.tab[11] && b.tab[11]) ||
	   (a.tab[13] && b.tab[13]) ||
	   (a.tab[14] && b.tab[14]) ||
	   (a.tab[15] && b.tab[15]))
	return false;
      return true;
    }
#endif
    const short * it=a.tab, * jt=b.tab;
#ifdef GBASIS_SWAP
    const short * itend=it+GROEBNER_VARS+1;
#endif
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
      ++it; ++jt;
    }
#ifndef GBASIS_SWAP
    const short * itend=it+dim;
#endif
    for (;it<itend;++jt,++it){
      if (*it && *jt)
	return false;
    }
    return true;
  }

  // polynomial are vector< T_unsigned<gen,tdeg_t> >
  template<class tdeg_t>
  struct poly8 {
    std::vector< T_unsigned<gen,tdeg_t> > coord;
    // lex order is implemented using tdeg_t as a list of degrees
    // tdeg uses total degree 1st then partial degree in lex order, max 7 vars
    // revlex uses total degree 1st then opposite of partial degree in reverse ordre, max 7 vars
    order_t order; // _PLEX_ORDER, _REVLEX_ORDER or _TDEG_ORDER or _7VAR_ORDER or _11VAR_ORDER
    short int dim;
    unsigned sugar;
    double logz; // unused, it's here for tripolymod_tri
    int age; // unused, it's here for tripolymod_tri
    void dbgprint() const;
    poly8():dim(0),sugar(0),logz(0),age(-1) {order.o=_PLEX_ORDER; order.lex=0; order.dim=0;}
    poly8(order_t o_,int dim_): order(o_),dim(dim_),sugar(0),logz(0),age(-1) {order.dim=dim_;}
    poly8(const polynome & p,order_t o_){
      order=o_;
      dim=p.dim;
      order.dim=p.dim;
      logz=0;
      age=-1;
      if (order.o%4!=3){
	if (p.is_strictly_greater==i_lex_is_strictly_greater)
	  order.o=_PLEX_ORDER;
	if (p.is_strictly_greater==i_total_revlex_is_strictly_greater)
	  order.o=_REVLEX_ORDER;
	if (p.is_strictly_greater==i_total_lex_is_strictly_greater)
	  order.o=_TDEG_ORDER;
      }
      if (
#ifdef GIAC_64VARS
	  0 &&
#endif
	  p.dim>GROEBNER_VARS)
	CERR << "Number of variables is too large to be handled by giac";
      else {
	coord.reserve(p.coord.size());
	for (unsigned i=0;i<p.coord.size();++i){
	  coord.push_back(T_unsigned<gen,tdeg_t>(p.coord[i].value,tdeg_t(p.coord[i].index,order)));
	}
      }
      if (coord.empty())
	sugar=0;
      else
	sugar=coord.front().u.total_degree(order);
    }
    void get_polynome(polynome & p) const {
      p.dim=dim;
      switch (order.o){
      case _REVLEX_ORDER:
	p.is_strictly_greater=i_total_revlex_is_strictly_greater;
	break;
      case _3VAR_ORDER:
	p.is_strictly_greater=i_3var_is_strictly_greater;
	break;
      case _7VAR_ORDER:
	p.is_strictly_greater=i_7var_is_strictly_greater;
	break;
      case _11VAR_ORDER:
	p.is_strictly_greater=i_11var_is_strictly_greater;
	break;
      case _TDEG_ORDER:
	p.is_strictly_greater=i_total_lex_is_strictly_greater;
	break;
      default:
      case _PLEX_ORDER:
	p.is_strictly_greater=i_lex_is_strictly_greater;
	break;
      }
      p.coord.clear();
      p.coord.reserve(coord.size());
      index_t idx(dim);
      for (unsigned i=0;i<coord.size();++i){
	get_index(coord[i].u,idx,order,dim);
	p.coord.push_back(monomial<gen>(coord[i].g,idx));
      }
      // if (order==_3VAR_ORDER || order==_7VAR_ORDER || order==_11VAR_ORDER) p.tsort();
    }
  };
  template<class tdeg_t>
  bool operator == (const poly8<tdeg_t> & p,const poly8<tdeg_t> &q){
    if (p.coord.size()!=q.coord.size())
      return false;
    for (unsigned i=0;i<p.coord.size();++i){
      if (p.coord[i].u!=q.coord[i].u || p.coord[i].g!=q.coord[i].g)
	return false;
    }
    return true;
  }

#ifdef NSPIRE
  template<class tdeg_t,class T> nio::ios_base<T> & operator << (nio::ios_base<T> & os, const poly8<tdeg_t> & p)
#else
  template<class tdeg_t>
    ostream & operator << (ostream & os, const poly8<tdeg_t> & p)
#endif
  {
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    int t2;
    if (it==itend)
      return os << 0 ;
    for (;it!=itend;){
      os << it->g  ;
#ifndef GBASIS_NO_OUTPUT
      if (it->u.vars64()){
	if (it->u.tdeg%2){
	  degtype * i=(degtype *)(it->u.ui+1);
	  int s=it->u.order_.dim;
	  for (int j=0;j<s;++j){
	    t2=i[j];
	    if (t2)
	      os << "*x"<< j << "^" << t2  ;
	  }
	  ++it;
	  if (it==itend)
	    break;
	  os << " + ";
	  continue;
	}
      }
#endif
      short tab[GROEBNER_VARS+1];
      it->u.get_tab(tab,p.order);
      switch (p.order.o){
      case _PLEX_ORDER:
	for (int i=0;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2)
	    os << "*x"<< i << "^" << t2  ;
	}
	break;
      case _REVLEX_ORDER:
	for (int i=1;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
#if GROEBNER_VARS==15
      case _3VAR_ORDER:
	for (int i=1;i<=3;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 3-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=5;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 7+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
      case _7VAR_ORDER:
	for (int i=1;i<=7;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 7-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=9;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 11+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
      case _11VAR_ORDER:
	for (int i=1;i<=11;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 11-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=13;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 15+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
#endif
      case _TDEG_ORDER:
	for (int i=1;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  if (t2)
	    os << "*x"<< i-1 << "^" << t2  ;
	}
	break;
      }
      ++it;
      if (it==itend)
	break;
      os << " + ";
    }
    return os;
  }


  template<class tdeg_t>
  void poly8<tdeg_t>::dbgprint() const {
    CERR << *this << '\n';
  }

  template<class tdeg_t>
  class vectpoly8:public vector<poly8<tdeg_t> >{
  public:
    void dbgprint() const { CERR << *this << '\n'; }
  };

  template<class tdeg_t>
  void vectpoly_2_vectpoly8(const vectpoly & v,order_t order,vectpoly8<tdeg_t> & v8){
    v8.clear();
    v8.reserve(v.size()); if (debug_infolevel>1000){ v8.dbgprint(); v8[0].dbgprint();}
    for (unsigned i=0;i<v.size();++i){
      v8.push_back(poly8<tdeg_t>(v[i],order));
    }
  }

  template<class tdeg_t>
  void vectpoly8_2_vectpoly(const vectpoly8<tdeg_t> & v8,vectpoly & v){
    v.clear();
    v.reserve(v8.size());
    for (unsigned i=0;i<v8.size();++i){
      v.push_back(polynome(v8[i].dim));
      v8[i].get_polynome(v[i]);
    }
  }

  // Groebner basis code begins here

  template<class tdeg_t>
  gen inplace_ppz(poly8<tdeg_t> & p,bool divide=true,bool quick=false){
    typename vector< T_unsigned<gen,tdeg_t> >::iterator it=p.coord.begin(),itend=p.coord.end();
    if (it==itend)
      return 1;
    gen res=(itend-1)->g;
    for (;it!=itend;++it){
      if (it->g.type==_INT_){
	res=it->g;
	if (quick)
	  return 1;
	break;
      }
    }
    if (res.type==_ZINT)
      res=*res._ZINTptr; // realloc for inplace gcd
    for (it=p.coord.begin();it!=itend;++it){
      if (res.type==_ZINT && it->g.type==_ZINT){
	mpz_gcd(*res._ZINTptr,*res._ZINTptr,*it->g._ZINTptr);
      }
      else
	res=gcd(res,it->g);
      if (is_one(res))
	return 1;
    }
    if (!divide)
      return res;
#ifndef USE_GMP_REPLACEMENTS
    if (res.type==_INT_ && res.val>0){
      for (it=p.coord.begin();it!=itend;++it){
	if (it->g.type!=_ZINT || it->g.ref_count()>1)
	  it->g=it->g/res;
	else
	  mpz_divexact_ui(*it->g._ZINTptr,*it->g._ZINTptr,res.val);
      }
      return res;
    }
    if (res.type==_ZINT){
      for (it=p.coord.begin();it!=itend;++it){
	if (it->g.type!=_ZINT || it->g.ref_count()>1)
	  it->g=it->g/res;
	else
	  mpz_divexact(*it->g._ZINTptr,*it->g._ZINTptr,*res._ZINTptr);
      }
      return res;
    }
#endif
    for (it=p.coord.begin();it!=itend;++it){
      it->g=it->g/res;
    }
    return res;
  }

  template<class tdeg_t>
  void inplace_mult(const gen & g,vector< T_unsigned<gen,tdeg_t> > & v){
    typename std::vector< T_unsigned<gen,tdeg_t> >::iterator it1=v.begin(),it1end=v.end();
    for (;it1!=it1end;++it1){
#if 0
      it1->g=g*(it1->g);
#else
      type_operator_times(g,it1->g,it1->g);
#endif
    }
  }

#define GBASIS_HEAP
#ifdef GBASIS_HEAP
  // heap: remove bitfields if that's not enough
  template<class tdeg_t>
  struct heap_t {
    unsigned i:16; // index in pairs of quotients/divisors
    unsigned qi:24;
    unsigned gj:24; // monomial index for quotient and divisor
    tdeg_t u; // product
  };

  // inline bool operator > (const heap_t & a,const heap_t & b){ return a.u>b.u; }

  // inline bool operator < (const heap_t & a,const heap_t & b){ return b>a; }
  template<class tdeg_t>
  struct heap_t_compare {
    order_t order;
    const heap_t<tdeg_t> * ptr;
    inline bool operator () (unsigned a,unsigned b){
      return !tdeg_t_greater((ptr+a)->u,(ptr+b)->u,order);
      // return (ptr+a)->u<(ptr+b)->u;
    }
    heap_t_compare(const vector<heap_t<tdeg_t> > & v,order_t o):order(o),ptr(v.empty()?0:&v.front()){};
  };

  template<class tdeg_t>
  struct compare_heap_t {
    order_t order;
    inline bool operator () (const heap_t<tdeg_t> & a,const heap_t<tdeg_t> & b){
      return !tdeg_t_greater(a.u,b.u,order);
      // return (ptr+a)->u<(ptr+b)->u;
    }
    compare_heap_t(order_t o):order(o) {}
  };

  template<class tdeg_t>
  struct heap_t_ptr {
    heap_t<tdeg_t> * ptr;
  };

  template<class tdeg_t>
  void heap_reduce(const poly8<tdeg_t> & f,const vectpoly8<tdeg_t> & g,const vector<unsigned> & G,unsigned excluded,vectpoly8<tdeg_t> & q,poly8<tdeg_t> & rem,poly8<tdeg_t>& R,gen & s,environment * env){
    // divides f by g[G[0]] to g[G[G.size()-1]] except maybe g[G[excluded]]
    // first implementation: use quotxsient heap for all quotient/divisor
    // do not use heap chain
    // ref Monaghan Pearce if g.size()==1
    // R is a temporary polynomial, should be different from f
    if (&rem==&f){
      R.dim=f.dim; R.order=f.order;
      heap_reduce(f,g,G,excluded,q,R,R,s,env);
      swap(rem.coord,R.coord);
      if (debug_infolevel>1000)
	g.dbgprint(); // instantiate dbgprint()
      return;
    }
    rem.coord.clear();
    if (f.coord.empty())
      return ;
    if (q.size()<G.size())
      q.resize(G.size());
    unsigned guess=0;
    for (unsigned i=0;i<G.size();++i){
      q[i].dim=f.dim;
      q[i].order=f.order;
      q[i].coord.clear();
      guess += unsigned(g[G[i]].coord.size());
    }
    vector<heap_t<tdeg_t> > H;
    compare_heap_t<tdeg_t> key(f.order);
    H.reserve(guess);
    vecteur invlcg(G.size());
    if (env && env->moduloon){
      for (unsigned i=0;i<G.size();++i){
	invlcg[i]=invmod(g[G[i]].coord.front().g,env->modulo);
      }
    }
    s=1;
    gen c,numer,denom;
    longlong C;
    unsigned k=0,i; // k=position in f
    tdeg_t m;
    bool finish=false;
    bool small0=env && env->moduloon && env->modulo.type==_INT_ && env->modulo.val;
    int p=env?env->modulo.val:0;
    while (!H.empty() || k<f.coord.size()){
      // is highest remaining degree in f or heap?
      if (k<f.coord.size() && (H.empty() || tdeg_t_greater(f.coord[k].u,H.front().u,f.order)) ){
	// it's in f or both
	m=f.coord[k].u;
	if (small0)
	  C=smod(f.coord[k].g,p).val;
	else {
	  if (s==1)
	    c=f.coord[k].g;
	  else
	    c=s*f.coord[k].g;
	}
	++k;
      }
      else {
	m=H.front().u;
	c=0;
	C=0;
      }
      // extract from heap all terms having m as monomials, substract from c
      while (!H.empty() && H.front().u==m){
	std::pop_heap(H.begin(),H.end(),key);
	heap_t<tdeg_t> & current=H.back(); // was root node of the heap
	const poly8<tdeg_t> & gcurrent = g[G[current.i]];
	if (small0)
	  C -= longlong(q[current.i].coord[current.qi].g.val) * smod(gcurrent.coord[current.gj].g,p).val;
	else {
	  if (env && env->moduloon){
	    c -= q[current.i].coord[current.qi].g * gcurrent.coord[current.gj].g;
	  }
	  else {
	    fxnd(q[current.i].coord[current.qi].g,numer,denom);
	    if (denom==s)
	      c -= numer*gcurrent.coord[current.gj].g;
	    else {
	      if (denom==1)
		c -= s*numer*gcurrent.coord[current.gj].g;
	      else
		c -= (s/denom)*numer*gcurrent.coord[current.gj].g;
	    }
	  }
	}
	if (current.gj<gcurrent.coord.size()-1){
	  ++current.gj;
	  current.u=q[current.i].coord[current.qi].u+gcurrent.coord[current.gj].u;
	  push_heap(H.begin(),H.end(),key);
	}
	else
	  H.pop_back();
      }
      if (small0){
	C %= p;
	if (C==0)
	  continue;
	c=C;
      }
      else {
	if (env && env->moduloon)
	  c=smod(c,env->modulo);
	if (c==0)
	  continue;
      }
      // divide (c,m) by one of the g if possible, otherwise push in remainder
      if (finish)
	i=unsigned(G.size());
      else {
	finish=true;
	for (i=0;i<G.size();++i){
	  if (i==excluded || g[G[i]].coord.empty())
	    continue;
	  if (tdeg_t_greater(m,g[G[i]].coord.front().u,f.order)){
	    finish=false;
	    if (tdeg_t_all_greater(m,g[G[i]].coord.front().u,f.order))
	      break;
	  }
	}
      }
      if (i==G.size()){
	if (s==1)
	  rem.coord.push_back(T_unsigned<gen,tdeg_t>(c,m)); // add c/s*m to remainder
	else {
	  //rem.coord.push_back(T_unsigned<gen,tdeg_t>(c/s,m)); // add c/s*m to remainder
	  rem.coord.push_back(T_unsigned<gen,tdeg_t>(Tfraction<gen>(c,s),m)); // add c/s*m to remainder
	}
	continue;
      }
      // add c/s*m/leading monomial of g[G[i]] to q[i]
      tdeg_t monom=m-g[G[i]].coord.front().u;
      if (env && env->moduloon){
	if (invlcg[i]!=1){
	  if (invlcg[i]==-1)
	    c=-c;
	  else
	    c=smod(c*invlcg[i],env->modulo);
	}
	q[i].coord.push_back(T_unsigned<gen,tdeg_t>(c,monom));
      }
      else {
	gen lcg=g[G[i]].coord.front().g;
	gen pgcd=simplify3(lcg,c);
	if (is_positive(-lcg,context0)){
	  lcg=-lcg;
	  c=-c;
	}
	s=s*lcg;
	if (s==1)
	  q[i].coord.push_back(T_unsigned<gen,tdeg_t>(c,monom));
	else
	  q[i].coord.push_back(T_unsigned<gen,tdeg_t>(Tfraction<gen>(c,s),monom));
      }
      // push in heap
      if (g[G[i]].coord.size()>1){
	heap_t<tdeg_t> current={i,int(q[i].coord.size())-1,1,g[G[i]].coord[1].u+monom};
	H.push_back(current);
	push_heap(H.begin(),H.end(),key);
      }
    } // end main heap pseudo-division loop
  }

  template<class tdeg_t>
  void heap_reduce(const poly8<tdeg_t> & f,const vectpoly8<tdeg_t> & g,const vector<unsigned> & G,unsigned excluded,vectpoly8<tdeg_t> & q,poly8<tdeg_t> & rem,poly8<tdeg_t>& TMP1,environment * env){
    gen s;
    if (debug_infolevel>2)
      CERR << f << " = " << '\n';
    heap_reduce(f,g,G,excluded,q,rem,TMP1,s,env);
    // end up by multiplying rem by s (so that everything is integer)
    if (debug_infolevel>2){
      for (unsigned i=0;i<G.size();++i)
	CERR << "(" << g[G[i]]<< ")*(" << q[i] << ")+ ";
      CERR << rem << '\n';
    }
    if (env && env->moduloon){
      if (!rem.coord.empty() && rem.coord.front().g!=1)
	smallmult(invmod(rem.coord.front().g,env->modulo),rem.coord,rem.coord,env->modulo.val);
      return;
    }
    if (s!=1)
      smallmult(s,rem.coord,rem.coord);
    gen tmp=inplace_ppz(rem);
    if (debug_infolevel>1)
      CERR << "ppz was " << tmp << '\n';
  }

#endif // GBASIS_HEAP


  template<class tdeg_t>
  void smallmult(const gen & a,poly8<tdeg_t> & p,gen & m){
    typename std::vector< T_unsigned<gen,tdeg_t> >::iterator pt=p.coord.begin(),ptend=p.coord.end();
    if (a.type==_INT_ && m.type==_INT_){
      for (;pt!=ptend;++pt){
	if (pt->g.type==_INT_)
	  pt->g=(longlong(pt->g.val)*a.val)%m.val;
	else
	  pt->g=smod(a*pt->g,m);
      }
    }
    else {
      for (;pt!=ptend;++pt){
	pt->g=smod(a*pt->g,m);
      }
    }
  }

  // p - a*q shifted mod m -> r
  template<class tdeg_t>
  void smallmultsub(const poly8<tdeg_t> & p,unsigned pos,int a,const poly8<tdeg_t> & q,const tdeg_t & shift,poly8<tdeg_t> & r,int m){
    r.coord.clear();
    r.coord.reserve(p.coord.size()+q.coord.size());
    typename vector< T_unsigned<gen,tdeg_t> >::const_iterator it=p.coord.begin()+pos,itend=p.coord.end(),jt=q.coord.begin(),jtend=q.coord.end();
    for (;jt!=jtend;++jt){
      tdeg_t v=jt->u+shift;
      for (;it!=itend && tdeg_t_strictly_greater(it->u,v,p.order);++it){
	r.coord.push_back(*it);
      }
      if (it!=itend && it->u==v){
	if (it->g.type==_INT_ && jt->g.type==_INT_){
	  int tmp=(it->g.val-longlong(a)*jt->g.val)%m;
	  if (tmp)
	    r.coord.push_back(T_unsigned<gen,tdeg_t>(tmp,v));
	}
	else
	  r.coord.push_back(T_unsigned<gen,tdeg_t>(smod(it->g-a*jt->g,m),v));
	++it;
      }
      else {
	if (jt->g.type==_INT_){
	  int tmp=(-longlong(a)*jt->g.val)%m;
	  r.coord.push_back(T_unsigned<gen,tdeg_t>(tmp,v));
	}
	else
	  r.coord.push_back(T_unsigned<gen,tdeg_t>(smod(-a*jt->g,m),v));
      }
    }
    for (;it!=itend;++it){
      r.coord.push_back(*it);
    }
  }

  // a and b are assumed to be _ZINT
  template<class tdeg_t>
  void linear_combination(const gen & a,const poly8<tdeg_t> &p,tdeg_t * ashift,const gen &b,const poly8<tdeg_t> & q,tdeg_t * bshift,poly8<tdeg_t> & r,environment * env){
    r.coord.clear();
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=q.coord.begin(),jtend=q.coord.end();
    r.coord.reserve((itend-it)+(jtend-jt));
    mpz_t tmpz;
    mpz_init(tmpz);
    if (jt!=jtend){
      tdeg_t v=jt->u;
      if (bshift)
	v=v+*bshift;
      for (;it!=itend;){
	tdeg_t u=it->u;
	if (ashift)
	  u=u+*ashift;
	if (u==v){
	  gen g;
#ifndef USE_GMP_REPLACEMENTS
	  if ( (it->g.type==_INT_ || it->g.type==_ZINT) &&
	       (jt->g.type==_INT_ || jt->g.type==_ZINT) ){
	    if (it->g.type==_INT_)
	      mpz_mul_si(tmpz,*a._ZINTptr,it->g.val);
	    else
	      mpz_mul(tmpz,*a._ZINTptr,*it->g._ZINTptr);
	    if (jt->g.type==_INT_){
	      if (jt->g.val>=0)
		mpz_addmul_ui(tmpz,*b._ZINTptr,jt->g.val);
	      else
		mpz_submul_ui(tmpz,*b._ZINTptr,-jt->g.val);
	    }
	    else
	      mpz_addmul(tmpz,*b._ZINTptr,*jt->g._ZINTptr);
	    if (mpz_sizeinbase(tmpz,2)<31)
	      g=int(mpz_get_si(tmpz));
	    else {
	      ref_mpz_t * ptr =new ref_mpz_t;
	      mpz_swap(ptr->z,tmpz);
	      g=ptr; // g=tmpz;
	    }
	  }
	  else
#endif
	    g=a*it->g+b*jt->g;
	  if (env && env->moduloon)
	    g=smod(g,env->modulo);
	  if (!is_zero(g))
	    r.coord.push_back(T_unsigned<gen,tdeg_t>(g,u));
	  ++it; ++jt;
	  if (jt==jtend)
	    break;
	  v=jt->u;
	  if (bshift)
	    v=v+*bshift;
	  continue;
	}
	if (tdeg_t_strictly_greater(u,v,p.order)){
	  gen g=a*it->g;
	  if (env && env->moduloon)
	    g=smod(g,env->modulo);
	  r.coord.push_back(T_unsigned<gen,tdeg_t>(g,u));
	  ++it;
	}
	else {
	  gen g=b*jt->g;
	  if (env && env->moduloon)
	    g=smod(g,env->modulo);
	  r.coord.push_back(T_unsigned<gen,tdeg_t>(g,v));
	  ++jt;
	  if (jt==jtend)
	    break;
	  v=jt->u;
	  if (bshift)
	    v=v+*bshift;
	}
      }
    }
    for (;it!=itend;++it){
      tdeg_t u=it->u;
      if (ashift)
	u=u+*ashift;
      gen g=a*it->g;
      if (env && env->moduloon)
	g=smod(g,env->modulo);
      r.coord.push_back(T_unsigned<gen,tdeg_t>(g,u));
    }
    for (;jt!=jtend;++jt){
      tdeg_t v=jt->u;
      if (bshift)
	v=v+*bshift;
      gen g=b*jt->g;
      if (env && env->moduloon)
	g=smod(g,env->modulo);
      r.coord.push_back(T_unsigned<gen,tdeg_t>(g,v));
    }
    mpz_clear(tmpz);
  }

  // check that &v1!=&v and &v2!=&v
  template<class tdeg_t>
  void sub(const poly8<tdeg_t> & v1,const poly8<tdeg_t> & v2,poly8<tdeg_t> & v,environment * env){
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator it1=v1.coord.begin(),it1end=v1.coord.end(),it2=v2.coord.begin(),it2end=v2.coord.end();
    gen g;
    v.coord.clear();
    v.coord.reserve((it1end-it1)+(it2end-it2)); // worst case
    for (;it1!=it1end && it2!=it2end;){
      if (it1->u==it2->u){
	g=it1->g-it2->g;
	if (env && env->moduloon)
	  g=smod(g,env->modulo);
	if (!is_zero(g))
	  v.coord.push_back(T_unsigned<gen,tdeg_t>(g,it1->u));
	++it1;
	++it2;
      }
      else {
	if (tdeg_t_strictly_greater(it1->u,it2->u,v1.order)){
	  v.coord.push_back(*it1);
	  ++it1;
	}
	else {
	  v.coord.push_back(T_unsigned<gen,tdeg_t>(-it2->g,it2->u));
	  ++it2;
	}
      }
    }
    for (;it1!=it1end;++it1)
      v.coord.push_back(*it1);
    for (;it2!=it2end;++it2)
      v.coord.push_back(T_unsigned<gen,tdeg_t>(-it2->g,it2->u));
  }

  template<class tdeg_t>
  void reduce(const poly8<tdeg_t> & p,const vectpoly8<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,vectpoly8<tdeg_t> & quo,poly8<tdeg_t> & rem,poly8<tdeg_t> & TMP1, poly8<tdeg_t> & TMP2,gen & lambda,environment * env,vector<bool> * Gusedptr=0){
    lambda=1;
    // last chance of improving = modular method for reduce or modular algo
    if (&p!=&rem)
      rem=p;
    if (p.coord.empty())
      return ;
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator pt,ptend;
    unsigned i,rempos=0;
    bool small0=env && env->moduloon && env->modulo.type==_INT_ && env->modulo.val;
    TMP1.order=p.order; TMP1.dim=p.dim; TMP2.order=p.order; TMP2.dim=p.dim; TMP1.coord.clear();
    for (unsigned count=0;;++count){
      ptend=rem.coord.end();
      // this branch search first in all leading coeff of G for a monomial
      // <= to the current rem monomial
      pt=rem.coord.begin()+rempos;
      if (pt>=ptend)
	break;
      for (i=0;i<G.size();++i){
	if (i==excluded || res[G[i]].coord.empty())
	  continue;
	if (tdeg_t_all_greater(pt->u,res[G[i]].coord.front().u,p.order))
	  break;
      }
      if (i==G.size()){ // no leading coeff of G is smaller than the current coeff of rem
	++rempos;
	// if (small0) TMP1.coord.push_back(*pt);
	continue;
      }
      if (Gusedptr)
	(*Gusedptr)[i]=true;
      gen a(pt->g),b(res[G[i]].coord.front().g);
      if (small0){
	smallmultsub(rem,0,smod(a*invmod(b,env->modulo),env->modulo).val,res[G[i]],pt->u-res[G[i]].coord.front().u,TMP2,env->modulo.val);
	// smallmultsub(rem,rempos,smod(a*invmod(b,env->modulo),env->modulo).val,res[G[i]],pt->u-res[G[i]].coord.front().u,TMP2,env->modulo.val);
	// rempos=0; // since we have removed the beginning of rem (copied in TMP1)
	swap(rem.coord,TMP2.coord);
	continue;
      }
      TMP1.coord.clear();
      TMP2.coord.clear();
      tdeg_t resshift=pt->u-res[G[i]].coord.front().u;
      if (env && env->moduloon){
	gen ab=a;
	if (b!=1)
	  ab=a*invmod(b,env->modulo);
	ab=smod(ab,env->modulo);
	smallshift(res[G[i]].coord,resshift,TMP1.coord);
	if (ab!=1)
	  smallmult(ab,TMP1,env->modulo);
	sub(rem,TMP1,TMP2,env);
      }
      else {
	// -b*rem+a*shift(res[G[i]])
	simplify(a,b);
	if (b==-1){
	  b=-b;
	  a=-a;
	}
	gen c=-b;
	if (a.type==_ZINT && c.type==_ZINT && !is_one(a) && !is_one(b)){
	  linear_combination<tdeg_t>(c,rem,0,a,res[G[i]],&resshift,TMP2,0);
	  lambda=c*lambda;
	}
	else {
	  smallshift(res[G[i]].coord,resshift,TMP1.coord);
	  if (!is_one(a))
	    inplace_mult(a,TMP1.coord);
	  if (!is_one(b)){
	    inplace_mult(b,rem.coord);
	    lambda=b*lambda;
	  }
	  sub(rem,TMP1,TMP2,0);
	}
	//if (count % 6==5) inplace_ppz(TMP2,true,true); // quick gcd check
      }
      swap(rem.coord,TMP2.coord);
    }
    if (env && env->moduloon){
      // if (small0) swap(rem.coord,TMP1.coord);
      if (!rem.coord.empty() && rem.coord.front().g!=1)
	smallmult(invmod(rem.coord.front().g,env->modulo),rem.coord,rem.coord,env->modulo.val);
      return;
    }
    gen g=inplace_ppz(rem);
    lambda=lambda/g;
    if (debug_infolevel>2){
      if (rem.coord.empty())
	CERR << "0 reduction" << '\n';
      if (g.type==_ZINT && mpz_sizeinbase(*g._ZINTptr,2)>16)
	CERR << "ppz size was " << mpz_sizeinbase(*g._ZINTptr,2) << '\n';
    }
  }

  template<class tdeg_t>
  void reduce(const poly8<tdeg_t> & p,const vectpoly8<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,vectpoly8<tdeg_t> & quo,poly8<tdeg_t> & rem,poly8<tdeg_t> & TMP1, poly8<tdeg_t> & TMP2,environment * env,vector<bool> * Gusedptr=0){
    gen lambda;
    reduce(p,res,G,excluded,quo,rem,TMP1,TMP2,lambda,env,Gusedptr);
  }

  template<class tdeg_t>
  void reduce1small(poly8<tdeg_t> & p,const poly8<tdeg_t> & q,poly8<tdeg_t> & TMP1, poly8<tdeg_t> & TMP2,environment * env){
    if (p.coord.empty())
      return ;
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator pt,ptend;
    unsigned rempos=0;
    TMP1.coord.clear();
    const tdeg_t & u = q.coord.front().u;
    const gen g=q.coord.front().g;
    for (unsigned count=0;;++count){
      ptend=p.coord.end();
      // this branch search first in all leading coeff of G for a monomial
      // <= to the current rem monomial
      pt=p.coord.begin()+rempos;
      if (pt>=ptend)
	break;
      if (!tdeg_t_all_greater(pt->u,u,p.order)){
	++rempos;
	// TMP1.coord.push_back(*pt);
	continue;
      }
      smallmultsub(p,0,smod(pt->g*invmod(g,env->modulo),env->modulo).val,q,pt->u-u,TMP2,env->modulo.val);
      // smallmultsub(p,rempos,smod(a*invmod(b,env->modulo),env->modulo).val,q,pt->u-u,TMP2,env->modulo.val);
      // rempos=0; // since we have removed the beginning of rem (copied in TMP1)
      swap(p.coord,TMP2.coord);
    }
    // if (small0) swap(p.coord,TMP1.coord);
    if (env && env->moduloon && !p.coord.empty() && p.coord.front().g!=1)
      smallmult(invmod(p.coord.front().g,env->modulo),p.coord,p.coord,env->modulo.val);
  }

  // reduce with respect to itself the elements of res with index in G
  template<class tdeg_t>
  void reduce(vectpoly8<tdeg_t> & res,vector<unsigned> G,environment * env){
    if (res.empty() || G.empty())
      return;
    poly8<tdeg_t> pred(res.front().order,res.front().dim),
      TMP1(res.front().order,res.front().dim),
      TMP2(res.front().order,res.front().dim);
    vectpoly8<tdeg_t> q;
    // reduce res
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      poly8<tdeg_t> & p=res[i];
      reduce(p,res,G,i,q,pred,TMP1,TMP2,env);
      swap(res[i].coord,pred.coord);
      pred.sugar=res[i].sugar;
    }
  }

  template<class tdeg_t>
  void spoly(const poly8<tdeg_t> & p,const poly8<tdeg_t> & q,poly8<tdeg_t> & res,poly8<tdeg_t> & TMP1, environment * env){
    if (p.coord.empty()){
      res=q;
      return ;
    }
    if (q.coord.empty()){
      res= p;
      return;
    }
    const tdeg_t & pi = p.coord.front().u;
    const tdeg_t & qi = q.coord.front().u;
    tdeg_t lcm;
    index_lcm(pi,qi,lcm,p.order);
    tdeg_t pshift=lcm-pi;
    unsigned sugarshift=pshift.total_degree(p.order);
    // adjust sugar for res
    res.sugar=p.sugar+sugarshift;
    // CERR << "spoly " << res.sugar << " " << pi << qi << '\n';
    gen a=p.coord.front().g,b=q.coord.front().g;
    simplify3(a,b);
    if (debug_infolevel>2)
      CERR << "spoly " << a << " " << b << '\n';
    if (a.type==_ZINT && b.type==_ZINT){
      tdeg_t u=lcm-pi,v=lcm-qi;
      linear_combination(b,p,&u,a,q,&v,res,env);
    }
    else {
      poly8<tdeg_t> tmp1(p),tmp2(q);
      smallshift(tmp1.coord,lcm-pi,tmp1.coord);
      smallmult(b,tmp1.coord,tmp1.coord);
      smallshift(tmp2.coord,lcm-qi,tmp2.coord);
      smallmult(a,tmp2.coord,tmp2.coord);
      sub(tmp1,tmp2,res,env);
    }
    a=inplace_ppz(res);
    if (debug_infolevel>2)
      CERR << "spoly ppz " << a << '\n';
  }

  template<class tdeg_t>
  void gbasis_update(vector<unsigned> & G,vector< paire > & B,vectpoly8<tdeg_t> & res,unsigned pos,poly8<tdeg_t> & TMP1,poly8<tdeg_t> & TMP2,vectpoly8<tdeg_t> & vtmp,environment * env){
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin gbasis update " << '\n';
    const poly8<tdeg_t> & h = res[pos];
    order_t order=h.order;
    vector<unsigned> C;
    C.reserve(G.size());
    const tdeg_t & h0=h.coord.front().u;
    tdeg_t tmp1,tmp2;
    // C is used to construct new pairs
    // create pairs with h and elements g of G, then remove
    // -> if g leading monomial is prime with h, remove the pair
    // -> if g leading monomial is not disjoint from h leading monomial
    //    keep it only if lcm of leading monomial is not divisible by another one
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (res[G[i]].coord.empty() || disjoint(h0,res[G[i]].coord.front().u,res.front().order,res.front().dim))
	continue;
      index_lcm(h0,res[G[i]].coord.front().u,tmp1,order); // h0 and G[i] leading monomial not prime together
      unsigned j;
      for (j=0;j<G.size();++j){
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return;
	if (i==j || res[G[j]].coord.empty())
	  continue;
	index_lcm(h0,res[G[j]].coord.front().u,tmp2,order);
	if (tdeg_t_all_greater(tmp1,tmp2,order)){
	  // found another pair, keep the smallest, or the first if equal
	  if (tmp1!=tmp2)
	    break;
	  if (i>j)
	    break;
	}
      } // end for j
      if (j==G.size())
	C.push_back(G[i]);
    }
    vector< paire > B1;
    B1.reserve(B.size()+C.size());
    for (unsigned i=0;i<B.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (res[B[i].first].coord.empty() || res[B[i].second].coord.empty())
	continue;
      index_lcm(res[B[i].first].coord.front().u,res[B[i].second].coord.front().u,tmp1,order);
      if (!tdeg_t_all_greater(tmp1,h0,order)){
	B1.push_back(B[i]);
	continue;
      }
      index_lcm(res[B[i].first].coord.front().u,h0,tmp2,order);
      if (tmp2==tmp1){
	B1.push_back(B[i]);
	continue;
      }
      index_lcm(res[B[i].second].coord.front().u,h0,tmp2,order);
      if (tmp2==tmp1){
	B1.push_back(B[i]);
	continue;
      }
    }
    // B <- B union pairs(h,g) with g in C
    for (unsigned i=0;i<C.size();++i)
      B1.push_back(paire(pos,C[i]));
    swap(B1,B);
    // Update G by removing elements with leading monomial >= leading monomial of h
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin Groebner interreduce " << '\n';
    C.clear();
    C.reserve(G.size());
    vector<unsigned> hG(1,pos);
    bool small0=env && env->moduloon && env->modulo.type==_INT_ && env->modulo.val;
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (!res[G[i]].coord.empty() && !tdeg_t_all_greater(res[G[i]].coord.front().u,h0,order)){
	// reduce res[G[i]] with respect to h
	if (small0)
	  reduce1small(res[G[i]],h,TMP1,TMP2,env);
	else
	  reduce(res[G[i]],res,hG,-1,vtmp,res[G[i]],TMP1,TMP2,env);
	C.push_back(G[i]);
      }
      // NB: removing all pairs containing i in it does not work
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " end Groebner interreduce " << '\n';
    C.push_back(pos);
    swap(C,G);
  }

  template<class tdeg_t>
  bool in_gbasis(vectpoly8<tdeg_t> & res,vector<unsigned> & G,environment * env,bool sugar){
    poly8<tdeg_t> TMP1(res.front().order,res.front().dim),TMP2(res.front().order,res.front().dim);
    vectpoly8<tdeg_t> vtmp;
    vector< paire > B;
    order_t order=res.front().order;
    //if (order==_PLEX_ORDER)
      sugar=false; // otherwise cyclic6 fails (bus error), don't know why
    for (unsigned l=0;l<res.size();++l){
      gbasis_update(G,B,res,l,TMP1,TMP2,vtmp,env);
    }
    for (int age=1;!B.empty() && !interrupted && !ctrl_c;++age){
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " number of pairs: " << B.size() << ", base size: " << G.size() << '\n';
      // find smallest lcm pair in B
      tdeg_t small0,cur;
      unsigned smallpos,smallsugar=0,cursugar=0;
      for (smallpos=0;smallpos<B.size();++smallpos){
	if (!res[B[smallpos].first].coord.empty() && !res[B[smallpos].second].coord.empty())
	  break;
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return false;
      }
      index_lcm(res[B[smallpos].first].coord.front().u,res[B[smallpos].second].coord.front().u,small0,order);
      if (sugar)
	smallsugar=res[B[smallpos].first].sugar+(small0-res[B[smallpos].first].coord.front().u).total_degree(order);
      for (unsigned i=smallpos+1;i<B.size();++i){
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return false;
	if (res[B[i].first].coord.empty() || res[B[i].second].coord.empty())
	  continue;
	index_lcm(res[B[i].first].coord.front().u,res[B[i].second].coord.front().u,cur,order);
	if (sugar)
	  cursugar=res[B[smallpos].first].sugar+(cur-res[B[smallpos].first].coord.front().u).total_degree(order);
	bool doswap;
	if (order.o==_PLEX_ORDER)
	  doswap=tdeg_t_strictly_greater(small0,cur,order);
	else {
	  if (cursugar!=smallsugar)
	    doswap = smallsugar > cursugar;
	  else
	    doswap=tdeg_t_strictly_greater(small0,cur,order);
	}
	if (doswap){
	  // CERR << "swap " << cursugar << " " << res[B[i].first].coord.front().u << " " << res[B[i].second].coord.front().u << '\n';
	  swap(small0,cur); // small0=cur;
	  swap(smallsugar,cursugar);
	  smallpos=i;
	}
      }
      paire bk=B[smallpos];
      if (debug_infolevel>1 && (equalposcomp(G,bk.first)==0 || equalposcomp(G,bk.second)==0))
	CERR << CLOCK()*1e-6 << " reducing pair with 1 element not in basis " << bk << '\n';
      B.erase(B.begin()+smallpos);
      poly8<tdeg_t> h(res.front().order,res.front().dim);
      spoly(res[bk.first],res[bk.second],h,TMP1,env);
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " reduce begin, pair " << bk << " remainder size " << h.coord.size() << '\n';
      reduce(h,res,G,-1,vtmp,h,TMP1,TMP2,env);
      if (debug_infolevel>1){
	if (debug_infolevel>3){ CERR << h << '\n'; }
	CERR << CLOCK()*1e-6 << " reduce end, remainder size " << h.coord.size() << '\n';
      }
      if (!h.coord.empty()){
	res.push_back(h);
	gbasis_update(G,B,res,unsigned(res.size()-1),TMP1,TMP2,vtmp,env);
	if (debug_infolevel>2)
	  CERR << CLOCK()*1e-6 << " basis indexes " << G << " pairs indexes " << B << '\n';
      }
    }
    return true;
  }

  longlong invmod(longlong a,longlong b){
    if (a==1 || a==-1 || a==1-b)
      return a;
    longlong aa(1),ab(0),ar(0);
#ifdef VISUALC
    longlong q,r;
    while (b){
      q=a/b;
      r=a-q*b;
      ar=aa-q*ab;
      a=b;
      b=r;
      aa=ab;
      ab=ar;
    }
#else
    lldiv_t qr;
    while (b){
      qr=lldiv(a,b);
      ar=aa-qr.quot*ab;
      a=b;
      b=qr.rem;
      aa=ab;
      ab=ar;
    }
#endif
    if (a==1)
      return aa;
    if (a!=-1){
#ifndef NO_STDEXCEPT
      setsizeerr(gettext("Not invertible"));
#endif
      return 0;
    }
    return -aa;
  }

  longlong smod(longlong a,longlong b){
    longlong r=a%b;
    if (r>b/2)
      r -= b;
    else {
      if (r<=-b/2)
	r += b;
    }
    return r;
  }

#if 0 // def x86_64
  typedef longlong modint;
  typedef int128_t modint2;
  longlong smod(int128_t a,longlong b){
    longlong r=a%b;
    if (r>b/2)
      r -= b;
    else {
      if (r<=-b/2)
	r += b;
    }
  }

#else
  typedef int modint;
  typedef longlong modint2;
#endif

  template<class tdeg_t>
  struct polymod {
    std::vector< T_unsigned<modint,tdeg_t> > coord;
    // lex order is implemented using tdeg_t as a list of degrees
    // tdeg uses total degree 1st then partial degree in lex order, max 7 vars
    // revlex uses total degree 1st then opposite of partial degree in reverse ordre, max 7 vars
    order_t order; // _PLEX_ORDER, _REVLEX_ORDER or _TDEG_ORDER or _7VAR_ORDER or _11VAR_ORDER
    short int dim;
    unsigned sugar;
    int fromleft,fromright,age;
    double logz; // trace origin as a s-polynomial
    void dbgprint() const;
    void swap(polymod & q){
      order_t tmp;
      tmp=order; order=q.order; q.order=tmp;
      int tmp2=dim; dim=q.dim; q.dim=tmp2;
      tmp2=sugar; sugar=q.sugar; q.sugar=tmp2;
      coord.swap(q.coord);
      tmp2=fromleft; fromleft=q.fromleft; q.fromleft=tmp2;
      tmp2=fromright; fromright=q.fromright; q.fromright=tmp2;
      tmp2=age; age=q.age; q.age=tmp2;
      double tmp3=logz; logz=q.logz; q.logz=tmp3;
   }
    polymod():dim(0),fromleft(-1),fromright(-1),logz(1) {order_t tmp={_PLEX_ORDER,0}; order=tmp;}
    polymod(order_t o_,int dim_): dim(dim_),fromleft(-1),fromright(-1),logz(1) {order=o_; order.dim=dim_;}
    polymod(const polynome & p,order_t o_,modint m):fromleft(-1),fromright(-1),logz(1){
      order=o_;
      dim=p.dim;
      order.dim=dim;
      if (order.o%4!=3){
	if (p.is_strictly_greater==i_lex_is_strictly_greater)
	  order.o=_PLEX_ORDER;
	if (p.is_strictly_greater==i_total_revlex_is_strictly_greater)
	  order.o=_REVLEX_ORDER;
	if (p.is_strictly_greater==i_total_lex_is_strictly_greater)
	  order.o=_TDEG_ORDER;
      }
      if (p.dim>GROEBNER_VARS-(order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER))
	CERR << "Number of variables is too large to be handled by giac";
      else {
	if (!p.coord.empty()){
	  coord.reserve(p.coord.size());
	  for (unsigned i=0;i<p.coord.size();++i){
	    modint n;
	    if (p.coord[i].value.type==_ZINT)
	      n=modulo(*p.coord[i].value._ZINTptr,m);
	    else
	      n=p.coord[i].value.val % m;
	    coord.push_back(T_unsigned<modint,tdeg_t>(n,tdeg_t(p.coord[i].index,order)));
	  }
	  sugar=coord.front().u.total_degree(order);
	}
      }
    }
    void get_polynome(polynome & p) const {
      p.dim=dim;
      switch (order.o){
      case _PLEX_ORDER:
	p.is_strictly_greater=i_lex_is_strictly_greater;
	break;
      case _REVLEX_ORDER:
	p.is_strictly_greater=i_total_revlex_is_strictly_greater;
	break;
      case _3VAR_ORDER:
	p.is_strictly_greater=i_3var_is_strictly_greater;
	break;
      case _7VAR_ORDER:
	p.is_strictly_greater=i_7var_is_strictly_greater;
	break;
      case _11VAR_ORDER:
	p.is_strictly_greater=i_11var_is_strictly_greater;
	break;
      case _TDEG_ORDER:
	p.is_strictly_greater=i_total_lex_is_strictly_greater;
	break;
      }
      p.coord.clear();
      p.coord.reserve(coord.size());
      index_t idx(dim);
      for (unsigned i=0;i<coord.size();++i){
	get_index(coord[i].u,idx,order,dim);
	p.coord.push_back(monomial<gen>(coord[i].g,idx));
      }
      // if (order==_3VAR_ORDER || order==_7VAR_ORDER || order==_11VAR_ORDER) p.tsort();
    }
  }; // end polymod

  template<class T>
  void increase(vector<T> &v){
    if (v.size()!=v.capacity())
      return;
    vector<T> w;
    w.reserve(v.size()*2);
    for (unsigned i=0;i<v.size();++i){
      w.push_back(T(v[i].order,v[i].dim));
      w[i].coord.swap(v[i].coord);
    }
    v.swap(w);
  }

  template<class tdeg_t>
  struct polymod_sort_t {
    polymod_sort_t() {}
    bool operator () (const polymod<tdeg_t> & p,const polymod<tdeg_t> & q) const {
      if (q.coord.empty())
	return false;
      if (p.coord.empty())
	return true;
      if (p.coord.front().u==q.coord.front().u)
	return false;
      return tdeg_t_greater(q.coord.front().u,p.coord.front().u,p.order); // p.coord.front().u<q.coord.front().u;
      // this should be enough to sort groebner basis
    }
  };

  template<class tdeg_t>
  void smallmultmod(modint a,polymod<tdeg_t> & p,modint m,bool makepositive=true){
    if (a==1 || a==1-m)
      return;
    typename std::vector< T_unsigned<modint,tdeg_t> >::iterator pt=p.coord.begin(),ptend=p.coord.end();
    if (makepositive){
      for (;pt!=ptend;++pt){
	modint tmp=(longlong(pt->g)*a)%m;
	if (tmp<0) tmp += m;
	pt->g=tmp;
      }
    }
    else {
      for (;pt!=ptend;++pt){
	modint tmp=(longlong(pt->g)*a)%m;
	pt->g=tmp;
      }
    }
  }

  template<class tdeg_t>
  struct tdeg_t_sort_t {
    order_t order;
    tdeg_t_sort_t() {order_t tmp={_REVLEX_ORDER,0}; order=tmp;}
    tdeg_t_sort_t(order_t o):order(o) {}
    bool operator ()(const T_unsigned<modint,tdeg_t> & a,const T_unsigned<modint,tdeg_t> & b) const {return !tdeg_t_greater(b.u,a.u,order);}
    bool operator ()(const T_unsigned<gen,tdeg_t> & a,const T_unsigned<gen,tdeg_t> & b) const {return !tdeg_t_greater(b.u,a.u,order);}
  };
  template<class tdeg_t>
  void convert(const poly8<tdeg_t> & p,polymod<tdeg_t> &q,modint env){
#if 0
    q.coord.reserve(p.coord.size());
    q.dim=p.dim;
    q.order=p.order;
    q.sugar=0;
    for (unsigned i=0;i<p.coord.size();++i){
      int g=1;
      if (env){
	if (p.coord[i].g.type==_ZINT)
	  g=modulo(*p.coord[i].g._ZINTptr,env);
	else
	  g=(p.coord[i].g.val)%env;
      }
      if (g!=0)
	q.coord.push_back(T_unsigned<int,tdeg_t>(g,p.coord[i].u));
    }
#else
    q.coord.resize(p.coord.size());
    q.dim=p.dim;
    q.order=p.order;
    q.sugar=0;
    for (unsigned i=0;i<p.coord.size();++i){
      if (!env)
	q.coord[i].g=1;
      else {
	if (p.coord[i].g.type==_ZINT)
	  q.coord[i].g=modulo(*p.coord[i].g._ZINTptr,env);
	else
	  q.coord[i].g=(p.coord[i].g.val)%env;
      }
      q.coord[i].u=p.coord[i].u;
    }
#endif
    if (env && !q.coord.empty()){
      q.sugar=q.coord.front().u.total_degree(p.order);
      if (q.coord.front().g!=1)
	smallmultmod(invmod(q.coord.front().g,env),q,env);
      q.coord.front().g=1;
    }
    sort(q.coord.begin(),q.coord.end(),tdeg_t_sort_t<tdeg_t>(p.order));
  }
  template<class tdeg_t>
  void convert(const polymod<tdeg_t> & p,poly8<tdeg_t> &q,modint env){
    q.coord.resize(p.coord.size());
    q.dim=p.dim;
    q.order=p.order;
    for (unsigned i=0;i<p.coord.size();++i){
      modint n=p.coord[i].g % env;
      if (n>env/2)
	n-=env;
      else {
	if (n<=-env/2)
	  n += env;
      }
      q.coord[i].g=n;
      q.coord[i].u=p.coord[i].u;
    }
    if (!q.coord.empty())
      q.sugar=q.coord.front().u.total_degree(p.order);
    else
      q.sugar=0;
  }

  template<class tdeg_t>
  bool operator == (const polymod<tdeg_t> & p,const polymod<tdeg_t> &q){
    if (p.coord.size()!=q.coord.size())
      return false;
    for (unsigned i=0;i<p.coord.size();++i){
      if (p.coord[i].u!=q.coord[i].u || p.coord[i].g!=q.coord[i].g)
	return false;
    }
    return true;
  }

#ifdef NSPIRE
  template<class T,class tdeg_t>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os, const polymod<tdeg_t> & p)
#else
  template<class tdeg_t>
  ostream & operator << (ostream & os, const polymod<tdeg_t> & p)
#endif
  {
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    int t2;
    if (it==itend)
      return os << 0 ;
    for (;it!=itend;){
      os << it->g  ;
#ifndef GBASIS_NO_OUTPUT
      if (it->u.vars64()){
	if (it->u.tdeg%2){
	  degtype * i=(degtype *)(it->u.ui+1);
	  for (int j=0;j<it->u.order_.dim;++j){
	    t2=i[j];
	    if (t2)
	      os << "*x"<< j << "^" << t2  ;
	  }
	  ++it;
	  if (it==itend)
	    break;
	  os << " + ";
	  continue;
	}
      }
#endif
      short tab[GROEBNER_VARS+1];
      tab[GROEBNER_VARS]=0;
      it->u.get_tab(tab,p.order);
      switch (p.order.o){
      case _PLEX_ORDER:
	for (int i=0;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2)
	    os << "*x"<< i << "^" << t2  ;
	}
	break;
      case _TDEG_ORDER:
	for (int i=1;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  if (t2)
	    os << "*x"<< i-1 << "^" << t2  ;
	}
	break;
      case _REVLEX_ORDER:
	for (int i=1;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
#if GROEBNER_VARS==15
      case _3VAR_ORDER:
	for (int i=1;i<=3;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 3-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=5;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 7+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
      case _7VAR_ORDER:
	for (int i=1;i<=7;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 7-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=9;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 11+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
      case _11VAR_ORDER:
	for (int i=1;i<=11;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 11-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=13;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 15+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
#endif
      }
      ++it;
      if (it==itend)
	break;
      os << " + ";
    }
    return os;
  }

  template<class tdeg_t>
  void polymod<tdeg_t>::dbgprint() const {
    CERR << *this << '\n';
  }

  template<class tdeg_t>
  class vectpolymod:public vector<polymod<tdeg_t> >{
  public:
    void dbgprint() const { CERR << *this << '\n'; }
  };

  template<class tdeg_t>
  void vectpoly_2_vectpolymod(const vectpoly & v,order_t order,vectpolymod<tdeg_t> & v8,modint m){
    v8.clear();
    v8.reserve(v.size());
    for (unsigned i=0;i<v.size();++i){
      v8.push_back(polymod<tdeg_t>(v[i],order,m));
      v8.back().order=order;
    }
  }

  template<class tdeg_t>
  void vectpolymod_2_vectpoly(const vectpoly8<tdeg_t> & v8,vectpoly & v){
    v.clear();
    v.reserve(v8.size());
    for (unsigned i=0;i<v8.size();++i){
      v.push_back(polynome(v8[i].dim));
      v8[i].get_polynome(v[i]);
    }
  }

  template<class tdeg_t>
  void convert(const vectpoly8<tdeg_t> & v,vectpolymod<tdeg_t> & w,modint env){
    if (w.size()<v.size())
      w.resize(v.size());
    for (unsigned i=0;i<v.size();++i){
      convert(v[i],w[i],env);
    }
  }

  template<class tdeg_t>
  void convert(const vectpolymod<tdeg_t> & v,vectpoly8<tdeg_t> & w,modint env){
    w.resize(v.size());
    for (unsigned i=0;i<v.size();++i){
      convert(v[i],w[i],env);
    }
  }

  template<class tdeg_t>
  void convert(const vectpolymod<tdeg_t> & v,const vector<unsigned> & G,vectpoly8<tdeg_t> & w,modint env){
    w.resize(v.size());
    for (unsigned i=0;i<G.size();++i){
      convert(v[G[i]],w[G[i]],env);
    }
  }

#ifdef GBASIS_HEAP
  template<class tdeg_t>
  void in_heap_reducemod(const polymod<tdeg_t> & f,const vectpolymod<tdeg_t> & g,const vector<unsigned> & G,unsigned excluded,vectpolymod<tdeg_t> & q,polymod<tdeg_t> & rem,polymod<tdeg_t> * R,modint env){
    // divides f by g[G[0]] to g[G[G.size()-1]] except maybe g[G[excluded]]
    // first implementation: use quotxsient heap for all quotient/divisor
    // do not use heap chain
    // ref Monaghan Pearce if g.size()==1
    // R is the list of all monomials
    if (R){
      R->dim=f.dim; R->order=f.order;
      R->coord.clear();
    }
    if (&rem==&f){
      polymod<tdeg_t> TMP;
      in_heap_reducemod(f,g,G,excluded,q,TMP,R,env);
      swap(rem.coord,TMP.coord);
      if (debug_infolevel>1000)
	g.dbgprint(); // instantiate dbgprint()
      return;
    }
    rem.coord.clear();
    if (f.coord.empty())
      return ;
    if (q.size()<G.size())
      q.resize(G.size());
    unsigned guess=0;
    for (unsigned i=0;i<G.size();++i){
      q[i].dim=f.dim;
      q[i].order=f.order;
      q[i].coord.clear();
      guess += unsigned(g[G[i]].coord.size());
    }
    vector<heap_t<tdeg_t> > H;
    compare_heap_t<tdeg_t> key(f.order);
    H.reserve(guess);
    vector<modint> invlcg(G.size());
    for (unsigned i=0;i<G.size();++i){
      invlcg[i]=invmod(g[G[i]].coord.front().g,env);
    }
    modint c=1;
#ifdef x86_64
    int128_t C=0; // int128_t to avoid %
#else
    modint2 C=0; // int128_t to avoid %
#endif
    unsigned k=0,i; // k=position in f
    tdeg_t m;
    bool finish=false;
    while (!H.empty() || k<f.coord.size()){
      // is highest remaining degree in f or heap?
      if (k<f.coord.size() && (H.empty() || tdeg_t_greater(f.coord[k].u,H.front().u,f.order)) ){
	// it's in f or both
	m=f.coord[k].u;
	C=f.coord[k].g;
	++k;
      }
      else {
	m=H[0].u;
	C=0;
      }
      if (R)
	R->coord.push_back(T_unsigned<modint,tdeg_t>(1,m));
      // extract from heap all terms having m as monomials, substract from c
      while (!H.empty() && H.front().u==m){
	std::pop_heap(H.begin(),H.end(),key);
	heap_t<tdeg_t> & current=H.back(); // was root node of the heap
	const polymod<tdeg_t> & gcurrent = g[G[current.i]];
	if (!R){
#ifdef x86_64
	  C -= modint2(q[current.i].coord[current.qi].g) * gcurrent.coord[current.gj].g;
#else
	  C = (C-modint2(q[current.i].coord[current.qi].g) * gcurrent.coord[current.gj].g) % env;
#endif
	}
	if (current.gj<gcurrent.coord.size()-1){
	  ++current.gj;
	  current.u=q[current.i].coord[current.qi].u+gcurrent.coord[current.gj].u;
	  push_heap(H.begin(),H.end(),key);
	}
	else
	  H.pop_back();
      }
      if (!R){
#ifdef x86_64
	c = C % env;
#else
	c=modint(C);
#endif
	if (c==0)
	  continue;
      }
      // divide (c,m) by one of the g if possible, otherwise push in remainder
      if (finish){
	rem.coord.push_back(T_unsigned<modint,tdeg_t>(c,m)); // add c*m to remainder
	continue;
      }
      finish=true;
#if 0
      for (i=G.size()-1;i!=-1;--i){
	if (i==excluded || g[G[i]].coord.empty())
	  continue;
	if (tdeg_t_greater(m,g[G[i]].coord.front().u,f.order)){
	  finish=false;
	  if (tdeg_t_all_greater(m,g[G[i]].coord.front().u,f.order))
	    break;
	}
      }
      if (i==-1){
	rem.coord.push_back(T_unsigned<modint,tdeg_t>(c,m)); // add c*m to remainder
	continue;
      }
#else
      for (i=0;i<G.size();++i){
	if (i==excluded || g[G[i]].coord.empty())
	  continue;
	if (tdeg_t_greater(m,g[G[i]].coord.front().u,f.order)){
	  finish=false;
	  if (tdeg_t_all_greater(m,g[G[i]].coord.front().u,f.order))
	    break;
	}
      }
      if (i==G.size()){
	rem.coord.push_back(T_unsigned<modint,tdeg_t>(c,m)); // add c*m to remainder
	continue;
      }
#endif
      // add c*m/leading monomial of g[G[i]] to q[i]
      tdeg_t monom=m-g[G[i]].coord.front().u;
      if (!R){
	if (invlcg[i]!=1){
	  if (invlcg[i]==-1)
	    c=-c;
	  else
	    c=(modint2(c)*invlcg[i]) % env;
	}
      }
      q[i].coord.push_back(T_unsigned<modint,tdeg_t>(c,monom));
      // push in heap
      if (g[G[i]].coord.size()>1){
	heap_t<tdeg_t> current={i,unsigned(q[i].coord.size())-1,1,g[G[i]].coord[1].u+monom};
	H.push_back(current);
	push_heap(H.begin(),H.end(),key);
      }
    } // end main heap pseudo-division loop
  }

  template<class tdeg_t>
  void heap_reducemod(const polymod<tdeg_t> & f,const vectpolymod<tdeg_t> & g,const vector<unsigned> & G,unsigned excluded,vectpolymod<tdeg_t> & q,polymod<tdeg_t> & rem,modint env){
    in_heap_reducemod(f,g,G,excluded,q,rem,0,env);
    // end up by multiplying rem by s (so that everything is integer)
    if (debug_infolevel>2){
      for (unsigned i=0;i<G.size();++i)
	CERR << "(" << g[G[i]]<< ")*(" << q[i] << ")+ ";
      CERR << rem << '\n';
    }
    if (!rem.coord.empty() && rem.coord.front().g!=1){
      smallmult(invmod(rem.coord.front().g,env),rem.coord,rem.coord,env);
      rem.coord.front().g=1;
    }
  }

#endif

  template<class tdeg_t>
  void symbolic_preprocess(const polymod<tdeg_t> & f,const vectpolymod<tdeg_t> & g,const vector<unsigned> & G,unsigned excluded,vectpolymod<tdeg_t> & q,polymod<tdeg_t> & rem,polymod<tdeg_t> * R){
    // divides f by g[G[0]] to g[G[G.size()-1]] except maybe g[G[excluded]]
    // first implementation: use quotient heap for all quotient/divisor
    // do not use heap chain
    // ref Monaghan Pearce if g.size()==1
    // R is the list of all monomials
    if (R){
      R->dim=f.dim; R->order=f.order;
      R->coord.clear();
    }
    rem.coord.clear();
    if (f.coord.empty())
      return ;
    if (q.size()<G.size())
      q.resize(G.size());
    unsigned guess=0;
    for (unsigned i=0;i<G.size();++i){
      q[i].dim=f.dim;
      q[i].order=f.order;
      q[i].coord.clear();
      guess += unsigned(g[G[i]].coord.size());
    }
    vector<heap_t<tdeg_t> > H_;
    vector<unsigned> H;
    H_.reserve(guess);
    H.reserve(guess);
    heap_t_compare<tdeg_t> keyheap(H_,f.order);
    unsigned k=0,i; // k=position in f
    tdeg_t m;
    bool finish=false;
    while (!H.empty() || k<f.coord.size()){
      // is highest remaining degree in f or heap?
      if (k<f.coord.size() && (H.empty() || tdeg_t_greater(f.coord[k].u,H_[H.front()].u,f.order)) ){
	// it's in f or both
	m=f.coord[k].u;
	++k;
      }
      else {
	m=H_[H.front()].u;
      }
      if (R)
	R->coord.push_back(T_unsigned<modint,tdeg_t>(1,m));
      // extract from heap all terms having m as monomials, substract from c
      while (!H.empty() && H_[H.front()].u==m){
	std::pop_heap(H.begin(),H.end(),keyheap);
	heap_t<tdeg_t> & current=H_[H.back()]; // was root node of the heap
	const polymod<tdeg_t> & gcurrent = g[G[current.i]];
	if (current.gj<gcurrent.coord.size()-1){
	  ++current.gj;
	  current.u=q[current.i].coord[current.qi].u+gcurrent.coord[current.gj].u;
	  std::push_heap(H.begin(),H.end(),keyheap);
	}
	else
	  H.pop_back();
      }
      // divide (c,m) by one of the g if possible, otherwise push in remainder
      if (finish){
	rem.coord.push_back(T_unsigned<modint,tdeg_t>(1,m)); // add to remainder
	continue;
      }
      finish=true;
      for (i=0;i<G.size();++i){
	const vector< T_unsigned<modint,tdeg_t> > & gGicoord=g[G[i]].coord;
	if (i==excluded || gGicoord.empty())
	  continue;
	if (tdeg_t_greater(m,gGicoord.front().u,f.order)){
	  finish=false;
	  if (tdeg_t_all_greater(m,gGicoord.front().u,f.order))
	    break;
	}
      }
      if (i==G.size()){
	rem.coord.push_back(T_unsigned<modint,tdeg_t>(1,m)); // add to remainder
	continue;
      }
      // add m/leading monomial of g[G[i]] to q[i]
      tdeg_t monom=m-g[G[i]].coord.front().u;
      q[i].coord.push_back(T_unsigned<modint,tdeg_t>(1,monom));
      // push in heap
      if (g[G[i]].coord.size()>1){
	heap_t<tdeg_t> current={i,unsigned(q[i].coord.size())-1,1,g[G[i]].coord[1].u+monom};
	H.push_back(unsigned(H_.size()));
	H_.push_back(current);
	keyheap.ptr=&H_.front();
	std::push_heap(H.begin(),H.end(),keyheap);
      }
    } // end main heap pseudo-division loop
    // CERR << H_.size() << '\n';
  }

  // p - a*q shifted mod m -> r
  template<class tdeg_t>
  void smallmultsubmodshift(const polymod<tdeg_t> & p,unsigned pos,modint a,const polymod<tdeg_t> & q,const tdeg_t & shift,polymod<tdeg_t> & r,modint m){
    r.coord.clear();
    r.coord.reserve(p.coord.size()+q.coord.size());
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it0=p.coord.begin(),it=it0+pos,itend=p.coord.end(),jt=q.coord.begin(),jtend=q.coord.end();
    // for (;it0!=it;++it0){ r.coord.push_back(*it0); }
    tdeg_t v=shift+shift; // new memory slot
    int dim=p.dim;
    for (;jt!=jtend;++jt){
      add(jt->u,shift,v,dim);
      for (;it!=itend && tdeg_t_strictly_greater(it->u,v,p.order);++it){
	r.coord.push_back(*it);
      }
      if (it!=itend && it->u==v){
	modint tmp=(it->g-modint2(a)*jt->g)%m;
	if (tmp)
	  r.coord.push_back(T_unsigned<modint,tdeg_t>(tmp,v));
	++it;
      }
      else {
	modint tmp=(-modint2(a)*jt->g)%m;
	r.coord.push_back(T_unsigned<modint,tdeg_t>(tmp,v));
      }
    }
    for (;it!=itend;++it){
      r.coord.push_back(*it);
    }
  }

  // p - a*q mod m -> r
  template<class tdeg_t>
  void smallmultsubmod(const polymod<tdeg_t> & p,modint a,const polymod<tdeg_t> & q,polymod<tdeg_t> & r,modint m){
    r.coord.clear();
    r.coord.reserve(p.coord.size()+q.coord.size());
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=q.coord.begin(),jtend=q.coord.end();
    for (;jt!=jtend;++jt){
      const tdeg_t & v=jt->u;
      for (;it!=itend && tdeg_t_strictly_greater(it->u,v,p.order);++it){
	r.coord.push_back(*it);
      }
      if (it!=itend && it->u==v){
	modint tmp=(it->g-modint2(a)*jt->g)%m;
	if (tmp)
	  r.coord.push_back(T_unsigned<modint,tdeg_t>(tmp,v));
	++it;
      }
      else {
	int tmp=(-modint2(a)*jt->g)%m;
	r.coord.push_back(T_unsigned<modint,tdeg_t>(tmp,v));
      }
    }
    for (;it!=itend;++it){
      r.coord.push_back(*it);
    }
  }

  // p + q  -> r
  template<class tdeg_t>
  void smallmerge(polymod<tdeg_t> & p,polymod<tdeg_t> & q,polymod<tdeg_t> & r){
    if (p.coord.empty()){
      swap(q.coord,r.coord);
      return;
    }
    if (q.coord.empty()){
      swap(p.coord,r.coord);
      return;
    }
    r.coord.clear();
    r.coord.reserve(p.coord.size()+q.coord.size());
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=q.coord.begin(),jtend=q.coord.end();
    for (;jt!=jtend;++jt){
      const tdeg_t & v=jt->u;
      for (;it!=itend && tdeg_t_strictly_greater(it->u,v,p.order);++it){
	r.coord.push_back(*it);
      }
      r.coord.push_back(*jt);
    }
    for (;it!=itend;++it){
      r.coord.push_back(*it);
    }
  }

  template<class tdeg_t>
  void reducemod(const polymod<tdeg_t> & p,const vectpolymod<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,polymod<tdeg_t> & rem,modint env,bool topreduceonly=false){
    if (&p!=&rem)
      rem=p;
    if (p.coord.empty())
      return ;
    polymod<tdeg_t> TMP2(p.order,p.dim);
    unsigned i,rempos=0;
    for (unsigned count=0;;++count){
      // this branch search first in all leading coeff of G for a monomial
      // <= to the current rem monomial
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator pt=rem.coord.begin()+rempos;
      if (pt>=rem.coord.end())
	break;
      for (i=0;i<G.size();++i){
	if (i==excluded || res[G[i]].coord.empty())
	  continue;
	if (tdeg_t_all_greater(pt->u,res[G[i]].coord.front().u,p.order))
	  break;
      }
      if (i==G.size()){ // no leading coeff of G is smaller than the current coeff of rem
	++rempos;
	if (topreduceonly)
	  break;
	// if (small0) TMP1.coord.push_back(*pt);
	continue;
      }
      modint a(pt->g),b(res[G[i]].coord.front().g);
      if (pt->u==res[G[i]].coord.front().u){
	smallmultsubmod(rem,smod(modint2(a)*invmod(b,env),env),res[G[i]],TMP2,env);
	// Gpos=i; // assumes basis element in G are sorted wrt >
      }
      else
	smallmultsubmodshift(rem,0,smod(modint2(a)*invmod(b,env),env),res[G[i]],pt->u-res[G[i]].coord.front().u,TMP2,env);
      swap(rem.coord,TMP2.coord);
    }
    if (!rem.coord.empty() && rem.coord.front().g!=1){
      smallmultmod(invmod(rem.coord.front().g,env),rem,env);
      rem.coord.front().g=1;
    }
  }

#if 0
  // reduce with respect to itself the elements of res with index in G
  template<class tdeg_t>
  void reducemod(vectpolymod<tdeg_t> & res,vector<unsigned> G,modint env){
    if (res.empty() || G.empty())
      return;
    polymod<tdeg_t> pred(res.front().order,res.front().dim),
      TMP2(res.front().order,res.front().dim);
    vectpolymod<tdeg_t> q;
    // reduce res
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      polymod<tdeg_t> & p=res[i];
      reducemod(p,res,G,i,q,pred,TMP2,env);
      swap(res[i].coord,pred.coord);
      pred.sugar=res[i].sugar;
    }
  }
#endif

  template<class tdeg_t>
  void spolymod(const polymod<tdeg_t> & p,const polymod<tdeg_t> & q,polymod<tdeg_t> & res,polymod<tdeg_t> & TMP1,modint env){
    if (p.coord.empty()){
      res=q;
      return ;
    }
    if (q.coord.empty()){
      res= p;
      return;
    }
    const tdeg_t & pi = p.coord.front().u;
    const tdeg_t & qi = q.coord.front().u;
    tdeg_t lcm;
    index_lcm(pi,qi,lcm,p.order);
    //polymod<tdeg_t> TMP1(p);
    TMP1=p;
    // polymod<tdeg_t> TMP2(q);
    const polymod<tdeg_t> &TMP2=q;
    modint a=p.coord.front().g,b=q.coord.front().g;
    tdeg_t pshift=lcm-pi;
    unsigned sugarshift=pshift.total_degree(p.order);
    // adjust sugar for res
    res.sugar=p.sugar+sugarshift;
    // CERR << "spoly mod " << res.sugar << " " << pi << qi << '\n';
    if (p.order.o==_PLEX_ORDER || sugarshift!=0)
      smallshift(TMP1.coord,pshift,TMP1.coord);
    // smallmultmod(b,TMP1,env);
    if (lcm==qi)
      smallmultsubmod(TMP1,smod(modint2(a)*invmod(b,env),env),TMP2,res,env);
    else
      smallmultsubmodshift(TMP1,0,smod(modint2(a)*invmod(b,env),env),TMP2,lcm-qi,res,env);
    if (!res.coord.empty() && res.coord.front().g!=1){
      smallmultmod(invmod(res.coord.front().g,env),res,env);
      res.coord.front().g=1;
    }
    if (debug_infolevel>2)
      CERR << "spolymod " << res << '\n';
  }

  template<class tdeg_t>
  void reduce1smallmod(polymod<tdeg_t> & p,const polymod<tdeg_t> & q,polymod<tdeg_t> & TMP2,modint env){
    if (p.coord.empty())
      return ;
    unsigned rempos=0;
    const tdeg_t & u = q.coord.front().u;
    const modint invg=invmod(q.coord.front().g,env);
    for (unsigned count=0;;++count){
      // this branch search first in all leading coeff of G for a monomial
      // <= to the current rem monomial
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator pt=p.coord.begin()+rempos;
      if (pt>=p.coord.end())
	break;
      if (pt->u==u){
	smallmultsubmodshift(p,0,smod(modint2(pt->g)*invg,env),q,pt->u-u,TMP2,env);
	swap(p.coord,TMP2.coord);
	break;
      }
      if (!tdeg_t_all_greater(pt->u,u,p.order)){
	++rempos;
	// TMP1.coord.push_back(*pt);
	continue;
      }
      smallmultsubmodshift(p,0,smod(modint2(pt->g)*invg,env),q,pt->u-u,TMP2,env);
      // smallmultsubmodshift(p,rempos,smod(modint2(pt->g)*invmod(g,env),env),q,pt->u-u,TMP2,env);
      rempos=0;
      swap(p.coord,TMP2.coord);
    }
    // if (small0) swap(p.coord,TMP1.coord);
    if (!p.coord.empty() && p.coord.front().g!=1){
      smallmultmod(invmod(p.coord.front().g,env),p,env);
      p.coord.front().g=1;
    }
  }

#define GIAC_GBASIS_PERMUTATION
  //#define GIAC_GBASIS_PERMUTATION2
  template<class tdeg_t>
  struct zsymb_data {
    unsigned pos;
    tdeg_t deg;
    order_t o;
    unsigned terms;
    int age;
  };

#ifdef GIAC_GBASIS_PERMUTATION2
  template<class tdeg_t>
  bool tri (const zsymb_data<tdeg_t> & z1,const zsymb_data<tdeg_t> & z2){
    if (z1.terms!=z2.terms)
      return z1.terms<z2.terms;
    int d1=z1.deg.total_degree(z1.o),d2=z2.deg.total_degree(z2.o);
    if (d1!=d2) return d1<d2;
    if (z1.pos!=z2.pos)
      return z2.pos>z1.pos;
    return false;
  }
#endif

  // #define GIAC_DEG_FIRST

  template <class tdeg_t>
  bool operator < (const zsymb_data<tdeg_t> & z1,const zsymb_data<tdeg_t> & z2){
    // reductor choice: less terms is better
    // but small degree gives a reductor sooner
    // e.g. less terms is faster for botana* but slower for cyclic*
    // if (z1.terms!=z2.terms){ return z2.terms>z1.terms; }
    int d1=z1.deg.total_degree(z1.o),d2=z2.deg.total_degree(z2.o);
#ifdef GIAC_DEG_FIRST
    if (d1!=d2)
      return d1<d2;
#endif
    // double Z1=d1*double(z1.terms)*(z1.age+1); double Z2=d2*double(z2.terms)*(z2.age+1); if (Z1!=Z2) return Z2>Z1;
    double Z1=z1.terms*double(z1.terms)*d1; double Z2=z2.terms*double(z2.terms)*d2; if (Z1!=Z2) return Z2>Z1;
    //double Z1=double(z1.terms)*d1; double Z2=double(z2.terms)*d2; if (Z1!=Z2) return Z2>Z1;
    if (z1.terms!=z2.terms) return z2.terms>z1.terms;
    if (z1.deg!=z2.deg)
      return tdeg_t_greater(z1.deg,z2.deg,z1.o)!=0;
    if (z1.pos!=z2.pos)
      return z2.pos>z1.pos;
    return false;
  }

  template<class tdeg_t>
  void reducesmallmod(polymod<tdeg_t> & rem,const vectpolymod<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,modint env,polymod<tdeg_t> & TMP1,bool normalize,int start_index=0,bool topreduceonly=false){
    if (debug_infolevel>1000){
      rem.dbgprint();
      if (!rem.coord.empty()) rem.coord.front().u.dbgprint();
    }
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator pt,ptend;
    unsigned i,rempos=0;
    TMP1.coord.clear();
    unsigned Gs=unsigned(G.size());
    const order_t o=rem.order;
    int Gstart_index=0;
    // starting at excluded may fail because the batch of new basis element created
    // by f4mod is reduced with respect to the previous batches but not necessarily
    // with respect to itself
    if (start_index && excluded<Gs){
      for (int i=excluded;i>=0;--i){
	int Gi=G[i];
	if (Gi<=start_index){
	  Gstart_index=i;
	  break;
	}
      }
    }
#ifdef GIAC_GBASIS_PERMUTATION2
    if (excluded<Gs)
      excluded=G[excluded];
    else
      excluded=-1;
    vector<zsymb_data> zsGi(Gs);
    for (unsigned i=0;i<Gs;++i){
      int Gi=G[i];
      const polymod<tdeg_t> & cur=res[Gi];
      zsymb_data tmp={Gi,cur.coord.empty()?0:cur.coord.front().u,o,cur.coord.size(),0};
      zsGi[i]=tmp;
    }
    sort(zsGi.begin(),zsGi.end(),tri);
#else
    const tdeg_t ** resGi=(const tdeg_t **) malloc(Gs*sizeof(tdeg_t *));
    for (unsigned i=0;i<Gs;++i){
      resGi[i]=res[G[i]].coord.empty()?0:&res[G[i]].coord.front().u;
    }
#endif
    for (unsigned count=0;;++count){
      ptend=rem.coord.end();
      // this branch search first in all leading coeff of G for a monomial
      // <= to the current rem monomial
      pt=rem.coord.begin()+rempos;
      if (pt>=ptend)
	break;
      const tdeg_t &ptu=pt->u;
#ifdef GIAC_GBASIS_PERMUTATION2
      for (i=0;i<Gs;++i){
	zsymb_data & zs=zsGi[i];
	if (zs.terms && zs.pos!=excluded && tdeg_t_all_greater(ptu,zs.deg,o))
	  break;
      }
#else // GIAC_GBASIS_PERMUTATION2
      if (excluded<Gs){
	i=Gs;
	const tdeg_t ** resGi_=resGi+excluded+1,**resGiend=resGi+Gs;
	for (;resGi_<resGiend;++resGi_){
	  if (!*resGi_)
	    continue;
	  if (tdeg_t_all_greater(ptu,**resGi_,o)){
	    i=unsigned(resGi_-resGi);
	    break;
	  }
	}
	if (i==Gs){
	  resGi_=resGi+Gstart_index;
	  resGiend=resGi+excluded;
	  for (;resGi_<resGiend;++resGi_){
	    if (!*resGi_)
	      continue;
	    if (tdeg_t_all_greater(ptu,**resGi_,o)){
	      i=unsigned(resGi_-resGi);
	      break;
	    }
	  }
	}
      }
      else {
	const tdeg_t ** resGi_=resGi,**resGiend=resGi+Gs;
	for (;resGi_<resGiend;++resGi_){
	  if (!*resGi_)
	    continue;
#if 0
	  int res=tdeg_t_compare_all(ptu,**resGi_,o);
	  if (res==1) break;
	  if (res==-1)
	    *resGi_=0;
#else
	  if (tdeg_t_all_greater(ptu,**resGi_,o))
	    break;
#endif
	}
	i=unsigned(resGi_-resGi);
      }
#endif // GIAC_GBASIS_PERMUTATION2
      if (i==G.size()){ // no leading coeff of G is smaller than the current coeff of rem
	++rempos;
	if (topreduceonly)
	  break;
	// if (small0) TMP1.coord.push_back(*pt);
	continue;
      }
#ifdef GIAC_GBASIS_PERMUTATION2
      int Gi=zsGi[i].pos;
#else
      int Gi=G[i];
#endif
      modint a(pt->g),b(res[Gi].coord.front().g);
      smallmultsubmodshift(rem,0,smod(a*modint2(invmod(b,env)),env),res[Gi],pt->u-res[Gi].coord.front().u,TMP1,env);
      // smallmultsub(rem,rempos,smod(a*invmod(b,env->modulo),env->modulo).val,res[G[i]],pt->u-res[G[i]].coord.front().u,TMP2,env->modulo.val);
      // rempos=0; // since we have removed the beginning of rem (copied in TMP1)
      swap(rem.coord,TMP1.coord);
      continue;
    }
    if (normalize && !rem.coord.empty() && rem.coord.front().g!=1){
      // smallmult does %, smallmultmod also make the result positive
      // smallmult(invmod(rem.coord.front().g,env),rem.coord,rem.coord,env);
      smallmultmod(invmod(rem.coord.front().g,env),rem,env,false);
      rem.coord.front().g=1;
    }
#ifndef GIAC_GBASIS_PERMUTATION2
    free(resGi);
#endif
  }

  template<class tdeg_t>
  static void reducemod(vectpolymod<tdeg_t> &resmod,modint env){
    if (resmod.empty())
      return;
    // Initial interreduce step
    polymod<tdeg_t> TMP1(resmod.front().order,resmod.front().dim);
    vector<unsigned> G(resmod.size());
    for (unsigned j=0;j<G.size();++j)
      G[j]=j;
    for (unsigned j=0; j<resmod.size();++j){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      reducesmallmod(resmod[j],resmod,G,j,env,TMP1,true);
    }
  }

  template<class tdeg_t>
  void gbasis_updatemod(vector<unsigned> & G,vector< paire > & B,vectpolymod<tdeg_t> & res,unsigned pos,polymod<tdeg_t> & TMP2,modint env,bool reduce,const vector<unsigned> & oldG){
    if (debug_infolevel>2)
      CERR << CLOCK()*1e-6 << " mod begin gbasis update " << G.size() << '\n';
    if (debug_infolevel>3)
      CERR << G << '\n';
    const polymod<tdeg_t> & h = res[pos];
    if (h.coord.empty())
      return;
    order_t order=h.order;
    vector<unsigned> C;
    C.reserve(G.size()+1);
    const tdeg_t & h0=h.coord.front().u;
    // FIXME: should use oldG instead of G here
    for (unsigned i=0;i<oldG.size();++i){
      if (tdeg_t_all_greater(h0,res[oldG[i]].coord.front().u,order))
	return;
    }
    tdeg_t tmp1,tmp2;
    // C is used to construct new pairs
    // create pairs with h and elements g of G, then remove
    // -> if g leading monomial is prime with h, remove the pair
    // -> if g leading monomial is not disjoint from h leading monomial
    //    keep it only if lcm of leading monomial is not divisible by another one
#if 1
    size_t tmpsize=G.size();
    vector<tdeg_t> tmp(tmpsize);
    for (unsigned i=0;i<tmpsize;++i){
      if (res[G[i]].coord.empty()){
	tmp[i].tab[0]=-2;
      }
      else
	index_lcm(h0,res[G[i]].coord.front().u,tmp[i],order);
    }
#else
    // this would be faster but it does not work for
    // gbasis([25*y^2*x^6-10*y^2*x^5+59*y^2*x^4-20*y^2*x^3+43*y^2*x^2-10*y^2*x+9*y^2-80*y*x^6+136*y*x^5+56*y*x^4-240*y*x^3+104*y*x^2+64*x^6-192*x^5+192*x^4-64*x^3,25*y^2*6*x^5-10*y^2*5*x^4+59*y^2*4*x^3-20*y^2*3*x^2+43*y^2*2*x-10*y^2-80*y*6*x^5+136*y*5*x^4+56*y*4*x^3-240*y*3*x^2+104*y*2*x+64*6*x^5-192*5*x^4+192*4*x^3-64*3*x^2,25*2*y*x^6-10*2*y*x^5+59*2*y*x^4-20*2*y*x^3+43*2*y*x^2-10*2*y*x+9*2*y-80*x^6+136*x^5+56*x^4-240*x^3+104*x^2],[x,y],revlex);
    // pair <4,3> is not generated
    unsigned tmpsize=G.empty()?0:G.back()+1;
    vector<tdeg_t> tmp(tmpsize);
    for (unsigned i=0;i<tmpsize;++i){
      if (res[i].coord.empty()){
	tmp[i].tab[0]=-2;
      }
      else
	index_lcm(h0,res[i].coord.front().u,tmp[i],order);
    }
#endif
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (res[G[i]].coord.empty() || disjoint(h0,res[G[i]].coord.front().u,res.front().order,res.front().dim))
	continue;
      // h0 and G[i] leading monomial not prime together
#if 1
      tdeg_t * tmp1=&tmp[i];
#else
      tdeg_t * tmp1=&tmp[G[i]];
#endif
      tdeg_t * tmp2=&tmp[0],*tmpend=tmp2+tmpsize;
      for (;tmp2!=tmp1;++tmp2){
	if (tmp2->tab[0]<0)
	  continue;
	if (tdeg_t_all_greater(*tmp1,*tmp2,order))
	  break; // found another pair, keep the smallest, or the first if equal
      }
      if (tmp2!=tmp1)
	continue;
      for (++tmp2;tmp2<tmpend;++tmp2){
	if (tmp2->tab[0]<0)
	  continue;
	if (tdeg_t_all_greater(*tmp1,*tmp2,order) && *tmp1!=*tmp2){
	  break;
	}
      }
      if (tmp2==tmpend)
	C.push_back(G[i]);
    }
    vector< paire > B1;
    B1.reserve(B.size()+C.size());
    for (unsigned i=0;i<B.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (res[B[i].first].coord.empty() || res[B[i].second].coord.empty())
	continue;
      index_lcm(res[B[i].first].coord.front().u,res[B[i].second].coord.front().u,tmp1,order);
      if (!tdeg_t_all_greater(tmp1,h0,order)){
	B1.push_back(B[i]);
	continue;
      }
      index_lcm(res[B[i].first].coord.front().u,h0,tmp2,order);
      if (tmp2==tmp1){
	B1.push_back(B[i]);
	continue;
      }
      index_lcm(res[B[i].second].coord.front().u,h0,tmp2,order);
      if (tmp2==tmp1){
	B1.push_back(B[i]);
	continue;
      }
    }
    // B <- B union pairs(h,g) with g in C
    for (unsigned i=0;i<C.size();++i){
      B1.push_back(paire(pos,C[i]));
    }
    swap(B1,B);
    // Update G by removing elements with leading monomial >= leading monomial of h
    if (debug_infolevel>2){
      CERR << CLOCK()*1e-6 << " end, pairs:"<< '\n';
      if (debug_infolevel>3)
	CERR << B << '\n';
      CERR << "mod begin Groebner interreduce " << '\n';
    }
    C.clear();
    C.reserve(G.size()+1);
    // bool pos_pushed=false;
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (!res[G[i]].coord.empty() && !tdeg_t_all_greater(res[G[i]].coord.front().u,h0,order)){
	if (reduce){
	  // reduce res[G[i]] with respect to h
	  reduce1smallmod(res[G[i]],h,TMP2,env);
	}
	C.push_back(G[i]);
      }
      // NB: removing all pairs containing i in it does not work
    }
    if (debug_infolevel>2)
      CERR << CLOCK()*1e-6 << " mod end Groebner interreduce " << '\n';
    C.push_back(pos);
    swap(C,G);
#if 0
    // clear in res polymod<tdeg_t> that are no more referenced
    vector<bool> used(res.size(),false);
    for (unsigned i=0;i<G.size();++i){
      used[G[i]]=true;
    }
    for (unsigned i=0;i<B.size();++i){
      used[B[i].first]=true;
      used[B[i].first]=false;
    }
    for (unsigned i=0;i<res.size();++i){
      if (!used[i] && !res[i].coord.capacity()){
	polymod<tdeg_t> clearer;
	swap(res[i].coord,clearer.coord);
      }
    }
#endif
  }

#if 0
  // update G, G is a list of index of the previous gbasis + new spolys
  // new spolys index are starting at debut
  template<class tdeg_t>
  void gbasis_multiupdatemod(vector<unsigned> & G,vector< paire > & B,vectpolymod<tdeg_t> & res,unsigned debut,polymod<tdeg_t> & TMP2,modint env){
    if (debug_infolevel>2)
      CERR << CLOCK()*1e-6 << " mod begin gbasis update " << G.size() << "+" << add.size() << '\n';
    if (debug_infolevel>3)
      CERR << G << '\n';
    vector<unsigned> C;
    // C is used to construct new pairs
    tdeg_t tmp1,tmp2;
    order_t order;
    for (unsigned pos=debut;pos<G.size();++pos){
      const polymod<tdeg_t> & h = res[pos];
      const tdeg_t & h0=h.coord.front().u;
      // create pairs with h and elements g of G, then remove
      // -> if g leading monomial is prime with h, remove the pair
      // -> if g leading monomial is not disjoint from h leading monomial
      //    keep it only if lcm of leading monomial is not divisible by another one
      for (unsigned i=0;i<pos;++i){
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return;
	if (res[G[i]].coord.empty() || disjoint(h0,res[G[i]].coord.front().u,res.front().order,res.front().dim))
	  continue;
	index_lcm(h0,res[G[i]].coord.front().u,tmp1,order); // h0 and G[i] leading monomial not prime together
	unsigned j;
	for (j=0;j<G.size();++j){
	  if (i==j || res[G[j]].coord.empty())
	    continue;
	  index_lcm(h0,res[G[j]].coord.front().u,tmp2,order);
	  if (tdeg_t_all_greater(tmp1,tmp2,order)){
	    // found another pair, keep the smallest, or the first if equal
	    if (tmp1!=tmp2)
	      break;
	    if (i>j)
	      break;
	  }
	} // end for j
	if (j==G.size())
	  C.push_back(G[i]);
      }
    }
    vector< paire > B1;
    B1.reserve(B.size()+C.size());
    for (unsigned i=0;i<B.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (res[B[i].first].coord.empty() || res[B[i].second].coord.empty())
	continue;
      index_lcm(res[B[i].first].coord.front().u,res[B[i].second].coord.front().u,tmp1,order);
      for (unsigned pos=debut;pos<G.size();++pos){
	const tdeg_t & h0=res[G[pos]].front().u;
	if (!tdeg_t_all_greater(tmp1,h0,order)){
	  B1.push_back(B[i]);
	  break;
	}
	index_lcm(res[B[i].first].coord.front().u,h0,tmp2,order);
	if (tmp2==tmp1){
	  B1.push_back(B[i]);
	  break;
	}
	index_lcm(res[B[i].second].coord.front().u,h0,tmp2,order);
	if (tmp2==tmp1){
	  B1.push_back(B[i]);
	  break;
	}
      }
    }
    // B <- B union B2
    for (unsigned i=0;i<B2.size();++i)
      B1.push_back(B2[i]);
    swap(B1,B);
    // Update G by removing elements with leading monomial >= leading monomial of h
    if (debug_infolevel>2){
      CERR << CLOCK()*1e-6 << " end, pairs:"<< '\n';
      if (debug_infolevel>3)
	CERR << B << '\n';
      CERR << "mod begin Groebner interreduce " << '\n';
    }
    vector<unsigned> C;
    C.reserve(G.size());
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (res[G[i]].coord.empty())
	continue;
      const tdeg_t & Gi=res[G[i]].coord.front().u;
      unsigned j;
      for (j=i+1;j<G.size();++j){
	if (tdeg_t_all_greater(Gi,res[G[j]].coord.front().u,order))
	  break;
      }
      if (i==G.size()-1 || j==G.size())
	C.push_back(G[i]);
    }
    if (debug_infolevel>2)
      CERR << CLOCK()*1e-6 << " mod end Groebner interreduce " << '\n';
    swap(C,G);
  }
#endif

  template<class tdeg_t>
  bool in_gbasismod(vectpoly8<tdeg_t> & res8,vectpolymod<tdeg_t> &res,vector<unsigned> & G,modint env,bool sugar,vector< paire > * pairs_reducing_to_zero){
    convert(res8,res,env);
    unsigned ressize=unsigned(res8.size());
    unsigned learned_position=0;
    bool learning=pairs_reducing_to_zero && pairs_reducing_to_zero->empty();
    if (debug_infolevel>1000)
      res.dbgprint(); // instantiate dbgprint()
    polymod<tdeg_t> TMP1(res.front().order,res.front().dim),TMP2(res.front().order,res.front().dim);
    vector< paire > B;
    order_t order=res.front().order;
    if (order.o==_PLEX_ORDER)
      sugar=false;
    vector<unsigned> oldG(G);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " initial reduction/gbasis_updatemod: " << ressize << '\n';
    for (unsigned l=0;l<ressize;++l){
#ifdef GIAC_REDUCEMODULO
      reducesmallmod(res[l],res,G,-1,env,TMP2,env!=0);
#endif
      gbasis_updatemod(G,B,res,l,TMP2,env,true,oldG);
    }
    for (;!B.empty() && !interrupted && !ctrl_c;){
      oldG=G;
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " mod number of pairs: " << B.size() << ", base size: " << G.size() << '\n';
      // find smallest lcm pair in B
      tdeg_t small0,cur;
      unsigned smallpos,smallsugar=0,cursugar=0;
      for (smallpos=0;smallpos<B.size();++smallpos){
	if (!res[B[smallpos].first].coord.empty() && !res[B[smallpos].second].coord.empty())
	  break;
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return false;
      }
      index_lcm(res[B[smallpos].first].coord.front().u,res[B[smallpos].second].coord.front().u,small0,order);
      if (sugar)
	smallsugar=res[B[smallpos].first].sugar+(small0-res[B[smallpos].first].coord.front().u).total_degree(order);
      for (unsigned i=smallpos+1;i<B.size();++i){
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return false;
	if (res[B[i].first].coord.empty() || res[B[i].second].coord.empty())
	  continue;
	bool doswap=false;
	index_lcm(res[B[i].first].coord.front().u,res[B[i].second].coord.front().u,cur,order);
	if (sugar)
	  cursugar=res[B[smallpos].first].sugar+(cur-res[B[smallpos].first].coord.front().u).total_degree(order);
	if (order.o==_PLEX_ORDER)
	  doswap=tdeg_t_strictly_greater(small0,cur,order);
	else {
	  if (cursugar!=smallsugar)
	    doswap = smallsugar > cursugar;
	  else
	    doswap=tdeg_t_strictly_greater(small0,cur,order);
	}
	if (doswap){
	  // CERR << "swap mod " << cursugar << " " << res[B[i].first].coord.front().u << " " << res[B[i].second].coord.front().u << '\n';
	  swap(small0,cur); // small0=cur;
	  swap(smallsugar,cursugar);
	  smallpos=i;
	}
      }
      paire bk=B[smallpos];
      B.erase(B.begin()+smallpos);
      if (pairs_reducing_to_zero && learned_position<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[learned_position]){
	++learned_position;
	continue;
      }
      if (debug_infolevel>1 && (equalposcomp(G,bk.first)==0 || equalposcomp(G,bk.second)==0))
	CERR << CLOCK()*1e-6 << " mod reducing pair with 1 element not in basis " << bk << '\n';
      // polymod<tdeg_t> h(res.front().order,res.front().dim);
      spolymod<tdeg_t>(res[bk.first],res[bk.second],TMP1,TMP2,env);
      if (debug_infolevel>1){
	CERR << CLOCK()*1e-6 << " mod reduce begin, pair " << bk << " spoly size " << TMP1.coord.size() << " sugar deg " << TMP1.sugar << " degree " << TMP1.coord.front().u << '\n';
      }
      reducemod(TMP1,res,G,-1,TMP1,env);
      if (debug_infolevel>1){
	if (debug_infolevel>2){ CERR << TMP1 << '\n'; }
	CERR << CLOCK()*1e-6 << " mod reduce end, remainder size " << TMP1.coord.size() << '\n';
      }
      if (!TMP1.coord.empty()){
	if (ressize==res.size())
	  res.push_back(polymod<tdeg_t>(TMP1.order,TMP1.dim));
	swap(res[ressize],TMP1);
	++ressize;
	gbasis_updatemod(G,B,res,ressize-1,TMP2,env,true,oldG);
	if (debug_infolevel>2)
	  CERR << CLOCK()*1e-6 << " mod basis indexes " << G << " pairs indexes " << B << '\n';
      }
      else {
	if (learning && pairs_reducing_to_zero)
	  pairs_reducing_to_zero->push_back(bk);
      }
    }
    if (ressize<res.size())
      res.resize(ressize);
    // sort(res.begin(),res.end(),tripolymod<tdeg_t>);
    convert(res,G,res8,env);
    return true;
  }

  // F4BUCHBERGER algorithm
  template<class tdeg_t>
  struct heap_tt {
    bool left;
    unsigned f4buchbergervpos:31;
    unsigned polymodpos;
    tdeg_t u;
    heap_tt(bool l,unsigned a,unsigned b,tdeg_t t):left(l),f4buchbergervpos(a),polymodpos(b),u(t){};
    heap_tt(unsigned a,unsigned b,tdeg_t t):left(true),f4buchbergervpos(a),polymodpos(b),u(t){};
    heap_tt():left(true),f4buchbergervpos(0),polymodpos(0),u(){};
  };

  template<class tdeg_t>
  struct heap_tt_compare {
    order_t order;
    const heap_tt<tdeg_t> * ptr;
    inline bool operator () (unsigned a,unsigned b){
      return !tdeg_t_greater((ptr+a)->u,(ptr+b)->u,order);
      // return (ptr+a)->u<(ptr+b)->u;
    }
    heap_tt_compare(const vector<heap_tt<tdeg_t> > & v,order_t o):order(o),ptr(v.empty()?0:&v.front()){};
  };


  template<class tdeg_t>
  struct compare_heap_tt {
    order_t order;
    inline bool operator () (const heap_tt<tdeg_t> & a,const heap_tt<tdeg_t> & b){
      return !tdeg_t_greater(a.u,b.u,order);
      // return (ptr+a)->u<(ptr+b)->u;
    }
    compare_heap_tt(order_t o):order(o) {}
  };


  // inline bool operator > (const heap_tt & a,const heap_tt & b){ return a.u>b.u; }

  // inline bool operator < (const heap_tt & a,const heap_tt & b){ return b>a;}

  template<class tdeg_t>
  struct heap_tt_ptr {
    heap_tt<tdeg_t> * ptr;
    heap_tt_ptr(heap_tt<tdeg_t> * ptr_):ptr(ptr_){};
    heap_tt_ptr():ptr(0){};
  };


  // inline bool operator > (const heap_tt_ptr & a,const heap_tt_ptr & b){ return a.ptr->u > b.ptr->u; }

  // inline bool operator < (const heap_tt_ptr & a,const heap_tt_ptr & b){ return b>a; }
  template<class tdeg_t>
  struct compare_heap_tt_ptr {
    order_t order;
    inline bool operator () (const heap_tt_ptr<tdeg_t> & a,const heap_tt_ptr<tdeg_t> & b){
      return !tdeg_t_greater(a.ptr->u,b.ptr->u,order);
      // return (ptr+a)->u<(ptr+b)->u;
    }
    compare_heap_tt_ptr(order_t o):order(o) {}
  };


  template<class tdeg_t>
  void collect(const vectpolymod<tdeg_t> & f4buchbergerv,polymod<tdeg_t> & allf4buchberger,int start=0){
    typename vectpolymod<tdeg_t>::const_iterator it=f4buchbergerv.begin(),itend=f4buchbergerv.end();
    vector<heap_tt<tdeg_t> > Ht;
    vector<heap_tt_ptr<tdeg_t> > H;
    Ht.reserve(itend-it);
    H.reserve(itend-it);
    unsigned s=0;
    order_t keyorder={_REVLEX_ORDER,0};
    for (unsigned i=0;it!=itend;++i,++it){
      keyorder=it->order;
      if (int(it->coord.size())>start){
	s=giacmax(s,unsigned(it->coord.size()));
	Ht.push_back(heap_tt<tdeg_t>(i,start,it->coord[start].u));
	H.push_back(heap_tt_ptr<tdeg_t>(&Ht.back()));
      }
    }
    allf4buchberger.coord.reserve(s); // int(s*std::log(1+H.size())));
    compare_heap_tt_ptr<tdeg_t> key(keyorder);
    make_heap(H.begin(),H.end(),key);
    while (!H.empty()){
      std::pop_heap(H.begin(),H.end(),key);
      // push root node of the heap in allf4buchberger
      heap_tt<tdeg_t> & current = *H.back().ptr;
      if (allf4buchberger.coord.empty() || allf4buchberger.coord.back().u!=current.u)
	allf4buchberger.coord.push_back(T_unsigned<modint,tdeg_t>(1,current.u));
      ++current.polymodpos;
      if (current.polymodpos>=f4buchbergerv[current.f4buchbergervpos].coord.size()){
	H.pop_back();
	continue;
      }
      current.u=f4buchbergerv[current.f4buchbergervpos].coord[current.polymodpos].u;
      std::push_heap(H.begin(),H.end(),key);
    }
  }

  template<class tdeg_t>
  void collect(const vectpolymod<tdeg_t> & f4buchbergerv,const vector<unsigned> & G,polymod<tdeg_t> & allf4buchberger,unsigned start=0){
    unsigned Gsize=unsigned(G.size());
    if (!Gsize) return;
    vector<heap_tt<tdeg_t> > H;
    compare_heap_tt<tdeg_t> key(f4buchbergerv[G[0]].order);
    H.reserve(Gsize);
    for (unsigned i=0;i<Gsize;++i){
      if (f4buchbergerv[G[i]].coord.size()>start)
	H.push_back(heap_tt<tdeg_t>(i,start,f4buchbergerv[G[i]].coord[start].u));
    }
    make_heap(H.begin(),H.end(),key);
    while (!H.empty()){
      std::pop_heap(H.begin(),H.end(),key);
      // push root node of the heap in allf4buchberger
      heap_tt<tdeg_t> & current =H.back();
      if (allf4buchberger.coord.empty() || allf4buchberger.coord.back().u!=current.u)
	allf4buchberger.coord.push_back(T_unsigned<modint,tdeg_t>(1,current.u));
      ++current.polymodpos;
      if (current.polymodpos>=f4buchbergerv[G[current.f4buchbergervpos]].coord.size()){
	H.pop_back();
	continue;
      }
      current.u=f4buchbergerv[G[current.f4buchbergervpos]].coord[current.polymodpos].u;
      std::push_heap(H.begin(),H.end(),key);
    }
  }

  template<class tdeg_t>
  void leftright(const vectpolymod<tdeg_t> & res,vector< paire > & B,vector<tdeg_t> & leftshift,vector<tdeg_t> & rightshift){
    for (unsigned i=0;i<B.size();++i){
      const polymod<tdeg_t> & p=res[B[i].first];
      const polymod<tdeg_t> & q=res[B[i].second];
      if (debug_infolevel>2)
	CERR << "leftright " << p << "," << q << '\n';
      tdeg_t l(p.coord.front().u);
      index_lcm(p.coord.front().u,q.coord.front().u,l,p.order);
      leftshift[i]=l-p.coord.front().u;
      rightshift[i]=l-q.coord.front().u;
    }
  }

  // collect monomials from pairs of res (vector of polymods), shifted by lcm
  // does not collect leading monomial (since they cancel)
  template<class tdeg_t>
  void collect(const vectpolymod<tdeg_t> & res,vector< paire > & B,polymod<tdeg_t> & allf4buchberger,vector<tdeg_t> & leftshift,vector<tdeg_t> & rightshift){
    int start=1;
    vector<heap_tt<tdeg_t> > Ht;
    vector<heap_tt_ptr<tdeg_t> > H;
    Ht.reserve(2*B.size());
    H.reserve(2*B.size());
    unsigned s=0;
    order_t keyorder={_REVLEX_ORDER,0};
    for (unsigned i=0;i<B.size();++i){
      const polymod<tdeg_t> & p=res[B[i].first];
      const polymod<tdeg_t> & q=res[B[i].second];
      keyorder=p.order;
      if (int(p.coord.size())>start){
	s=giacmax(s,unsigned(p.coord.size()));
	Ht.push_back(heap_tt<tdeg_t>(true,i,start,p.coord[start].u+leftshift[i]));
	H.push_back(heap_tt_ptr<tdeg_t>(&Ht.back()));
      }
      if (int(q.coord.size())>start){
	s=giacmax(s,unsigned(q.coord.size()));
	Ht.push_back(heap_tt<tdeg_t>(false,i,start,q.coord[start].u+rightshift[i]));
	H.push_back(heap_tt_ptr<tdeg_t>(&Ht.back()));
      }
    }
    allf4buchberger.coord.reserve(s); // int(s*std::log(1+H.size())));
    compare_heap_tt_ptr<tdeg_t> key(keyorder);
    make_heap(H.begin(),H.end(),key);
    while (!H.empty()){
      std::pop_heap(H.begin(),H.end(),key);
      // push root node of the heap in allf4buchberger
      heap_tt<tdeg_t> & current = *H.back().ptr;
      if (allf4buchberger.coord.empty() || allf4buchberger.coord.back().u!=current.u)
	allf4buchberger.coord.push_back(T_unsigned<modint,tdeg_t>(1,current.u));
      ++current.polymodpos;
      unsigned vpos;
      if (current.left)
	vpos=B[current.f4buchbergervpos].first;
      else
	vpos=B[current.f4buchbergervpos].second;
      if (current.polymodpos>=res[vpos].coord.size()){
	H.pop_back();
	continue;
      }
      if (current.left)
	current.u=res[vpos].coord[current.polymodpos].u+leftshift[current.f4buchbergervpos];
      else
	current.u=res[vpos].coord[current.polymodpos].u+rightshift[current.f4buchbergervpos];
      std::push_heap(H.begin(),H.end(),key);
    }
  }

  struct sparse_element {
    modint val;
    unsigned pos;
    sparse_element(modint v,size_t u):val(v),pos(unsigned(u)){};
    sparse_element():val(0),pos(-1){};
  };

#ifdef NSPIRE
  template<class T>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os,const sparse_element & s){
    return os << '{' << s.val<<',' << s.pos << '}' ;
  }
#else
  ostream & operator << (ostream & os,const sparse_element & s){
    return os << '{' << s.val<<',' << s.pos << '}' ;
  }
#endif

#ifdef GBASISF4_BUCHBERGER
  bool reducef4buchbergerpos(vector<modint> &v,const vector< vector<modint> > & M,vector<int> pivotpos,modint env){
    unsigned pos=0;
    bool res=false;
    for (unsigned i=0;i<M.size();++i){
      const vector<modint> & m=M[i];
      pos=pivotpos[i];
      if (pos==-1)
	return res;
      modint c=v[pos];
      if (!c)
	continue;
      res=true;
      c=(modint2(invmod(m[pos],env))*c)%env;
      vector<modint>::const_iterator jt=m.begin()+pos+1;
      vector<modint>::iterator it=v.begin()+pos,itend=v.end();
      *it=0; ++it;
      for (;it!=itend;++jt,++it){
	if (*jt)
	  *it=(*it-modint2(c)*(*jt))%env;
      }
    }
    return res;
  }

#ifdef x86_64
  unsigned reducef4buchberger_64(vector<modint> &v,const vector< vector<sparse_element> > & M,modint env,vector<int128_t> & w){
    w.resize(v.size());
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    vector<int128_t>::iterator wt=w.begin();
    for (;vt!=vtend;++wt,++vt){
      *wt=*vt;
    }
    for (unsigned i=0;i<M.size();++i){
      const vector<sparse_element> & m=M[i];
      const sparse_element * it=&m.front(),*itend=it+m.size(),*it2;
      if (it==itend)
	continue;
      int128_t & ww=w[it->pos];
      if (ww==0)
	continue;
      modint c=(modint2(invmod(it->val,env))*ww)%env;
      // CERR << "multiplier ok line " << i << " value " << c << " " << w << '\n';
      if (!c)
	continue;
      ww=0;
      ++it;
      it2=itend-8;
      for (;it<=it2;){
#if 0
	w[it[0].pos] -= modint2(c)*(it[0].val);
	w[it[1].pos] -= modint2(c)*(it[1].val);
	w[it[2].pos] -= modint2(c)*(it[2].val);
	w[it[3].pos] -= modint2(c)*(it[3].val);
	w[it[4].pos] -= modint2(c)*(it[4].val);
	w[it[5].pos] -= modint2(c)*(it[5].val);
	w[it[6].pos] -= modint2(c)*(it[6].val);
	w[it[7].pos] -= modint2(c)*(it[7].val);
	it+=8;
#else
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
#endif
      }
      for (;it!=itend;++it){
	w[it->pos] -= modint2(c)*(it->val);
      }
    }
    for (vt=v.begin(),wt=w.begin();vt!=vtend;++wt,++vt){
      if (*wt)
	*vt=*wt % env;
      else
	*vt=0;
    }
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return vt-v.begin();
    }
    return v.size();
  }

#ifdef NSPIRE
  template<class T>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os,const int128_t & i){
    return os << longlong(i) ;
    // return os << "(" << longlong(i>>64) <<","<< longlong(i) <<")" ;
  }
#else
  ostream & operator << (ostream & os,const int128_t & i){
    return os << longlong(i) ;
    // return os << "(" << longlong(i>>64) <<","<< longlong(i) <<")" ;
  }
#endif

#endif
  // sparse element if prime is < 2^24
  // if shift == 0 the position is absolute in the next sparse32 of the vector
  struct sparse32 {
    modint val:25;
    unsigned shift:7;
    sparse32(modint v,unsigned s):val(v),shift(s){};
    sparse32():val(0),shift(0){};
  };

#ifdef NSPIRE
  template<class T>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os,const sparse32 & s){
    return os << "(" << s.val << "," << s.shift << ")" ;
  }
#else
  ostream & operator << (ostream & os,const sparse32 & s){
    return os << "(" << s.val << "," << s.shift << ")" ;
  }
#endif

  unsigned reducef4buchberger_32(vector<modint> &v,const vector< vector<sparse32> > & M,modint env,vector<modint2> & w){
    w.resize(v.size());
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    vector<modint2>::iterator wt=w.begin();
    for (;vt!=vtend;++wt,++vt){
      *wt=*vt;
    }
    for (unsigned i=0;i<M.size();++i){
      const vector<sparse32> & m=M[i];
      vector<sparse32>::const_iterator it=m.begin(),itend=m.end(),it2=itend-16;
      if (it==itend)
	continue;
      unsigned p=0;
      modint val;
      if (it->shift){
	p += it->shift;
	val=it->val;
      }
      else {
	val=it->val;
	++it;
	p=*(unsigned *)&(*it);
      }
      modint2 & ww=w[p];
      if (ww==0)
	continue;
      modint c=(modint2(invmod(val,env))*ww)%env;
      if (!c)
	continue;
      ww=0;
      ++it;
      for (;it<=it2;){
	sparse32 se = *it;
	unsigned seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
	se = *it;
	seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
	se = *it;
	seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
	se = *it;
	seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
	se = *it;
	seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
	se = *it;
	seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
	se = *it;
	seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
	se = *it;
	seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
	++it;
      }
      for (;it!=itend;++it){
	const sparse32 & se = *it;
	unsigned seshift=se.shift;
	if (seshift){
	  p += seshift;
	  w[p] -= modint2(c)*se.val;
	}
	else {
	  ++it;
	  p = *(unsigned *) &*it;
	  w[p] -= modint2(c)*se.val;
	}
      }
    }
    for (vt=v.begin(),wt=w.begin();vt!=vtend;++wt,++vt){
      if (*wt)
	*vt = *wt % env;
      else
	*vt =0;
    }
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return unsigned(vt-v.begin());
    }
    return unsigned(v.size());
  }

#ifdef PSEUDO_MOD
  // find pseudo remainder of x mod p, 2^nbits>=p>2^(nbits-1)
  // assumes invp=2^(2*nbits)/p+1 has been precomputed
  // and abs(x)<2^(31+nbits)
  // |remainder| <= max(2^nbits,|x|*p/2^(2nbits)), <=2*p if |x|<=p^2
  inline int pseudo_mod(longlong x,int p,unsigned invp,unsigned nbits){
    return int(x - (((x>>nbits)*invp)>>(nbits))*p);
  }
  // a <- (a+b*c) mod or smod p
  inline void pseudo_mod(int & a,int b,int c,int p,unsigned invp,unsigned nbits){
    a=pseudo_mod(a+((longlong)b)*c,p,invp,nbits);
  }
#endif

  unsigned reducef4buchberger(vector<modint> &v,const vector< vector<sparse_element> > & M,modint env){
#ifdef PSEUDO_MOD
    int nbits=sizeinbase2(env);
    unsigned invmodulo=((1ULL<<(2*nbits)))/env+1;
#endif
    for (unsigned i=0;i<M.size();++i){
      const vector<sparse_element> & m=M[i];
      vector<sparse_element>::const_iterator it=m.begin(),itend=m.end();
      if (it==itend)
	continue;
      modint c=(modint2(invmod(it->val,env))*v[it->pos])%env;
      v[it->pos]=0;
      if (!c)
	continue;
#ifdef PSEUDO_MOD
      if (env<(1<<29)){
	c=-c;
	for (++it;it!=itend;++it){
	  pseudo_mod(v[it->pos],c,it->val,env,invmodulo,nbits);
	}
	continue;
      }
#endif
      for (++it;it!=itend;++it){
	modint &x=v[it->pos];
	x=(x-modint2(c)*(it->val))%env;
      }
    }
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
#ifdef PSEUDO_MOD
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	*vt %= env;
    }
#endif
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return unsigned(vt-v.begin());
    }
    return unsigned(v.size());
  }


#if GIAC_SHORTSHIFTTYPE==8
  typedef unsigned char shifttype;
  // assumes that all shifts are less than 2^(3*sizeof()),
  // and almost all shifts are less than 2^sizeof()-1
  // for unsigned char here matrix density should be significantly above 0.004

  inline void next_index(unsigned & pos,const shifttype * & it){
    if (*it)
      pos+=(*it);
    else { // next 3 will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += (*it << 8);
      ++it;
      pos += *it;
    }
    ++it;
  }

  inline void next_index(vector<modint>::iterator & pos,const shifttype * & it){
    if (*it)
      pos+=(*it);
    else { // next 3 will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += (*it << 8);
      ++it;
      pos += *it;
    }
    ++it;
  }

  inline void next_index(vector<modint2>::iterator & pos,const shifttype * & it){
    if (*it)
      pos+=(*it);
    else { // next 3 will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += (*it << 8);
      ++it;
      pos += *it;
    }
    ++it;
  }

  inline void next_index(vector<double>::iterator & pos,const shifttype * & it){
    if (*it)
      pos+=(*it);
    else { // next 3 will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += (*it << 8);
      ++it;
      pos += *it;
    }
    ++it;
  }

#ifdef x86_64
  inline void next_index(vector<int128_t>::iterator & pos,const shifttype * & it){
    if (*it)
      pos+=(*it);
    else { // next 3 will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += (*it << 8);
      ++it;
      pos += *it;
    }
    ++it;
  }
#endif

  unsigned first_index(const vector<shifttype> & v){
    if (v.front())
      return v.front();
    return (v[1]<<16)+(v[2]<<8)+v[3];
  }

  inline void pushsplit(vector<shifttype> & v,unsigned & pos,unsigned newpos){
    unsigned shift=newpos-pos;
    if (shift && (shift <(1<<8)))
      v.push_back(shift);
    else {
      v.push_back(0);
      v.push_back(shift >> 16 );
      v.push_back(shift >> 8);
      v.push_back(shift);
    }
    pos=newpos;
  }
#endif

#if GIAC_SHORTSHIFTTYPE==16
  typedef unsigned short shifttype;

  inline void next_index(unsigned & pos,const shifttype * & it){
    if (*it)
      pos += (*it);
    else { // next will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += *it;
    }
    ++it;
  }

  inline void next_index(vector<modint>::iterator & pos,const shifttype * & it){
    if (*it)
      pos += (*it);
    else { // next will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += *it;
    }
    ++it;
  }

  inline void next_index(vector<modint2>::iterator & pos,const shifttype * & it){
    if (*it)
      pos += (*it);
    else { // next will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += *it;
    }
    ++it;
  }

  inline void next_index(vector<double>::iterator & pos,const shifttype * & it){
    if (*it)
      pos += (*it);
    else { // next will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += *it;
    }
    ++it;
  }

#ifdef x86_64
  inline void next_index(vector<int128_t>::iterator & pos,const shifttype * & it){
    if (*it)
      pos += (*it);
    else { // next will make the shift
      ++it;
      pos += (*it << 16);
      ++it;
      pos += *it;
    }
    ++it;
  }
#endif

  unsigned first_index(const vector<shifttype> & v){
    if (v.front())
      return v.front();
    return (v[1]<<16)+v[2];
  }

  inline void pushsplit(vector<shifttype> & v,unsigned & pos,unsigned newpos){
    unsigned shift=newpos-pos;
    if ( shift && (shift < (1<<16)) )
      v.push_back(shift);
    else {
      v.push_back(0);
      v.push_back(shift >> 16 );
      v.push_back(shift);
    }
    pos=newpos;
  }
#endif

#ifndef GIAC_SHORTSHIFTTYPE
  typedef unsigned shifttype;
  inline void next_index(unsigned & pos,const shifttype * & it){
    pos=(*it);
    ++it;
  }
  inline unsigned first_index(const vector<shifttype> & v){
    return v.front();
  }
  inline void pushsplit(vector<shifttype> & v,unsigned & pos,unsigned newpos){
    v.push_back(pos=newpos);
  }

#endif

  struct coeffindex_t {
    bool b;
    unsigned u:24;
    coeffindex_t(bool b_,unsigned u_):b(b_),u(u_) {};
    coeffindex_t():b(false),u(0) {};
  };

#ifdef x86_64
  unsigned reducef4buchbergersplit64(vector<modint> &v,const vector< vector<shifttype> > & M,const vector<unsigned> & firstpos,vector< vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env,vector<int128_t> & v128){
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    v128.resize(v.size());
    vector<int128_t>::iterator wt=v128.begin(),wt0=wt;
    for (;vt!=vtend;++wt,++vt)
      *wt=*vt;
    vector<unsigned>::const_iterator fit=firstpos.begin(),fit0=fit,fitend=firstpos.end();
    for (;fit!=fitend;++fit){
      if (*(wt0+*fit)==0)
	continue;
      unsigned i=fit-fit0;
      const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
      bool shortshifts=coeffindex[i].b;
      vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end(),jt_=jtend-8;
      if (jt==jtend)
	continue;
      const vector<shifttype> & mindex=M[i];
      const shifttype * it=&mindex.front();
      unsigned pos=0;
      next_index(pos,it);
      wt=wt0+pos;
      // if (*wt==0) continue;
      // if (pos>v.size()) CERR << "error" <<'\n';
      modint c=(invmod(*jt,env)*(*wt))%env;
      *wt=0;
      if (!c)
	continue;
      ++jt;
#ifdef GIAC_SHORTSHIFTTYPE
      if (shortshifts){
#if 0
	if (jt<jt_){
	  while (ulonglong(it)%4){
	    wt += *it; ++it;;
	    *wt -=modint2(c)*(*jt);
	    ++jt;
	  }
	}
#endif
	for (;jt<jt_;){
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  wt += *it; ++it;;
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	}
      } // if (shortshifts)
      else {
	for (;jt<jt_;){
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  *wt -=modint2(c)*(*jt);
	  ++jt;
	}
      }
#else // GIAC_SHORTSHIFTTYPE
      for (;jt<jt_;){
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
      }
#endif
      for (;jt!=jtend;++jt){
#ifdef GIAC_SHORTSHIFTTYPE
	next_index(wt,it);
	*wt -=modint2(c)*(*jt);
#else
	v128[*it]-=modint2(c)*(*jt);
	++it;
#endif
      }
    }
    unsigned res=v.size();
    for (vt=v.begin(),wt=v128.begin();vt!=vtend;++wt,++vt){
      int128_t i=*wt;
      if (i)
	i %= env;
      *vt = i;
      if (i){
	res=vt-v.begin();
	break;
      }
    }
    for (;vt!=vtend;++wt,++vt){
      if (*wt)
	*vt = *wt % env;
      else
	*vt = 0;
    }
    return res;
  }

  unsigned reducef4buchbergersplit64u(vector<modint> &v,const vector< vector<unsigned> > & M,vector< vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env,vector<int128_t> & v128){
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    v128.resize(v.size());
    vector<int128_t>::iterator wt=v128.begin();
    for (;vt!=vtend;++wt,++vt)
      *wt=*vt;
    for (unsigned i=0;i<M.size();++i){
      const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
      vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end(),jt_=jtend-8;
      if (jt==jtend)
	continue;
      const vector<unsigned> & mindex=M[i];
      const unsigned * it=&mindex.front();
      unsigned pos=*it;
      // if (pos>v.size()) CERR << "error" <<'\n';
      modint c=(invmod(*jt,env)*v128[pos])%env;
      v128[pos]=0;
      if (!c)
	continue;
      ++it;++jt;
      for (;jt<jt_;){
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
      }
      for (;jt!=jtend;++jt){
	v128[*it]-=modint2(c)*(*jt);
	++it;
      }
    }
    for (vt=v.begin(),wt=v128.begin();vt!=vtend;++wt,++vt){
      if (*wt)
	*vt = *wt % env;
      else
	*vt=0;
    }
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return vt-v.begin();
    }
    return v.size();
  }

  unsigned reducef4buchbergersplit64s(vector<modint> &v,const vector< vector<short unsigned> > & M,vector< vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env,vector<int128_t> & v128){
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    v128.resize(v.size());
    vector<int128_t>::iterator wt=v128.begin();
    for (;vt!=vtend;++wt,++vt)
      *wt=*vt;
    for (unsigned i=0;i<M.size();++i){
      const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
      vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end(),jt_=jtend-8;
      if (jt==jtend)
	continue;
      const vector<short unsigned> & mindex=M[i];
      const short unsigned * it=&mindex.front();
      unsigned pos=*it;
      // if (pos>v.size()) CERR << "error" <<'\n';
      modint c=(invmod(*jt,env)*v128[pos])%env;
      v128[pos]=0;
      if (!c)
	continue;
      ++it;++jt;
      for (;jt<jt_;){
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
	v128[*it]-=modint2(c)*(*jt);
	++it; ++jt;
      }
      for (;jt!=jtend;++jt){
	v128[*it]-=modint2(c)*(*jt);
	++it;
      }
    }
    for (vt=v.begin(),wt=v128.begin();vt!=vtend;++wt,++vt){
      if (*wt)
	*vt = *wt % env;
      else
	*vt=0;
    }
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return vt-v.begin();
    }
    return v.size();
  }

#endif

  // reducef4buchberger matrix M has band structure, to spare memory
  // we split coeffs/index in :
  // - M each line is a list of shift index,
  // - coeffindex, relative to coeffs, M[i][j] corresponds to coeffs[coeffinde[i]][j]
  // - coeffs is the list of coefficients
  unsigned reducef4buchbergersplit(vector<modint> &v,const vector< vector<shifttype> > & M,const vector<unsigned> & firstpos,vector< vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env,vector<modint2> & v64){
    vector<modint>::iterator vt=v.begin(),vt0=vt,vtend=v.end();
    vector<unsigned>::const_iterator fit=firstpos.begin(),fit0=fit,fitend=firstpos.end();
    if (env<(1<<24)){
      v64.resize(v.size());
      vector<modint2>::iterator wt=v64.begin(),wt0=wt,wtend=v64.end();
      for (;vt!=vtend;++wt,++vt){
	*wt=*vt;
	*vt=0;
      }
      bool fastcheck = (fitend-fit0)<=0xffff;
      for (;fit!=fitend;++fit){
	if (fastcheck && *(wt0+*fit)==0)
	  continue;
	unsigned i=unsigned(fit-fit0);
	if (!fastcheck){
	  if ((i&0xffff)==0xffff){
	    // reduce the line mod env
	    for (wt=v64.begin();wt!=wtend;++wt){
	      if (*wt)
		*wt %= env;
	    }
	  }
	  if (v64[*fit]==0)
	    continue;
	}
	const vector<shifttype> & mindex=M[i];
	const shifttype * it=&mindex.front();
	unsigned pos=0;
	next_index(pos,it);
	wt=wt0+pos;
	// if (*wt==0) continue;
	const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
	bool shortshifts=coeffindex[i].b;
	if (mcoeff.empty())
	  continue;
	const modint * jt=&mcoeff.front(),*jtend=jt+mcoeff.size(),*jt_=jtend-8;
	// if (pos>v.size()) CERR << "error" <<'\n';
	// if (*jt!=1) CERR << "not normalized" << '\n';
	modint c=(modint2(invmod(*jt,env))*(*wt % env))%env;
	*wt=0;
	if (!c)
	  continue;
	++jt;
#ifdef GIAC_SHORTSHIFTTYPE
	if (shortshifts){
	  for (;jt<jt_;){
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	  }
	  for (;jt!=jtend;++jt){
	    wt += *it; ++it;
	    *wt-=modint2(c)*(*jt);
	  }
	}
	else {
	  for (;jt!=jtend;++jt){
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	  }
	}
#else // def GIAC_SHORTSHIFTTYPE
	for (;jt<jt_;){
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	}
	for (;jt!=jtend;++jt){
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	}
#endif // def GIAC_SHORTSHIFTTYPE
      }
      unsigned res=unsigned(v.size());
      for (vt=v.begin(),wt=v64.begin();vt!=vtend;++wt,++vt){
	modint2 i=*wt;
	if (i) // if (i>=env || i<=-env)
	  i %= env;
	*vt = modint(i);
	if (i){
	  res=unsigned(vt-v.begin());
	  break;
	}
      }
      for (;vt!=vtend;++wt,++vt){
	if (modint2 i=*wt) // if (i>=env || i<=-env)
	  *vt = i % env;
      }
      return res;
    }
#ifdef PSEUDO_MOD
    int nbits=sizeinbase2(env);
    unsigned invmodulo=((1ULL<<(2*nbits)))/env+1;
#endif
    for (;fit!=fitend;++fit){
      if (*(vt0+*fit)==0)
	continue;
      unsigned i=unsigned(fit-fit0);
      const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
      bool shortshifts=coeffindex[i].b;
      vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end(),jt_=jt-8;
      if (jt==jtend)
	continue;
      const vector<shifttype> & mindex=M[i];
      const shifttype * it=&mindex.front();
      unsigned pos=0;
      next_index(pos,it);
      vt=v.begin()+pos;
      // if (pos>v.size()) CERR << "error" <<'\n';
      modint c=(modint2(invmod(*jt,env))*(*vt))%env;
      *vt=0;
      if (!c)
	continue;
      ++jt;
#ifdef PSEUDO_MOD
      if (env<(1<<29)){
	c=-c;
#ifdef GIAC_SHORTSHIFTTYPE
	if (shortshifts){
	  for (;jt<jt_;){
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	    ++jt;
	  }
	  for (;jt!=jtend;++jt){
	    vt += *it; ++it;
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	  }
	}
	else {
	  for (;jt<jt_;){
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	    next_index(vt,it); pseudo_mod(*vt,c,*jt,env,invmodulo,nbits); ++jt;
	  }
	  for (;jt!=jtend;++jt){
	    next_index(vt,it);
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(*vt,c,*jt,env,invmodulo,nbits);
	  }
	}
	continue;
#else
	for (;jt!=jtend;++jt){
	  // if (pos>v.size()) CERR << "error" <<'\n';
	  pseudo_mod(v[*it],c,*jt,env,invmodulo,nbits);
	  ++it;
	}
	continue;
#endif // GIAC_SHORTSHIFTTYPES
      } // end if env<1<<29
#endif // PSEUDOMOD
      for (;jt!=jtend;++jt){
#ifdef GIAC_SHORTSHIFTTYPE
	next_index(vt,it);
	*vt = (*vt-modint2(c)*(*jt))%env;
#else
	modint &x=v[*it];
	++it;
	x=(x-modint2(c)*(*jt))%env;
#endif
      }
    }
    vt=v.begin();vtend=v.end();
#ifdef PSEUDO_MOD
    unsigned res=v.size();
    for (;vt!=vtend;++vt){ // or if (*vt) *vt %= env;
      if (!*vt) continue;
      *vt %= env;
      if (*vt){
	res=vt-v.begin();
	break;
      }
    }
    for (;vt!=vtend;++vt){ // or if (*vt) *vt %= env;
      modint v=*vt;
      if (v>-env && v<env)
	continue;
      *vt = v % env;
    }
    return res;
#endif
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return unsigned(vt-v.begin());
    }
    return unsigned(v.size());
  }

  inline void special_mod(modint2 & x,modint2 c,modint d,modint env,modint2 env2){
    register modint2 y=x-c*d;
    //x=y%env;
    //if (y<0) x = y+env2; else x=y;// if y is negative make it positive by adding env^2
    x = y - (y>>63)*env2;
  }

  inline void special_mod(double & x,double c,modint d,modint env,double env2){
    register modint2 y=modint2(x-c*d);
    if (y<0) x = double(y+env2); else x=double(y);// if y is negative make it positive by adding env^2
  }

  typedef char used_t;
  // typedef bool used_t;

  void f4_innerloop_(modint2 * wt,const modint * jt,const modint * jtend,modint c,const shifttype* it){
    jtend -= 8;
    for (;jt<=jtend;){
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
      ++jt;
    }
    jtend+=8;
    for (;jt!=jtend;++jt){
      wt += *it; ++it;
      *wt-=modint2(c)*(*jt);
    }
  }

  inline void f4_innerloop(modint2 * wt,const modint * jt,const modint * jtend,modint C,const shifttype* it){
    jtend -= 16;
    for (;jt<=jtend;){
#if 1
      wt += it[0]; int b=it[1];
      *wt -= modint2(C)*jt[0];
      wt[b] -= modint2(C)*jt[1];
      wt += b+it[2]; b=it[3];
      *wt -= modint2(C)*jt[2];
      wt[b] -= modint2(C)*jt[3];
      wt += b+it[4]; b=it[5];
      *wt -= modint2(C)*jt[4];
      wt[b] -= modint2(C)*jt[5];
      wt += b+it[6]; b=it[7];
      *wt -= modint2(C)*jt[6];
      wt[b] -= modint2(C)*jt[7];
      wt += b+it[8]; b=it[9];
      *wt -= modint2(C)*jt[8];
      wt[b] -= modint2(C)*jt[9];
      wt += b+it[10]; b=it[11];
      *wt -= modint2(C)*jt[10];
      wt[b] -= modint2(C)*jt[11];
      wt += b+it[12]; b=it[13];
      *wt -= modint2(C)*jt[12];
      wt[b] -= modint2(C)*jt[13];
      wt += b+it[14]; b=it[15];
      *wt -= modint2(C)*jt[14];
      wt[b] -= modint2(C)*jt[15];
      wt += b;
      it += 16; jt+=16;
#else
      wt += it[0]; *wt -= modint2(C)*jt[0];
      wt += it[1]; *wt -= modint2(C)*jt[1];
      wt += it[2]; *wt -= modint2(C)*jt[2];
      wt += it[3]; *wt -= modint2(C)*jt[3];
      wt += it[4]; *wt -= modint2(C)*jt[4];
      wt += it[5]; *wt -= modint2(C)*jt[5];
      wt += it[6]; *wt -= modint2(C)*jt[6];
      wt += it[7]; *wt -= modint2(C)*jt[7];
      wt += it[8]; *wt -= modint2(C)*jt[8];
      wt += it[9]; *wt -= modint2(C)*jt[9];
      wt += it[10]; *wt -= modint2(C)*jt[10];
      wt += it[11]; *wt -= modint2(C)*jt[11];
      wt += it[12]; *wt -= modint2(C)*jt[12];
      wt += it[13]; *wt -= modint2(C)*jt[13];
      wt += it[14]; *wt -= modint2(C)*jt[14];
      wt += it[15]; *wt -= modint2(C)*jt[15];
      it += 16; jt+=16;
#endif
    }
    jtend += 16;
    for (;jt!=jtend;++jt){
      wt += *it; ++it;
      *wt-=modint2(C)*(*jt);
    }
  }

  void f4_innerloop_special_mod(modint2 * wt,const modint * jt,const modint * jtend,modint C,const shifttype* it,modint env){
    modint2 env2=modint2(env)*env;
    jtend -= 16;
    for (;jt<=jtend;){
      wt += it[0]; int b=it[1];
      special_mod(*wt,C,*jt,env,env2);
      special_mod(wt[b],C,jt[1],env,env2);
      wt += b+it[2]; b=it[3];
      special_mod(*wt,C,jt[2],env,env2);
      special_mod(wt[b],C,jt[3],env,env2);
      wt += b+it[4]; b=it[5];
      special_mod(*wt,C,jt[4],env,env2);
      special_mod(wt[b],C,jt[5],env,env2);
      wt += b+it[6]; b=it[7];
      special_mod(*wt,C,jt[6],env,env2);
      special_mod(wt[b],C,jt[7],env,env2);
      wt += b+it[8]; b=it[9];
      special_mod(*wt,C,jt[8],env,env2);
      special_mod(wt[b],C,jt[9],env,env2);
      wt += b+it[10]; b=it[11];
      special_mod(*wt,C,jt[10],env,env2);
      special_mod(wt[b],C,jt[11],env,env2);
      wt += b+it[12]; b=it[13];
      special_mod(*wt,C,jt[12],env,env2);
      special_mod(wt[b],C,jt[13],env,env2);
      wt += b+it[14]; b=it[15];
      special_mod(*wt,C,jt[14],env,env2);
      special_mod(wt[b],C,jt[15],env,env2);
      wt += b;
      it += 16; jt+=16;
    }
    jtend += 16;
    for (;jt!=jtend;++jt){
      wt += *it; ++it;
      special_mod(*wt,C,*jt,env,env2);
    }
  }

  void f4_innerloop_special_mod(double * wt,const modint * jt,const modint * jtend,modint C,const shifttype* it,modint env){
    double env2=double(env)*env;
    jtend -= 16;
    for (;jt<=jtend;){
      wt += it[0]; int b=it[1];
      special_mod(*wt,C,*jt,env,env2);
      special_mod(wt[b],C,jt[1],env,env2);
      wt += b+it[2]; b=it[3];
      special_mod(*wt,C,jt[2],env,env2);
      special_mod(wt[b],C,jt[3],env,env2);
      wt += b+it[4]; b=it[5];
      special_mod(*wt,C,jt[4],env,env2);
      special_mod(wt[b],C,jt[5],env,env2);
      wt += b+it[6]; b=it[7];
      special_mod(*wt,C,jt[6],env,env2);
      special_mod(wt[b],C,jt[7],env,env2);
      wt += b+it[8]; b=it[9];
      special_mod(*wt,C,jt[8],env,env2);
      special_mod(wt[b],C,jt[9],env,env2);
      wt += b+it[10]; b=it[11];
      special_mod(*wt,C,jt[10],env,env2);
      special_mod(wt[b],C,jt[11],env,env2);
      wt += b+it[12]; b=it[13];
      special_mod(*wt,C,jt[12],env,env2);
      special_mod(wt[b],C,jt[13],env,env2);
      wt += b+it[14]; b=it[15];
      special_mod(*wt,C,jt[14],env,env2);
      special_mod(wt[b],C,jt[15],env,env2);
      wt += b;
      it += 16; jt+=16;
    }
    jtend += 16;
    for (;jt!=jtend;++jt){
      wt += *it; ++it;
      special_mod(*wt,C,*jt,env,env2);
    }
  }

  unsigned store_coeffs(vector<modint2> &v64,unsigned firstcol,vector<modint> & lescoeffs,unsigned * bitmap,vector<used_t> & used,modint env){
    unsigned res=0;
    used_t * uit=&used.front();
    vector<modint2>::iterator wt0=v64.begin(),wt=v64.begin()+firstcol,wtend=v64.end();
    vector<modint2>::iterator wt1=wtend-4;
#if 1
    for (;wt<=wt1;wt+=4){
      if (wt[0] | wt[1] | wt[2] | wt[3])
	break;
    }
#endif
    if (!res){
      for (;wt<wtend;++wt){
	modint2 i=*wt;
	if (!i) continue;
	*wt = 0;
	i %= env;
	if (!i) continue;
	unsigned I=unsigned(wt-wt0);
	res=I;
	*(uit+I)=1; // used[i]=1;
	bitmap[I>>5] |= (1<<(I&0x1f));
	lescoeffs.push_back(modint(i));
	break;
      }
      if (!res)
	res=unsigned(v64.size());
    }
#if 1
    for (;wt<=wt1;){
      modint2 i=*wt;
      if (!i){
	if (!wt[1] && !wt[2] && !wt[3]){
	  wt += 4;
	  continue;
	}
	++wt; i=*wt;
	if (!i){
	  ++wt; i=*wt;
	  if (!i){
	    ++wt; i=*wt;
	  }
	}
      }
      *wt = 0;
      i %= env;
      if (!i){
	wt++; continue;
      }
      unsigned I=unsigned(wt-wt0);
      *(uit+I)=1; // used[i]=1;
      bitmap[I>>5] |= (1<<(I&0x1f));
      lescoeffs.push_back(modint(i));
      wt++;
    }
#endif
    for (;wt<wtend;++wt){
      modint2 i=*wt;
      if (!i) continue;
      *wt=0;
      i %= env;
      if (!i) continue;
      unsigned I=unsigned(wt-wt0);
      *(uit+I)=1; // used[i]=1;
      bitmap[I>>5] |= (1<<(I&0x1f));
      lescoeffs.push_back(modint(i));
    }
    return res;
  }

  unsigned reducef4buchbergersplit(vector<modint2> &v64,const vector< vector<shifttype> > & M,const vector<unsigned> & firstpos,unsigned firstcol,const vector< vector<modint> > & coeffs,const vector<coeffindex_t> & coeffindex,vector<modint> & lescoeffs,unsigned * bitmap,vector<used_t> & used,modint env){
    vector<unsigned>::const_iterator fit=firstpos.begin(),fit0=fit,fitend=firstpos.end(),fit1=fit+firstcol,fit2;
    if (fit1>fitend)
      fit1=fitend;
    vector<modint2>::iterator wt=v64.begin(),wt0=wt,wt1,wtend=v64.end();
    unsigned skip=0;
    while (fit+1<fit1){
      fit2=fit+(fit1-fit)/2;
      if (*fit2>firstcol)
	fit1=fit2;
      else
	fit=fit2;
    }
    if (debug_infolevel>2)
      CERR << "Firstcol " << firstcol << "/" << v64.size() << " ratio skipped " << (fit-fit0)/double(fitend-fit0) << '\n';
    if (env<(1<<24)){
#ifdef PSEUDO_MOD
      int nbits=sizeinbase2(env);
      unsigned invmodulo=((1ULL<<(2*nbits)))/env+1;
#endif
      bool fastcheck = (fitend-fit)<32768;
      unsigned redno=0;
      for (;fit<fitend;++fit){
	if (*(wt0+*fit)==0){
	  //if (fit<fit4 && *(wt0+fit[1])==0 && *(wt0+fit[2])==0 && *(wt0+fit[3])==0 ) fit += 3;
	  continue;
	}
	unsigned i=unsigned(fit-fit0);
	const vector<shifttype> & mindex=M[i];
	const shifttype * it=&mindex.front();
	unsigned pos=0;
	next_index(pos,it);
	skip=pos;
	wt=wt0+pos;
	//if (*wt==0) continue; // test already done with v64[*fit]==0
	const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
	bool shortshifts=coeffindex[i].b;
	if (mcoeff.empty())
	  continue;
	const modint * jt=&mcoeff.front(),*jtend=jt+mcoeff.size(),*jt_=jtend-8;
	// if (pos>v.size()) CERR << "error" <<'\n';
	// if (*jt!=1) CERR << "not normalized " << i << '\n';
#if 0 // def PSEUDO_MOD, does not work for cyclic8m
	modint c=pseudo_mod(*wt,env,invmodulo,nbits); // *jt should be 1
#else
	modint c=*wt % env; // (modint2(*jt)*(*wt % env))%env;
#endif
	*wt=0;
	if (!c)
	  continue;
	if (!fastcheck){
	  ++redno;
	  if (redno==32768){
	    redno=0;
	    // reduce the line mod env
	    //CERR << "reduce line" << '\n';
	    for (vector<modint2>::iterator wt=v64.begin()+pos;wt!=wtend;++wt){
	      // 2^63-1-p*p*32768 where p:=prevprime(2^24)
	      modint2 tmp=*wt;
	      if (tmp>=3298534588415LL || tmp<=-3298534588415LL)
		*wt = tmp % env;
	      // if (*wt) *wt %= env; // does not work pseudo_mod(*wt,env,invmodulo,nbits);
	    }
	  }
	}
	++jt;
#ifdef GIAC_SHORTSHIFTTYPE
	if (shortshifts){
	  f4_innerloop(&*wt,jt,jtend,c,it);
	}
	else {
	  for (;jt<jt_;){
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	    ++jt;
	  }
	  for (;jt!=jtend;++jt){
	    next_index(wt,it);
	    *wt-=modint2(c)*(*jt);
	  }
	}
#else // def GIAC_SHORTSHIFTTYPE
	for (;jt<jt_;){
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	}
	for (;jt!=jtend;++jt){
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	}
#endif // def GIAC_SHORTSHIFTTYPE
      }
    }
    else { // env > 2^24
      modint2 env2=modint2(env)*env;
      for (;fit!=fitend;++fit){
	if (v64[*fit]==0)
	  continue;
	unsigned i=unsigned(fit-fit0);
	const vector<shifttype> & mindex=M[i];
	const shifttype * it=&mindex.front();
	unsigned pos=0;
	next_index(pos,it);
	skip=pos;
	wt=wt0+pos;
	// if (*wt==0) continue;
	const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
	bool shortshifts=coeffindex[i].b;
	if (mcoeff.empty())
	  continue;
	const modint * jt=&mcoeff.front(),*jtend=jt+mcoeff.size(),*jt_=jtend-8;
	// if (pos>v.size()) CERR << "error" <<'\n';
	// if (*jt!=1) CERR << "not normalized" << '\n';
	modint c=*wt % env; // (modint2(*jt)*(*wt % env))%env;
	if (c<0) c += env;
	*wt=0;
	if (!c)
	  continue;
	++jt;
#ifdef GIAC_SHORTSHIFTTYPE
	if (shortshifts){
	  f4_innerloop_special_mod(&*wt,jt,jtend,c,it,env);
	}
	else {
	  for (;jt<jt_;){
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2); // 	    *wt-=modint2(c)*(*jt);
	    ++jt;
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	    ++jt;
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	    ++jt;
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	    ++jt;
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	    ++jt;
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	    ++jt;
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	    ++jt;
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	    ++jt;
	  }
	  for (;jt!=jtend;++jt){
	    next_index(wt,it);
	    special_mod(*wt,c,*jt,env,env2);
	  }
	}
#else // def GIAC_SHORTSHIFTTYPE
	for (;jt<jt_;){
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	}
	for (;jt!=jtend;++jt){
	  special_mod(v64[*it],c,*jt,env,env2);
	  ++it; ++jt;
	}
#endif // def GIAC_SHORTSHIFTTYPE
      }
    }
    if (!bitmap)
      return 0; // result in v64, for multiple uses
    return store_coeffs(v64,firstcol,lescoeffs,bitmap,used,env);
  }

  unsigned reducef4buchbergersplitdouble(vector<double> &v64,const vector< vector<shifttype> > & M,const vector<unsigned> & firstpos,unsigned firstcol,const vector< vector<modint> > & coeffs,const vector<coeffindex_t> & coeffindex,vector<modint> & lescoeffs,unsigned * bitmap,vector<used_t> & used,modint env){
    vector<unsigned>::const_iterator fit=firstpos.begin(),fit0=fit,fitend=firstpos.end(),fit1=fit+firstcol,fit2;
    if (fit1>fitend)
      fit1=fitend;
    vector<double>::iterator wt=v64.begin(),wt0=wt,wt1,wtend=v64.end();
    unsigned skip=0;
    while (fit+1<fit1){
      fit2=fit+(fit1-fit)/2;
      if (*fit2>firstcol)
	fit1=fit2;
      else
	fit=fit2;
    }
    if (debug_infolevel>2)
      CERR << "Firstcol " << firstcol << "/" << v64.size() << " ratio skipped " << (fit-fit0)/double(fitend-fit0) << '\n';
    double env2=double(env)*env;
    for (;fit!=fitend;++fit){
      if (v64[*fit]==0)
	continue;
      unsigned i=unsigned(fit-fit0);
      const vector<shifttype> & mindex=M[i];
      const shifttype * it=&mindex.front();
      unsigned pos=0;
      next_index(pos,it);
      skip=pos;
      wt=wt0+pos;
      // if (*wt==0) continue;
      const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
      bool shortshifts=coeffindex[i].b;
      if (mcoeff.empty())
	continue;
      const modint * jt=&mcoeff.front(),*jtend=jt+mcoeff.size(),*jt_=jtend-8;
      // if (pos>v.size()) CERR << "error" <<'\n';
      // if (*jt!=1) CERR << "not normalized" << '\n';
      modint c=modint2(*wt) % env; // (modint2(*jt)*(*wt % env))%env;
      if (c<0) c += env;
      *wt=0;
      if (!c)
	continue;
      ++jt;
#ifdef GIAC_SHORTSHIFTTYPE
      if (shortshifts){
	f4_innerloop_special_mod(&*wt,jt,jtend,c,it,env);
      }
      else {
	for (;jt<jt_;){
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2); // 	    *wt-=modint2(c)*(*jt);
	  ++jt;
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	  ++jt;
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	  ++jt;
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	  ++jt;
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	  ++jt;
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	  ++jt;
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	  ++jt;
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	  ++jt;
	}
	for (;jt!=jtend;++jt){
	  next_index(wt,it);
	  special_mod(*wt,c,*jt,env,env2);
	}
      }
#else // def GIAC_SHORTSHIFTTYPE
      for (;jt<jt_;){
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
      }
      for (;jt!=jtend;++jt){
	special_mod(v64[*it],c,*jt,env,env2);
	++it; ++jt;
      }
#endif // def GIAC_SHORTSHIFTTYPE
    }
    if (!bitmap)
      return 0; // result in v64, for multiple uses
    unsigned res=0;
    used_t * uit=&used.front();
    wt=v64.begin()+firstcol;
    wt1=wtend-4;
#if 1
    for (;wt<=wt1;++wt){
      double i=*wt;
      if (i) break;
      ++wt; i=*wt;
      if (i) break;
      ++wt; i=*wt;
      if (i) break;
      ++wt; i=*wt;
      if (i) break;
    }
#endif
    if (!res){
      for (;wt<wtend;++wt){
	modint2 i=modint2(*wt);
	if (!i) continue;
	*wt = 0;
	i %= env;
	if (!i) continue;
	unsigned I=unsigned(wt-wt0);
	res=I;
	*(uit+I)=1; // used[i]=1;
	bitmap[I>>5] |= (1<<(I&0x1f));
	lescoeffs.push_back(modint(i));
	break;
      }
      if (!res)
	res=unsigned(v64.size());
    }
#if 1
    for (;wt<=wt1;++wt){
      modint2 i=modint2(*wt);
      if (!i){
	++wt; i=modint2(*wt);
	if (!i){
	  ++wt; i=modint2(*wt);
	  if (!i){
	    ++wt; i=modint2(*wt);
	    if (!i)
	      continue;
	  }
	}
      }
      *wt = 0;
      i %= env;
      if (!i) continue;
      unsigned I=unsigned(wt-wt0);
      *(uit+I)=1; // used[i]=1;
      bitmap[I>>5] |= (1<<(I&0x1f));
      lescoeffs.push_back(modint(i));
    }
#endif
    for (;wt<wtend;++wt){
      modint2 i=modint2(*wt);
      if (!i) continue;
      *wt=0;
      i %= env;
      if (!i) continue;
      unsigned I=unsigned(wt-wt0);
      *(uit+I)=1; // used[i]=1;
      bitmap[I>>5] |= (1<<(I&0x1f));
      lescoeffs.push_back(modint(i));
    }
    return res;
  }

  unsigned reducef4buchbergersplitu(vector<modint> &v,const vector< vector<unsigned> > & M,vector< vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env,vector<modint2> & v64){
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    if (env<(1<<24)){
      v64.resize(v.size());
      vector<modint2>::iterator wt=v64.begin(),wtend=v64.end();
      for (;vt!=vtend;++wt,++vt)
	*wt=*vt;
      for (unsigned i=0;i<M.size();++i){
	if ((i&0xffff)==0xffff){
	  // reduce the line mod env
	  for (wt=v64.begin();wt!=wtend;++wt){
	    if (*wt)
	      *wt %= env;
	  }
	}
	const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
	if (mcoeff.empty())
	  continue;
	const modint * jt=&mcoeff.front(),*jtend=jt+mcoeff.size(),*jt_=jtend-8;
	const vector<unsigned> & mindex=M[i];
	const unsigned * it=&mindex.front();
	unsigned pos=*it;
	// if (pos>v.size()) CERR << "error" <<'\n';
	// if (*jt!=1) CERR << "not normalized" << '\n';
	modint c=(modint2(invmod(*jt,env))*(v64[pos] % env))%env;
	v64[pos]=0;
	if (!c)
	  continue;
	++it; ++jt;
	for (;jt<jt_;){
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	}
	for (;jt!=jtend;++it,++jt){
	  v64[*it]-=modint2(c)*(*jt);
	}
      }
      for (vt=v.begin(),wt=v64.begin();vt!=vtend;++wt,++vt){
	if (*wt)
	  *vt = *wt % env;
	else
	  *vt=0;
      }
    }
    else { // large modulo
#ifdef PSEUDO_MOD
      int nbits=sizeinbase2(env);
      unsigned invmodulo=((1ULL<<(2*nbits)))/env+1;
#endif
      for (unsigned i=0;i<M.size();++i){
	const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
	vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end();
	if (jt==jtend)
	  continue;
	const vector<unsigned> & mindex=M[i];
	const unsigned * it=&mindex.front();
	unsigned pos=*it;
	// if (pos>v.size()) CERR << "error" <<'\n';
	modint c=(modint2(invmod(*jt,env))*v[pos])%env;
	v[pos]=0;
	if (!c)
	  continue;
	++it; ++jt;
#ifdef PSEUDO_MOD
	if (env<(1<<29)){
	  c=-c;
	  for (;jt!=jtend;++jt){
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(v[*it],c,*jt,env,invmodulo,nbits);
	    ++it;
	  }
	  continue;
	}
#endif
	for (;jt!=jtend;++jt){
	  modint &x=v[*it];
	  ++it;
	  x=(x-modint2(c)*(*jt))%env;
	}
      }
      vector<modint>::iterator vt=v.begin(),vtend=v.end();
#ifdef PSEUDO_MOD
      for (vt=v.begin();vt!=vtend;++vt){
	if (*vt)
	  *vt %= env;
      }
#endif
    } // end else based on modulo size
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return unsigned(vt-v.begin());
    }
    return unsigned(v.size());
  }

  unsigned reducef4buchbergersplits(vector<modint> &v,const vector< vector<unsigned short> > & M,vector< vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env,vector<modint2> & v64){
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    if (env<(1<<24)){
      v64.resize(v.size());
      vector<modint2>::iterator wt=v64.begin(),wtend=v64.end();
      for (;vt!=vtend;++wt,++vt)
	*wt=*vt;
      for (unsigned i=0;i<M.size();++i){
	if ((i&0xffff)==0xffff){
	  // reduce the line mod env
	  for (wt=v64.begin();wt!=wtend;++wt){
	    if (*wt)
	      *wt %= env;
	  }
	}
	const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
	if (mcoeff.empty())
	  continue;
	const modint * jt=&mcoeff.front(),*jtend=jt+mcoeff.size(),*jt_=jtend-8;
	const vector<unsigned short> & mindex=M[i];
	const unsigned short * it=&mindex.front();
	unsigned pos=*it;
	// if (pos>v.size()) CERR << "error" <<'\n';
	// if (*jt!=1) CERR << "not normalized" << '\n';
	modint c=(modint2(invmod(*jt,env))*(v64[pos] % env))%env;
	v64[pos]=0;
	if (!c)
	  continue;
	++it; ++jt;
	for (;jt<jt_;){
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	  v64[*it]-=modint2(c)*(*jt);
	  ++it; ++jt;
	}
	for (;jt!=jtend;++it,++jt){
	  v64[*it]-=modint2(c)*(*jt);
	}
      }
      for (vt=v.begin(),wt=v64.begin();vt!=vtend;++wt,++vt){
	if (*wt)
	  *vt = *wt % env;
	else
	  *vt=0;
      }
    }
    else { // large modulo
#ifdef PSEUDO_MOD
      int nbits=sizeinbase2(env);
      unsigned invmodulo=((1ULL<<(2*nbits)))/env+1;
#endif
      for (unsigned i=0;i<M.size();++i){
	const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
	vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end();
	if (jt==jtend)
	  continue;
	const vector<unsigned short> & mindex=M[i];
	const unsigned short * it=&mindex.front();
	unsigned pos=*it;
	// if (pos>v.size()) CERR << "error" <<'\n';
	modint c=(modint2(invmod(*jt,env))*v[pos])%env;
	v[pos]=0;
	if (!c)
	  continue;
	++it; ++jt;
#ifdef PSEUDO_MOD
	if (env<(1<<29)){
	  c=-c;
	  for (;jt!=jtend;++jt){
	    // if (pos>v.size()) CERR << "error" <<'\n';
	    pseudo_mod(v[*it],c,*jt,env,invmodulo,nbits);
	    ++it;
	  }
	  continue;
	}
#endif
	for (;jt!=jtend;++jt){
	  modint &x=v[*it];
	  ++it;
	  x=(x-modint2(c)*(*jt))%env;
	}
      }
      vector<modint>::iterator vt=v.begin(),vtend=v.end();
#ifdef PSEUDO_MOD
      for (vt=v.begin();vt!=vtend;++vt){
	if (*vt)
	  *vt %= env;
      }
#endif
    } // end else based on modulo size
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return unsigned(vt-v.begin());
    }
    return unsigned(v.size());
  }

  bool tri(const vector<sparse_element> & v1,const vector<sparse_element> & v2){
    return v1.front().pos<v2.front().pos;
  }

  struct sparse_element_tri1 {
    sparse_element_tri1(){}
    bool operator() (const sparse_element & v1,const sparse_element & v2){
      return v1.val<v2.val;
    }
  };

  void sort_vector_sparse_element(vector<sparse_element>::iterator it,vector<sparse_element>::iterator itend){
    sort(it,itend,sparse_element_tri1());
  }

  // if sorting with presumed size, adding reconstructed generators will
  // not work...
  template <class poly>
  struct tripolymod_tri {
    int sort_by_logz_age;
    tripolymod_tri(int b):sort_by_logz_age(b){}
    bool operator() (const poly & v1,const poly & v2){
      if (sort_by_logz_age==1 && v1.logz!=v2.logz)
	return v1.logz<v2.logz;
      if (sort_by_logz_age==2 && v1.age!=v2.age)
	return v1.age<v2.age;
      return tdeg_t_strictly_greater(v2.coord.front().u,v1.coord.front().u,v1.order);
    }
  };

  template<class tdeg_t>
  void makeline(const polymod<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<modint> & v,int start=0){
    v.resize(R.coord.size());
    v.assign(R.coord.size(),0);
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin()+start,itend=p.coord.end(),jt=R.coord.begin(),jtbeg=jt,jtend=R.coord.end();
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v[jt-jtbeg]=it->g;
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v[jt-jtbeg]=it->g;
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  template<class tdeg_t>
  void makelinesub(const polymod<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<modint> & v,int start,modint env){
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin()+start,itend=p.coord.end(),jt=R.coord.begin(),jtbeg=jt,jtend=R.coord.end();
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    // v[jt-jtbeg] -= it->g;
	    modint & vv=v[jt-jtbeg];
	    vv = (vv-longlong(it->g))%env;
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    // v[jt-jtbeg]-=it->g;
	    modint & vv=v[jt-jtbeg];
	    vv = (vv-longlong(it->g))%env;
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  // put in v coeffs of polymod corresponding to R, and in rem those who do not match
  // returns false if v is null
  template<class tdeg_t>
  bool makelinerem(const polymod<tdeg_t> & p,polymod<tdeg_t> & rem,const polymod<tdeg_t> & R,vector<modint> & v){
    rem.coord.clear();
    v.clear();
    v.resize(R.coord.size());
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=R.coord.begin(),jtend=R.coord.end();
    bool res=false;
    for (;it!=itend;++it){
      const tdeg_t & u=it->u;
      for (;jt!=jtend;++jt){
	if (tdeg_t_greater(u,jt->u,p.order)
	    // u>=jt->u
	    ){
	  if (u==jt->u){
	    res=true;
	    v[jt-R.coord.begin()]=it->g;
	    ++jt;
	  }
	  else
	    rem.coord.push_back(*it);
	  break;
	}
      }
    }
    return res;
  }

  template<class tdeg_t>
  void makeline(const polymod<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<sparse_element> & v){
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=R.coord.begin(),jtend=R.coord.end();
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v.push_back(sparse_element(it->g,jt-R.coord.begin()));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v.push_back(sparse_element(it->g,jt-R.coord.begin()));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }


#if 1
  void convert(const vector<modint> & v,vector<sparse_element> & w,vector<used_t> & used){
    unsigned count=0;
    vector<modint>::const_iterator it=v.begin(),itend=v.end();
    vector<used_t>::iterator ut=used.begin();
    for (;it!=itend;++ut,++it){
      if (!*it)
	continue;
      *ut=1;
      ++count;
    }
    w.clear();
    w.reserve(count);
    for (count=0,it=v.begin();it!=itend;++count,++it){
      if (*it)
	w.push_back(sparse_element(*it,count));
    }
  }

#else
  void convert(const vector<modint> & v,vector<sparse_element> & w,vector<used_t> & used){
    unsigned count=0;
    vector<modint>::const_iterator it=v.begin(),itend=v.end();
    for (;it!=itend;++it){
      if (*it){
#if 1
	used[it-v.begin()]=1;
#else
	++used[it-v.begin()];
	if (used[it-v.begin()]>100)
	  used[it-v.begin()]=100;
#endif
	++count;
      }
    }
    w.clear();
    w.reserve(count);
    for (it=v.begin();it!=itend;++it){
      if (*it)
	w.push_back(sparse_element(*it,it-v.begin()));
    }
  }
#endif

  // add to w non-zero coeffs of v, set bit i in bitmap to 1 if v[i]!=0
  // bitmap size is rounded to a multiple of 32
  void zconvert(const vector<modint> & v,vector<modint>::iterator & coeffit,unsigned * bitmap,vector<used_t> & used){
    vector<modint>::const_iterator it=v.begin(),itend=v.end();
    used_t * uit=&used.front();
    for (unsigned i=0;it!=itend;++i,++it){
      if (!*it)
	continue;
      *(uit+i)=1; // used[i]=1;
      bitmap[i>>5] |= (1<<(i&0x1f));
      *coeffit=*it;
      ++coeffit;
    }
  }

  template<class modint_t>
  void zconvert_(vector<modint_t> & v,vector<modint> & lescoeffs,unsigned * bitmap,vector<used_t> & used){
    typename vector<modint_t>::iterator it0=v.begin(),it=it0,itend=v.end(),itend4=itend-4;
    used_t * uit=&used.front();
    for (;it<=itend4;++it){
      if (!*it ){
	++it;
	if (!*it){
	  ++it;
	  if (!*it){
	    ++it;
	    if (!*it)
	      continue;
	  }
	}
      }
      unsigned i=unsigned(it-it0);
      *(uit+i)=1; // used[i]=1;
      bitmap[i>>5] |= (1<<(i&0x1f));
      lescoeffs.push_back(*it);
      *it=0;
    }
    for (;it!=itend;++it){
      if (!*it)
	continue;
      unsigned i=unsigned(it-it0);
      *(uit+i)=1; // used[i]=1;
      bitmap[i>>5] |= (1<<(i&0x1f));
      lescoeffs.push_back(*it);
      *it=0;
    }
  }

  // create matrix from list of coefficients and bitmap of non-zero positions
  // M must already have been created with the right number of rows
  void create_matrix(const vector<modint> & lescoeffs,const unsigned * bitmap,unsigned bitmapcols,const vector<used_t> & used,vector< vector<modint> > & M){
    unsigned nrows=unsigned(M.size());
    int ncols=0;
    vector<used_t>::const_iterator ut=used.begin(),utend=used.end();
    unsigned jend=unsigned(utend-ut);
    vector<modint>::const_iterator it=lescoeffs.begin();
    for (;ut!=utend;++ut){
      ncols += *ut;
    }
    // do all memory allocation at once, trying to speed up threaded execution
    for (unsigned i=0;i<nrows;++i)
      M[i].resize(ncols);
    for (unsigned i=0;i<nrows;++i){
      const unsigned * bitmapi = bitmap + i*bitmapcols;
      vector<modint>::iterator mi=M[i].begin();
      unsigned j=0;
      for (;j<jend;++j){
	if (!used[j])
	  continue;
	if (bitmapi[j>>5] & (1<<(j&0x1f))){
	  *mi=*it;
	  ++it;
	}
	++mi;
      }
    }
  }

  void create_matrix(const unsigned * bitmap,unsigned bitmapcols,const vector<used_t> & used,vector< vector<modint> > & M){
    unsigned nrows=unsigned(M.size()),zeros=0;
    int ncols=0;
    vector<used_t>::const_iterator ut=used.begin(),utend=used.end();
    unsigned jend=unsigned(utend-ut);
    for (;ut!=utend;++ut){
      ncols += *ut;
    }
    vector<modint> tmp;
    for (unsigned i=0;i<nrows;++i){
      if (M[i].empty()){ ++zeros; continue; }
      const unsigned * bitmapi = bitmap + i*bitmapcols;
      tmp.clear();
      tmp.resize(ncols);
      tmp.swap(M[i]);
      vector<modint>::iterator mi=M[i].begin(),it=tmp.begin();
      unsigned j=0;
      for (;j<jend;++j){
	if (!used[j])
	  continue;
	if (bitmapi[j>>5] & (1<<(j&0x1f))){
	  *mi=*it;
	  ++it;
	}
	++mi;
      }
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " " << zeros << " null lines over " << M.size() << '\n';
  }

  inline void push32(vector<sparse32> & v,modint val,unsigned & pos,unsigned newpos){
    unsigned shift=newpos-pos;
    if (newpos && (shift <(1<<7)))
      v.push_back(sparse32(val,shift));
    else {
      v.push_back(sparse32(val,0));
      v.push_back(sparse32());
      * (unsigned *) & v.back() =newpos;
    }
    pos=newpos;
  }

  template <class tdeg_t>
  void makeline32(const polymod<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<sparse32> & v){
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=R.coord.begin(),jtend=R.coord.end();
    unsigned pos=0;
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    push32(v,it->g,pos,unsigned(jt-R.coord.begin()));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    push32(v,it->g,pos,unsigned(jt-R.coord.begin()));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  void convert32(const vector<modint> & v,vector<sparse32> & w,vector<used_t> & used){
    unsigned count=0;
    vector<modint>::const_iterator it=v.begin(),itend=v.end();
    for (;it!=itend;++it){
      if (*it){
#if 1
	used[it-v.begin()]=1;
#else
	++used[it-v.begin()];
	if (used[it-v.begin()]>100)
	  used[it-v.begin()]=100;
#endif
	++count;
      }
    }
    w.clear();
    w.reserve(1+int(count*1.1));
    unsigned pos=0;
    for (it=v.begin();it!=itend;++it){
      if (*it)
	push32(w,*it,pos,unsigned(it-v.begin()));
    }
  }

  template<class tdeg_t>
  void rref_f4buchbergermod_interreduce(vectpolymod<tdeg_t> & f4buchbergerv,const vector<unsigned> & f4buchbergervG,vectpolymod<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,const vectpolymod<tdeg_t> & quo,const polymod<tdeg_t> & R,modint env,vector<int> & permutation){
    // step2: for each monomials of quo[i], shift res[G[i]] by monomial
    // set coefficient in a line of a matrix M, columns are R monomials indices
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin build M" << '\n';
    unsigned N=unsigned(R.coord.size()),i,j=0;
    unsigned c=N;
    double sknon0=0;
    vector<used_t> used(N,0);
    unsigned usedcount=0,zerolines=0,Msize=0;
    vector< vector<modint> > K(f4buchbergervG.size());
    for (i=0;i<G.size();++i){
      Msize += unsigned(quo[i].coord.size());
    }
    if ( env<(1<<24) && env*double(env)*Msize<9.223e18 ){
      vector< vector<sparse32> > M;
      M.reserve(N);
      vector<sparse_element> atrier;
      atrier.reserve(N);
      for (i=0;i<G.size();++i){
	typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
	for (;jt!=jtend;++j,++jt){
	  M.push_back(vector<sparse32>(0));
	  M[j].reserve(1+int(1.1*res[G[i]].coord.size()));
	  makeline32(res[G[i]],&jt->u,R,M[j]);
	  // CERR << M[j] << '\n';
	  if (M[j].front().shift)
	    atrier.push_back(sparse_element(M[j].front().shift,j));
	  else
	    atrier.push_back(sparse_element(*(unsigned *) &M[j][1],j));
	}
      }
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " end build M32" << '\n';
      // should not sort but compare res[G[i]]*quo[i] monomials to build M already sorted
      // CERR << "before sort " << M << '\n';
      sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
      vector< vector<sparse32> > M1(atrier.size());
      double mem=0; // mem*4=number of bytes allocated for M1
      for (i=0;i<atrier.size();++i){
	swap(M1[i],M[atrier[i].pos]);
	mem += M1[i].size();
      }
      swap(M,M1);
      bool freemem=mem>4e7; // should depend on real memory available
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " M32 sorted, rows " << M.size() << " columns " << N << " terms " << mem << " ratio " << (mem/M.size())/N <<'\n';
      // CERR << "after sort " << M << '\n';
      // step3 reduce
      vector<modint> v(N); vector<modint2> w(N);
      vector< vector<sparse32> > SK(f4buchbergerv.size());
      for (i=0;i<f4buchbergervG.size();++i){
	if (!f4buchbergerv[f4buchbergervG[i]].coord.empty()){
	  makeline<tdeg_t>(f4buchbergerv[f4buchbergervG[i]],0,R,v);
	  if (freemem){
	    polymod<tdeg_t> clearer; swap(f4buchbergerv[f4buchbergervG[i]].coord,clearer.coord);
	  }
	  c=giacmin(c,reducef4buchberger_32(v,M,env,w));
	  // convert v to a sparse vector in SK and update used
	  convert32(v,SK[i],used);
	  //CERR << v << '\n' << SK[i] << '\n';
	}
      }
      M.clear();
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " f4buchbergerv reduced " << f4buchbergervG.size() << " polynoms over " << N << " monomials, start at " << c << '\n';
      for (i=0;i<N;++i)
	usedcount += (used[i]>0);
      if (debug_infolevel>1){
	CERR << CLOCK()*1e-6 << " number of non-zero columns " << usedcount << " over " << N << '\n'; // usedcount should be approx N-M.size()=number of cols of M-number of rows
	if (debug_infolevel>2)
	  CERR << " column32 used " << used << '\n';
      }
      // create dense matrix K
      for (i=0; i<K.size(); ++i){
	vector<modint> & v =K[i];
	if (SK[i].empty()){
	  ++zerolines;
	  continue;
	}
	v.resize(usedcount);
	vector<modint>::iterator vt=v.begin();
	vector<used_t>::const_iterator ut=used.begin(),ut0=ut;
	vector<sparse32>::const_iterator st=SK[i].begin(),stend=SK[i].end();
	unsigned p=0;
	for (j=0;st!=stend;++j,++ut){
	  if (!*ut)
	    continue;
	  if (st->shift){
	    if (j==p + st->shift){
	      p += st->shift;
	      *vt=st->val;
	      ++st;
	      ++sknon0;
	    }
	  }
	  else {
	    if (j==* (unsigned *) &(*(st+1))){
	      *vt=st->val;
	      ++st;
	      p = * (unsigned *) &(*st);
	      ++st;
	      ++sknon0;
	    }
	  }
	  ++vt;
	}
#if 0
	vector<sparse32> clearer;
	swap(SK[i],clearer); // clear SK[i] memory
#endif
      }
    }
    else {
      vector< vector<sparse_element> > M;
      M.reserve(N);
      vector<sparse_element> atrier;
      atrier.reserve(N);
      for (i=0;i<G.size();++i){
	typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
	for (;jt!=jtend;++j,++jt){
	  M.push_back(vector<sparse_element>(0));
	  M[j].reserve(res[G[i]].coord.size());
	  makeline(res[G[i]],&jt->u,R,M[j]);
	  atrier.push_back(sparse_element(M[j].front().pos,j));
	}
      }
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " end build M" << '\n';
      // should not sort but compare res[G[i]]*quo[i] monomials to build M already sorted
      // CERR << "before sort " << M << '\n';
      sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
      vector< vector<sparse_element> > M1(atrier.size());
      double mem=0; // mem*8=number of bytes allocated for M1
      unsigned firstpart=0;
      for (i=0;i<atrier.size();++i){
	swap(M1[i],M[atrier[i].pos]);
	mem += M1[i].size();
	if (!M1[i].empty())
	  firstpart=giacmax(firstpart,M1[i].front().pos);
      }
      swap(M,M1);
      bool freemem=mem>4e7; // should depend on real memory available
      // sort(M.begin(),M.end(),tri);
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " M sorted, rows " << M.size() << " columns " << N << "[" << firstpart << "] terms " << mem << " ratio " << (mem/N)/M.size() << '\n';
      // CERR << "after sort " << M << '\n';
      // step3 reduce
      vector<modint> v(N);
      vector< vector<sparse_element> > SK(f4buchbergerv.size());
#ifdef x86_64
      vector<int128_t> v128(N);
      vector<modint> multiplier(M.size()); vector<unsigned> pos(M.size());
#endif
      for (i=0;i<f4buchbergervG.size();++i){
	if (!f4buchbergerv[f4buchbergervG[i]].coord.empty()){
	  makeline<tdeg_t>(f4buchbergerv[f4buchbergervG[i]],0,R,v);
	  if (freemem){
	    polymod<tdeg_t> clearer; swap(f4buchbergerv[f4buchbergervG[i]].coord,clearer.coord);
	  }
#ifdef x86_64
	  /* vector<modint> w(v);
	  // CERR << "reduce " << v << '\n' << M << '\n';
	  c=giacmin(c,reducef4buchbergerslice(w,M,env,v128,multiplier,pos));
	  c=giacmin(c,reducef4buchberger_64(v,M,env,v128));
	  if (w!=v) CERR << w << '\n' << v << '\n'; else CERR << "ok" << '\n';
	  */
	  // c=giacmin(c,reducef4buchbergerslice(v,M,env,v128,multiplier,pos));
	  if (0 && env<(1<<29) && N>10000) // it's slower despite v128 not in cache
	    c=giacmin(c,reducef4buchberger(v,M,env));
	  else
	    c=giacmin(c,reducef4buchberger_64(v,M,env,v128));
#else // x86_64
	  c=giacmin(c,reducef4buchberger(v,M,env));
#endif // x86_64
	  // convert v to a sparse vector in SK and update used
	  convert(v,SK[i],used);
	  // CERR << v << '\n' << SK[i] << '\n';
	}
      }
      M.clear();
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " f4buchbergerv reduced " << f4buchbergervG.size() << " polynoms over " << N << " monomials, start at " << c << '\n';
      for (i=0;i<N;++i)
	usedcount += (used[i]>0);
      if (debug_infolevel>1){
	CERR << CLOCK()*1e-6 << " number of non-zero columns " << usedcount << " over " << N << '\n'; // usedcount should be approx N-M.size()=number of cols of M-number of rows
	// if (debug_infolevel>2) CERR << " column use " << used << '\n';
      }
      // create dense matrix K
      for (i=0; i<K.size(); ++i){
	vector<modint> & v =K[i];
	if (SK[i].empty()){
	  ++zerolines;
	  continue;
	}
	sknon0 += SK[i].size();
	v.resize(usedcount);
	vector<modint>::iterator vt=v.begin();
	vector<used_t>::const_iterator ut=used.begin(),ut0=ut;
	vector<sparse_element>::const_iterator st=SK[i].begin(),stend=SK[i].end();
	for (j=0;st!=stend;++j,++ut){
	  if (!*ut)
	    continue;
	  if (j==st->pos){
	    *vt=st->val;
	    ++st;
	  }
	  ++vt;
	}
#if 1
	vector<sparse_element> clearer;
	swap(SK[i],clearer); // clear SK[i] memory
#endif
	// CERR << used << '\n' << SK[i] << '\n' << K[i] << '\n';
      }
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " rref " << K.size() << "x" << usedcount << " non0 " << sknon0 << " ratio " << (sknon0/K.size())/usedcount << " nulllines " << zerolines << '\n';
    vecteur pivots; vector<int> maxrankcols; longlong idet;
    // CERR << K << '\n';
    smallmodrref(1,K,pivots,permutation,maxrankcols,idet,0,int(K.size()),0,usedcount,1/* fullreduction*/,0/*dontswapbelow*/,env,0/* rrefordetorlu*/,true,0,true,-1);
    //CERR << K << "," << permutation << '\n';
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it=R.coord.begin(),itend=R.coord.end();
    vector<int> permu=perminv(permutation);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv interreduced" << '\n';
    for (i=0;i<f4buchbergervG.size();++i){
#if 0 // spare memory, keep exactly the right number of monomials in f4buchbergerv[]
      polymod<tdeg_t> tmpP(f4buchbergerv[f4buchbergervG[i]].order,f4buchbergerv[f4buchbergervG[i]].dim);
      vector<modint> & v =K[permu[i]];
      if (v.empty())
	continue;
      unsigned vcount=0;
      vector<modint>::const_iterator vt=v.begin(),vtend=v.end();
      for (;vt!=vtend;++vt){
	if (*vt)
	  ++vcount;
      }
      vector< T_unsigned<modint,tdeg_t> > & Pcoord=tmpP.coord;
      Pcoord.reserve(vcount);
      vector<used_t>::const_iterator ut=used.begin();
      for (vt=v.begin(),it=R.coord.begin();it!=itend;++ut,++it){
	if (!*ut)
	  continue;
	modint coeff=*vt;
	++vt;
	if (coeff!=0)
	  Pcoord.push_back(T_unsigned<modint,tdeg_t>(coeff,it->u));
      }
      if (!Pcoord.empty() && Pcoord.front().g!=1){
	smallmultmod(invmod(Pcoord.front().g,env),tmpP,env);
	Pcoord.front().g=1;
      }
      swap(tmpP.coord,f4buchbergerv[f4buchbergervG[i]].coord);
#else
      // CERR << v << '\n';
      vector< T_unsigned<modint,tdeg_t> > & Pcoord=f4buchbergerv[f4buchbergervG[i]].coord;
      Pcoord.clear();
      vector<modint> & v =K[permu[i]];
      if (v.empty())
	continue;
      unsigned vcount=0;
      vector<modint>::const_iterator vt=v.begin(),vtend=v.end();
      for (;vt!=vtend;++vt){
	if (*vt)
	  ++vcount;
      }
      Pcoord.reserve(vcount);
      vector<used_t>::const_iterator ut=used.begin();
      for (vt=v.begin(),it=R.coord.begin();it!=itend;++ut,++it){
	if (!*ut)
	  continue;
	modint coeff=*vt;
	++vt;
	if (coeff!=0)
	  Pcoord.push_back(T_unsigned<modint,tdeg_t>(coeff,it->u));
      }
      if (!Pcoord.empty() && Pcoord.front().g!=1){
	smallmultmod(invmod(Pcoord.front().g,env),f4buchbergerv[f4buchbergervG[i]],env);
	Pcoord.front().g=1;
      }
#endif
    }
  }


  template<class tdeg_t>
  void copycoeff(const polymod<tdeg_t> & p,vector<modint> & v){
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    v.clear();
    v.reserve(itend-it);
    for (;it!=itend;++it)
      v.push_back(it->g);
  }

  template<class tdeg_t>
  void copycoeff(const poly8<tdeg_t> & p,vector<gen> & v){
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    v.clear();
    v.reserve(itend-it);
    for (;it!=itend;++it)
      v.push_back(it->g);
  }

  // dichotomic seach for jt->u==u in [jt,jtend[
  template <class tdeg_t>
  bool dicho(typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator & jt,typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jtend,const tdeg_t & u,order_t order){
    if (jt->u==u) return true;
    for (;;){
      int step=int((jtend-jt)/2);
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator j=jt+step;
      if (j==jt)
	return j->u==u;
      //PREFETCH(&*(j+step/2));
      //PREFETCH(&*(jt+step/2));
      if (int res=tdeg_t_greater(j->u,u,order)){
	jt=j;
	if (res==2)
	  return true;
      }
      else
	jtend=j;
    }
  }

  template<class tdeg_t>
  void makelinesplit(const polymod<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<shifttype> & v){
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=R.coord.begin(),jtend=R.coord.end();
    unsigned pos=0;
    double nop1=double(R.coord.size());
    double nop2=4*p.coord.size()*std::log(nop1)/std::log(2.0);
    bool dodicho=nop2<nop1;
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	/* new faster code */
	if (dodicho && dicho(jt,jtend,u,R.order)){
	  pushsplit(v,pos,unsigned(jt-R.coord.begin()));
	  ++jt;
	  continue;
	}
	/* end new faster code */
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    pushsplit(v,pos,unsigned(jt-R.coord.begin()));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	/* new faster code */
	if (dodicho && dicho(jt,jtend,u,R.order)){
	  pushsplit(v,pos,unsigned(jt-R.coord.begin()));
	  ++jt;
	  continue;
	}
	/* end new faster code */
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    pushsplit(v,pos,unsigned(jt-R.coord.begin()));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  // return true if all shifts are <=0xffff
  bool checkshortshifts(const vector<shifttype> & v){
    if (v.empty())
      return false;
    const shifttype * it=&v.front(),*itend=it+v.size();
    // ignore first, it's not a shift
    unsigned pos;
    next_index(pos,it);
    for (;it!=itend;++it){
      if (!*it)
	return false;
    }
    return true;
  }

  template<class tdeg_t>
  void makelinesplit(const poly8<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<shifttype> & v){
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=R.coord.begin(),jt0=jt,jtend=R.coord.end();
    unsigned pos=0;
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    pushsplit(v,pos,unsigned(jt-jt0));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    pushsplit(v,pos,unsigned(jt-jt0));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  template<class tdeg_t>
  void makelinesplitu(const polymod<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<unsigned> & vu){
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=R.coord.begin(),jt0=jt,jtend=R.coord.end();
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    vu.push_back(hashgcd_U(jt-jt0));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    vu.push_back(hashgcd_U(jt-jt0));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  template<class tdeg_t>
  void makelinesplits(const polymod<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<unsigned short> & vu){
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end(),jt=R.coord.begin(),jt0=jt,jtend=R.coord.end();
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    vu.push_back(hashgcd_U(jt-jt0));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    vu.push_back(hashgcd_U(jt-jt0));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }


#define GIAC_Z

  // perhaps a good idea to lock when memory allocation occur?
#ifdef HAVE_LIBPTHREAD
    pthread_mutex_t gbasismutex = PTHREAD_MUTEX_INITIALIZER;
#endif

  template<class tdeg_t>
  void rref_f4buchbergermodsplit_interreduce(vectpolymod<tdeg_t> & f4buchbergerv,const vector<unsigned> & f4buchbergervG,vectpolymod<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,const vectpolymod<tdeg_t> & quo,const polymod<tdeg_t> & R,modint env,vector<int> & permutation){
    // step2: for each monomials of quo[i], shift res[G[i]] by monomial
    // set coefficient in a line of a matrix M, columns are R monomials indices
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin build M" << '\n';
    unsigned N=unsigned(R.coord.size()),i,j=0;
    if (N==0) return;
#if GIAC_SHORTSHIFTTYPE==16
    bool useshort=true;
#else
    bool useshort=N<=0xffff;
#endif
    unsigned nrows=0;
    for (i=0;i<G.size();++i){
      nrows += unsigned(quo[i].coord.size());
    }
    unsigned c=N;
    double sknon0=0;
    vector<used_t> used(N,0);
    unsigned usedcount=0,zerolines=0;
    vector< vector<modint> > K(f4buchbergervG.size());
    vector<vector<unsigned short> > Mindex;
    vector<vector<unsigned> > Muindex;
    vector< vector<modint> > Mcoeff(G.size());
    vector<coeffindex_t> coeffindex;
    if (useshort)
      Mindex.reserve(nrows);
    else
      Muindex.reserve(nrows);
    coeffindex.reserve(nrows);
    vector<sparse_element> atrier;
    atrier.reserve(nrows);
    for (i=0;i<G.size();++i){
      Mcoeff[i].reserve(res[G[i]].coord.size());
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      if (useshort){
	for (;jt!=jtend;++j,++jt){
	  Mindex.push_back(vector<unsigned short>(0));
#if GIAC_SHORTSHIFTTYPE==16
	  Mindex[j].reserve(int(1.1*res[G[i]].coord.size()));
#else
	  Mindex[j].reserve(res[G[i]].coord.size());
#endif
	}
      }
      else {
	for (;jt!=jtend;++j,++jt){
	  Muindex.push_back(vector<unsigned>(0));
	  Muindex[j].reserve(res[G[i]].coord.size());
	}
      }
    }
    for (i=0,j=0;i<G.size();++i){
      // copy coeffs of res[G[i]] in Mcoeff
      copycoeff(res[G[i]],Mcoeff[i]);
      // for each monomial of quo[i], find indexes and put in Mindex
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      for (;jt!=jtend;++j,++jt){
	coeffindex.push_back(coeffindex_t(N<=0xffff,i));
	if (useshort){
#if GIAC_SHORTSHIFTTYPE==16
	  makelinesplit(res[G[i]],&jt->u,R,Mindex[j]);
	  if (!coeffindex.back().b)
	    coeffindex.back().b=checkshortshifts(Mindex[j]);
	  atrier.push_back(sparse_element(first_index(Mindex[j]),j));
#else
	  makelinesplits(res[G[i]],&jt->u,R,Mindex[j]);
	  atrier.push_back(sparse_element(Mindex[j].front(),j));
#endif
	}
	else {
	  makelinesplitu(res[G[i]],&jt->u,R,Muindex[j]);
	  atrier.push_back(sparse_element(Muindex[j].front(),j));
	}
      }
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " end build Mindex/Mcoeff rref_f4buchbergermodsplit_interreduce" << '\n';
    // should not sort but compare res[G[i]]*quo[i] monomials to build M already sorted
    // CERR << "before sort " << M << '\n';
    sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
    vector<coeffindex_t> coeffindex1(atrier.size());
    double mem=0; // mem*4=number of bytes allocated for M1
    if (useshort){
      vector< vector<unsigned short> > Mindex1(atrier.size());
      for (i=0;i<atrier.size();++i){
	swap(Mindex1[i],Mindex[atrier[i].pos]);
	mem += Mindex1[i].size();
	swap(coeffindex1[i],coeffindex[atrier[i].pos]);
      }
      swap(Mindex,Mindex1);
      nrows=unsigned(Mindex.size());
    }
    else {
      vector< vector<unsigned> > Muindex1(atrier.size());
      for (i=0;i<atrier.size();++i){
	swap(Muindex1[i],Muindex[atrier[i].pos]);
	mem += Muindex1[i].size();
	swap(coeffindex1[i],coeffindex[atrier[i].pos]);
      }
      swap(Muindex,Muindex1);
      nrows=unsigned(Muindex.size());
    }
    swap(coeffindex,coeffindex1);
    vector<unsigned> firstpos(atrier.size());
    for (i=0;i < atrier.size();++i){
      firstpos[i]=atrier[i].val;
    }
    bool freemem=true; // mem>4e7; // should depend on real memory available
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " Mindex sorted, rows " << nrows << " columns " << N << " terms " << mem << " ratio " << (mem/nrows)/N <<'\n';
    // CERR << "after sort " << M << '\n';
    // step3 reduce
    vector<modint> v(N);
    vector<modint2> v64(N);
#ifdef x86_64
    vector<int128_t> v128(N);
#endif
#ifdef GIAC_Z
    if (N<nrows){
      CERR << "Error " << N << "," << nrows << '\n';
      return;
    }
    unsigned Kcols=N-nrows;
    unsigned effectivef4buchbergervGsize=0;
    for (unsigned i=0;i<f4buchbergervG.size();++i){
      if (!f4buchbergerv[f4buchbergervG[i]].coord.empty())
	++effectivef4buchbergervGsize;
    }
    vector<modint> lescoeffs;
    lescoeffs.reserve(Kcols*effectivef4buchbergervGsize);
    // vector<modint> lescoeffs(Kcols*effectivef4buchbergervGsize);
    // vector<modint>::iterator coeffit=lescoeffs.begin();
    if (debug_infolevel>1)
      CERR << "Capacity for coeffs " << lescoeffs.size() << '\n';
    vector<unsigned> lebitmap(((N>>5)+1)*effectivef4buchbergervGsize);
    unsigned * bitmap=&lebitmap.front();
#else
    vector< vector<sparse_element> > SK(f4buchbergerv.size());
#endif
    for (i=0;i<f4buchbergervG.size();++i){
      if (!f4buchbergerv[f4buchbergervG[i]].coord.empty()){
	makeline<tdeg_t>(f4buchbergerv[f4buchbergervG[i]],0,R,v);
	//CERR << v << '\n';
#ifdef x86_64
	if (useshort){
	  if (env<(1<<24)){
#if GIAC_SHORTSHIFTTYPE==16
	    c=giacmin(c,reducef4buchbergersplit(v,Mindex,firstpos,Mcoeff,coeffindex,env,v64));
#else
	    c=giacmin(c,reducef4buchbergersplits(v,Mindex,Mcoeff,coeffindex,env,v64));
#endif
	  }
	  else {
#if GIAC_SHORTSHIFTTYPE==16
	    c=giacmin(c,reducef4buchbergersplit64(v,Mindex,firstpos,Mcoeff,coeffindex,env,v128));
#else
	    c=giacmin(c,reducef4buchbergersplit64s(v,Mindex,Mcoeff,coeffindex,env,v128));
#endif
	  }
	}
	else {
	  if (env<(1<<24))
	    c=giacmin(c,reducef4buchbergersplitu(v,Muindex,Mcoeff,coeffindex,env,v64));
	  else
	    c=giacmin(c,reducef4buchbergersplit64u(v,Muindex,Mcoeff,coeffindex,env,v128));
	}
#else
	if (useshort){
#if GIAC_SHORTSHIFTTYPE==16
	  c=giacmin(c,reducef4buchbergersplit(v,Mindex,firstpos,Mcoeff,coeffindex,env,v64));
#else
	  c=giacmin(c,reducef4buchbergersplits(v,Mindex,Mcoeff,coeffindex,env,v64));
#endif
	}
	else
	  c=giacmin(c,reducef4buchbergersplitu(v,Muindex,Mcoeff,coeffindex,env,v64));
#endif
	// convert v to a sparse vector in SK and update used
	if (freemem){
	  polymod<tdeg_t> clearer; swap(f4buchbergerv[f4buchbergervG[i]].coord,clearer.coord);
	}
#ifdef GIAC_Z
	// zconvert(v,coeffit,bitmap,used); bitmap += (N>>5)+1;
	zconvert_(v,lescoeffs,bitmap,used); bitmap += (N>>5)+1;
#else
	convert(v,SK[i],used);
#endif
	//CERR << v << '\n' << SK[i] << '\n';
      }
    }
#if 0 // def GIAC_Z
    if (debug_infolevel>1) CERR << "Total size for coeffs " << coeffit-lescoeffs.begin() << '\n';
    if (freemem){
      for (i=0;i<f4buchbergervG.size();++i){
	polymod<tdeg_t> clearer; swap(f4buchbergerv[f4buchbergervG[i]].coord,clearer.coord);
      }
    }
#endif
    Mindex.clear(); Muindex.clear();
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv split reduced " << f4buchbergervG.size() << " polynoms over " << N << " monomials, start at " << c << '\n';
    for (i=0;i<N;++i)
      usedcount += (used[i]>0);
    if (debug_infolevel>1){
      CERR << CLOCK()*1e-6 << " number of non-zero columns " << usedcount << " over " << N << '\n'; // usedcount should be approx N-M.size()=number of cols of M-number of rows
      if (debug_infolevel>3)
	CERR << " column split used " << used << '\n';
    }
    // create dense matrix K
#ifdef GIAC_Z
    bitmap=&lebitmap.front();
    create_matrix(lescoeffs,bitmap,(N>>5)+1,used,K);
    if (freemem){
      // clear memory required for lescoeffs
      vector<modint> tmp; lescoeffs.swap(tmp);
      vector<unsigned> tmp1; lebitmap.swap(tmp1);
    }
#else
    for (i=0; i<K.size(); ++i){
      vector<modint> & v =K[i];
      if (SK[i].empty()){
	++zerolines;
	continue;
      }
      sknon0 += SK[i].size();
      v.resize(usedcount);
      vector<modint>::iterator vt=v.begin();
      vector<used_t>::const_iterator ut=used.begin(),ut0=ut;
      vector<sparse_element>::const_iterator st=SK[i].begin(),stend=SK[i].end();
      for (j=0;st!=stend;++j,++ut){
	if (!*ut)
	  continue;
	if (j==st->pos){
	  *vt=st->val;
	  ++st;
	}
	++vt;
      }
#if 1
      vector<sparse_element> clearer;
      swap(SK[i],clearer); // clear SK[i] memory
#endif
      // CERR << used << '\n' << SK[i] << '\n' << K[i] << '\n';
    } // end create dense matrix K
#endif // GIAC_Z
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " rref " << K.size() << "x" << usedcount << " non0 " << sknon0 << " ratio " << (sknon0/K.size())/usedcount << " nulllines " << zerolines << '\n';
    vecteur pivots; vector<int> maxrankcols; longlong idet;
    //CERR << K << '\n';
    smallmodrref(1,K,pivots,permutation,maxrankcols,idet,0,int(K.size()),0,usedcount,1/* fullreduction*/,0/*dontswapbelow*/,env,0/* rrefordetorlu*/,true,0,true,-1);
    //CERR << K << "," << permutation << '\n';
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it=R.coord.begin(),itend=R.coord.end();
    vector<int> permu=perminv(permutation);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv interreduced" << '\n';
    for (i=0;i<f4buchbergervG.size();++i){
      // CERR << v << '\n';
      vector< T_unsigned<modint,tdeg_t> > & Pcoord=f4buchbergerv[f4buchbergervG[i]].coord;
      Pcoord.clear();
      vector<modint> & v =K[permu[i]];
      if (v.empty())
	continue;
      unsigned vcount=0;
      vector<modint>::const_iterator vt=v.begin(),vtend=v.end();
      for (;vt!=vtend;++vt){
	if (*vt)
	  ++vcount;
      }
      Pcoord.reserve(vcount);
      vector<used_t>::const_iterator ut=used.begin();
      for (vt=v.begin(),it=R.coord.begin();it!=itend;++ut,++it){
	if (!*ut)
	  continue;
	modint coeff=*vt;
	++vt;
	if (coeff!=0)
	  Pcoord.push_back(T_unsigned<modint,tdeg_t>(coeff,it->u));
      }
      if (!Pcoord.empty() && Pcoord.front().g!=1){
	smallmultmod(invmod(Pcoord.front().g,env),f4buchbergerv[f4buchbergervG[i]],env);
	Pcoord.front().g=1;
      }
    }
  }

  template<class tdeg_t>
  void rref_f4buchbergermod_nointerreduce(vectpolymod<tdeg_t> & f4buchbergerv,const vector<unsigned> & f4buchbergervG,vectpolymod<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,const vectpolymod<tdeg_t> & quo,const polymod<tdeg_t> & R,modint env,vector<int> & permutation){
    unsigned N=unsigned(R.coord.size()),i,j=0;
    for (i=0;i<G.size();++i){
      if (!quo[i].coord.empty())
	break;
    }
    if (i==G.size()){
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " No inter-reduction" << '\n';
      return;
    }
    // step2: for each monomials of quo[i], shift res[G[i]] by monomial
    // set coefficient in a line of a matrix M, columns are R monomials indices
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin build M" << '\n';
    vector< vector<sparse_element> > M;
    M.reserve(N);
    vector<sparse_element> atrier;
    atrier.reserve(N);
    for (i=0;i<G.size();++i){
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      for (;jt!=jtend;++j,++jt){
	M.push_back(vector<sparse_element>(0));
	M[j].reserve(res[G[i]].coord.size());
	makeline(res[G[i]],&jt->u,R,M[j]);
	atrier.push_back(sparse_element(M[j].front().pos,j));
      }
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " end build M" << '\n';
    // should not sort but compare res[G[i]]*quo[i] monomials to build M already sorted
    // CERR << "before sort " << M << '\n';
    sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
    vector< vector<sparse_element> > M1(atrier.size());
    for (i=0;i<atrier.size();++i){
      swap(M1[i],M[atrier[i].pos]);
    }
    swap(M,M1);
    // sort(M.begin(),M.end(),tri);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " M sorted, rows " << M.size() << " columns " << N << " #basis to reduce" << f4buchbergervG.size() << '\n';
    // CERR << "after sort " << M << '\n';
    // step3 reduce
    unsigned c=N;
    vector<modint> v(N);
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it=R.coord.begin(),itend=R.coord.end();
#ifdef x86_64
    vector<int128_t> v128(N);
#endif
    for (i=0;i<f4buchbergervG.size();++i){
      if (!f4buchbergerv[f4buchbergervG[i]].coord.empty()){
	makeline(f4buchbergerv[f4buchbergervG[i]],0,R,v);
	// CERR << v << '\n';
#ifdef x86_64
	/* if (N>=4096)
	  c=giacmin(c,reducef4buchberger(v,M,env));
	  else */
	  c=giacmin(c,reducef4buchberger_64(v,M,env,v128));
#else
	c=giacmin(c,reducef4buchberger(v,M,env));
#endif
	vector< T_unsigned<modint,tdeg_t> > & Pcoord=f4buchbergerv[f4buchbergervG[i]].coord;
	Pcoord.clear();
	unsigned vcount=0;
	vector<modint>::const_iterator vt=v.begin(),vtend=v.end();
	for (;vt!=vtend;++vt){
	  if (*vt)
	    ++vcount;
	}
	Pcoord.reserve(vcount);
	for (vt=v.begin(),it=R.coord.begin();it!=itend;++it){
	  modint coeff=*vt;
	  ++vt;
	  if (coeff!=0)
	    Pcoord.push_back(T_unsigned<modint,tdeg_t>(coeff,it->u));
	}
	if (!Pcoord.empty() && Pcoord.front().g!=1){
	  smallmultmod(invmod(Pcoord.front().g,env),f4buchbergerv[f4buchbergervG[i]],env);
	  Pcoord.front().g=1;
	}
      }
    }
  }

  template<class tdeg_t>
  void rref_f4buchbergermod(vectpolymod<tdeg_t> & f4buchbergerv,vectpolymod<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded,const vectpolymod<tdeg_t> & quo,const polymod<tdeg_t> & R,modint env,vector<int> & permutation,bool split){
    vector<unsigned> f4buchbergervG(f4buchbergerv.size());
    for (unsigned i=0;i<f4buchbergerv.size();++i)
      f4buchbergervG[i]=i;
#if 0
    rref_f4buchbergermod(f4buchbergerv,f4buchbergervG,res,G,excluded,quo,R,env,permutation,true);
#else
    if (//1
	split
	//0
	)
      rref_f4buchbergermodsplit_interreduce(f4buchbergerv,f4buchbergervG,res,G,excluded,quo,R,env,permutation);
    else
      rref_f4buchbergermod_interreduce(f4buchbergerv,f4buchbergervG,res,G,excluded,quo,R,env,permutation);
#endif
  }

  template<class tdeg_t>
  struct info_t {
    vectpolymod<tdeg_t> quo,quo2;
    polymod<tdeg_t> R,R2;
    vector<int> permu;
    vector< paire > B;
    vector<unsigned> G;
    unsigned nonzero;
  };

  template<class tdeg_t>
  void reducemodf4buchberger(vectpolymod<tdeg_t> & f4buchbergerv,vectpolymod<tdeg_t> & res,const vector<unsigned> & G,unsigned excluded, modint env,info_t<tdeg_t> & info_tmp){
    polymod<tdeg_t> allf4buchberger(f4buchbergerv.front().order,f4buchbergerv.front().dim),rem(f4buchbergerv.front().order,f4buchbergerv.front().dim);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchberger begin collect monomials on #polys " << f4buchbergerv.size() << '\n';
    // collect all terms in f4buchbergerv
    collect(f4buchbergerv,allf4buchberger);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchberger symbolic preprocess" << '\n';
    // find all monomials required to reduce all polymod<tdeg_t> in f4buchberger with res[G[.]]
    symbolic_preprocess(allf4buchberger,res,G,excluded,info_tmp.quo,rem,&info_tmp.R);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchberger end symbolic preprocess" << '\n';
    // build a matrix with first lines res[G[.]]*quo[.] in terms of monomials in S
    // and finishing with lines of f4buchbergerv
    // rref (below) the matrix and find the last lines in f4buchbergerv
    rref_f4buchbergermod(f4buchbergerv,res,G,excluded,info_tmp.quo,info_tmp.R,env,info_tmp.permu,true); // do splitting
  }

  // v -= v1, assumes that no overflow can happen (modulo must be 2^30)
  void sub(vector<modint> & v,const vector<modint> & v1,modint env){
    vector<modint>::const_iterator jt=v1.begin();
    vector<modint>::iterator it=v.begin(),itend=v.end();
    for (;it!=itend;++jt,++it){
      *it -= *jt;
      if (*it>-env && *it<env)
	continue;
      if (*it>env)
	*it -=env;
      else
	*it += env;
    }
#if 0
    // normalize first element to 1
    for (it=v.begin();it!=itend;++it){
      if (*it)
	break;
    }
    if (it!=itend){
      modint c=invmod(*it,env);
      *it=1;
      for (++it;it!=itend;++it){
	if (*it)
	  *it=(longlong(c)*(*it))%env;
      }
    }
#endif
  }

  void sub(vector<modint2> & v,const vector<modint2> & v1){
    vector<modint2>::const_iterator jt=v1.begin();
    vector<modint2>::iterator it=v.begin(),itend=v.end();
    for (;it!=itend;++jt,++it){
      *it -= *jt;
    }
  }

  void sub(vector<double> & v,const vector<modint2> & v1){
    vector<modint2>::const_iterator jt=v1.begin();
    vector<double>::iterator it=v.begin(),itend=v.end();
    for (;it!=itend;++jt,++it){
      *it -= *jt;
    }
  }

  template<class tdeg_t>
  int f4mod(vectpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,vector< paire > & B,vectpolymod<tdeg_t> & f4buchbergerv,bool learning,unsigned & learned_position,vector< paire > * pairs_reducing_to_zero,vector< info_t<tdeg_t> >* f4buchberger_info,unsigned & f4buchberger_info_position,bool recomputeR){
    if (B.empty())
      return 0;
    vector<tdeg_t> leftshift(B.size());
    vector<tdeg_t> rightshift(B.size());
    leftright(res,B,leftshift,rightshift);
    f4buchbergerv.resize(B.size());
    info_t<tdeg_t> info_tmp;
    unsigned nonzero=unsigned(B.size());
    info_t<tdeg_t> * info_ptr=&info_tmp;
    if (!learning && f4buchberger_info && f4buchberger_info_position<f4buchberger_info->size()){
      info_ptr=&(*f4buchberger_info)[f4buchberger_info_position];
      ++f4buchberger_info_position;
      nonzero=info_ptr->nonzero;
    }
    else {
      polymod<tdeg_t> all(res[B[0].first].order,res[B[0].first].dim),rem;
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " f4buchberger begin collect monomials on #polys " << f4buchbergerv.size() << '\n';
      collect(res,B,all,leftshift,rightshift);
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " f4buchberger symbolic preprocess" << '\n';
      symbolic_preprocess(all,res,G,-1,info_tmp.quo,rem,&info_tmp.R);
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " end symbolic preprocess, rem size " << rem.coord.size() << '\n';
    }
    polymod<tdeg_t> & R = info_ptr->R;
    vectpolymod<tdeg_t> & quo = info_ptr->quo;
	unsigned N = unsigned(R.coord.size()), i, j = 0;
    if (N==0){
      if (learning && f4buchberger_info)
	f4buchberger_info->push_back(*info_ptr);
      return 1;
    }
#if GIAC_SHORTSHIFTTYPE==16
    bool useshort=true;
#else
    bool useshort=N<=0xffff;
#endif
    unsigned nrows=0;
    for (i=0;i<G.size();++i){
		nrows += unsigned(quo[i].coord.size());
    }
    unsigned c=N;
    double sknon0=0;
    vector<used_t> used(N,0);
    unsigned usedcount=0,zerolines=0;
    vector< vector<modint> > K(B.size());
    vector<vector<unsigned short> > Mindex;
    vector<vector<unsigned> > Muindex;
    vector< vector<modint> > Mcoeff(G.size());
    vector<coeffindex_t> coeffindex;
    if (useshort)
      Mindex.reserve(nrows);
    else
      Muindex.reserve(nrows);
    coeffindex.reserve(nrows);
    vector<sparse_element> atrier;
    atrier.reserve(nrows);
    for (i=0;i<G.size();++i){
      Mcoeff[i].reserve(res[G[i]].coord.size());
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      if (useshort){
	for (;jt!=jtend;++j,++jt){
	  Mindex.push_back(vector<unsigned short>(0));
#if GIAC_SHORTSHIFTTYPE==16
	  Mindex[j].reserve(int(1.1*res[G[i]].coord.size()));
#else
	  Mindex[j].reserve(res[G[i]].coord.size());
#endif
	}
      }
      else {
	for (;jt!=jtend;++j,++jt){
	  Muindex.push_back(vector<unsigned>(0));
	  Muindex[j].reserve(res[G[i]].coord.size());
	}
      }
    }
    for (i=0,j=0;i<G.size();++i){
      // copy coeffs of res[G[i]] in Mcoeff
      copycoeff(res[G[i]],Mcoeff[i]);
      // for each monomial of quo[i], find indexes and put in Mindex
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      for (;jt!=jtend;++j,++jt){
	coeffindex.push_back(coeffindex_t(N<=0xffff,i));
	if (useshort){
#if GIAC_SHORTSHIFTTYPE==16
	  makelinesplit(res[G[i]],&jt->u,R,Mindex[j]);
	  if (!coeffindex.back().b)
	    coeffindex.back().b=checkshortshifts(Mindex[j]);
	  atrier.push_back(sparse_element(first_index(Mindex[j]),j));
#else
	  makelinesplits(res[G[i]],&jt->u,R,Mindex[j]);
	  atrier.push_back(sparse_element(Mindex[j].front(),j));
#endif
	}
	else {
	  makelinesplitu(res[G[i]],&jt->u,R,Muindex[j]);
	  atrier.push_back(sparse_element(Muindex[j].front(),j));
	}
      }
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " end build Mindex/Mcoeff f4mod" << '\n';
    // should not sort but compare res[G[i]]*quo[i] monomials to build M already sorted
    // CERR << "before sort " << M << '\n';
    sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
    vector<coeffindex_t> coeffindex1(atrier.size());
    double mem=0; // mem*4=number of bytes allocated for M1
    if (useshort){
      vector< vector<unsigned short> > Mindex1(atrier.size());
      for (i=0;i<atrier.size();++i){
	swap(Mindex1[i],Mindex[atrier[i].pos]);
	mem += Mindex1[i].size();
	swap(coeffindex1[i],coeffindex[atrier[i].pos]);
      }
      swap(Mindex,Mindex1);
	  nrows = unsigned(Mindex.size());
    }
    else {
      vector< vector<unsigned> > Muindex1(atrier.size());
      for (i=0;i<atrier.size();++i){
	swap(Muindex1[i],Muindex[atrier[i].pos]);
	mem += Muindex1[i].size();
	swap(coeffindex1[i],coeffindex[atrier[i].pos]);
      }
      swap(Muindex,Muindex1);
	  nrows = unsigned(Muindex.size());
    }
    swap(coeffindex,coeffindex1);
    vector<unsigned> firstpos(atrier.size());
    for (i=0;i < atrier.size();++i){
      firstpos[i]=atrier[i].val;
    }
    bool freemem=mem>4e7; // should depend on real memory available
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " Mindex sorted, rows " << nrows << " columns " << N << " terms " << mem << " ratio " << (mem/nrows)/N <<'\n';
    // CERR << "after sort " << M << '\n';
    // step3 reduce
    vector<modint> v(N);
    vector<modint2> v64(N);
#ifdef x86_64
    vector<int128_t> v128(N);
#endif
    if (N<nrows){
      CERR << "Error " << N << "," << nrows << '\n';
      return -1;
    }
    unsigned Kcols=N-nrows;
    vector<unsigned> lebitmap(((N>>5)+1)*B.size());
    unsigned * bitmap=&lebitmap.front();
    for (i=0;i<B.size();++i){
      paire bk=B[i];
      if (!learning && pairs_reducing_to_zero && learned_position<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[learned_position]){
	if (debug_infolevel>2)
	  CERR << bk << " f4buchberger learned " << learned_position << '\n';
	++learned_position;
	unsigned tofill=(N>>5)+1;
	fill(bitmap,bitmap+tofill,0);
	bitmap += tofill;
	continue;
      }
      makeline(res[bk.first],&leftshift[i],R,v,1);
      makelinesub(res[bk.second],&rightshift[i],R,v,1,env);
      // CERR << v << '\n' << v2 << '\n';
      // sub(v,v2,env);
      // CERR << v << '\n';
#ifdef x86_64
      if (useshort){
	if (env<(1<<24)){
#if GIAC_SHORTSHIFTTYPE==16
	  c=giacmin(c,reducef4buchbergersplit(v,Mindex,firstpos,Mcoeff,coeffindex,env,v64));
#else
	  c=giacmin(c,reducef4buchbergersplits(v,Mindex,Mcoeff,coeffindex,env,v64));
#endif
	}
	else {
#if GIAC_SHORTSHIFTTYPE==16
	  c=giacmin(c,reducef4buchbergersplit64(v,Mindex,firstpos,Mcoeff,coeffindex,env,v128));
#else
	  c=giacmin(c,reducef4buchbergersplit64s(v,Mindex,Mcoeff,coeffindex,env,v128));
#endif
	}
      }
      else {
	if (env<(1<<24))
	  c=giacmin(c,reducef4buchbergersplitu(v,Muindex,Mcoeff,coeffindex,env,v64));
	else
	  c=giacmin(c,reducef4buchbergersplit64u(v,Muindex,Mcoeff,coeffindex,env,v128));
      }
#else // x86_64
      if (useshort){
#if GIAC_SHORTSHIFTTYPE==16
	c=giacmin(c,reducef4buchbergersplit(v,Mindex,firstpos,Mcoeff,coeffindex,env,v64));
#else
	c=giacmin(c,reducef4buchbergersplits(v,Mindex,Mcoeff,coeffindex,env,v64));
#endif
      }
      else
	c=giacmin(c,reducef4buchbergersplitu(v,Muindex,Mcoeff,coeffindex,env,v64));
#endif // x86_64
      // zconvert(v,coeffit,bitmap,used); bitmap += (N>>5)+1;
      K[i].reserve(Kcols);
      zconvert_(v,K[i],bitmap,used); bitmap += (N>>5)+1;
      //CERR << v << '\n' << SK[i] << '\n';
    } // end for (i=0;i<B.size();++i)
    Mindex.clear(); Muindex.clear();
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv split reduced " << B.size() << " polynoms over " << N << " monomials, start at " << c << '\n';
    for (i=0;i<N;++i)
      usedcount += (used[i]>0);
    if (debug_infolevel>1){
      CERR << CLOCK()*1e-6 << " number of non-zero columns " << usedcount << " over " << N << '\n'; // usedcount should be approx N-M.size()=number of cols of M-number of rows
      if (debug_infolevel>3)
	CERR << " column split used " << used << '\n';
    }
    // create dense matrix K
    bitmap=&lebitmap.front();
    create_matrix(bitmap,(N>>5)+1,used,K);
    // clear memory required for lescoeffs
    //vector<modint> tmp; lescoeffs.swap(tmp);
    { vector<unsigned> tmp1; lebitmap.swap(tmp1); }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " rref " << K.size() << "x" << usedcount << '\n';
    vecteur pivots; vector<int> permutation,maxrankcols; longlong idet;
    // CERR << K << '\n';
    smallmodrref(1,K,pivots,permutation,maxrankcols,idet,0,int(K.size()),0,usedcount,1/* fullreduction*/,0/*dontswapbelow*/,env,0/* rrefordetorlu*/,true,0,true,-1);
    //CERR << K << '\n';
    unsigned first0 = unsigned(pivots.size());
    if (first0<K.size() && (learning || !f4buchberger_info)){
      vector<modint> & tmpv=K[first0];
      for (i=0;i<tmpv.size();++i){
	if (tmpv[i])
	  break;
      }
      if (i==tmpv.size()){
	unsigned Ksize = unsigned(K.size());
	K.resize(first0);
	K.resize(Ksize);
      }
    }
    //CERR << permutation << K << '\n';
    if (!learning && f4buchberger_info){
      // check that permutation is the same as learned permutation
      for (unsigned j=0;j<permutation.size();++j){
	if (permutation[j]!=info_ptr->permu[j]){
	  CERR << "learning failed"<<'\n';
	  return -1;
	}
      }
    }
    if (learning)
      info_ptr->permu=permutation;
    // CERR << K << "," << permutation << '\n';
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it=R.coord.begin(),itend=R.coord.end();
    // vector<int> permu=perminv(permutation);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv interreduced" << '\n';
    for (i=0;i<f4buchbergerv.size();++i){
      // CERR << v << '\n';
      vector< T_unsigned<modint,tdeg_t> > & Pcoord=f4buchbergerv[permutation[i]].coord;
      Pcoord.clear();
      vector<modint> & v =K[i];
      if (v.empty()){
	continue;
      }
      unsigned vcount=0;
      vector<modint>::const_iterator vt=v.begin(),vtend=v.end();
      for (;vt!=vtend;++vt){
	if (*vt)
	  ++vcount;
      }
      Pcoord.reserve(vcount);
      vector<used_t>::const_iterator ut=used.begin();
      for (vt=v.begin(),it=R.coord.begin();it!=itend;++ut,++it){
	if (!*ut)
	  continue;
	modint coeff=*vt;
	++vt;
	if (coeff!=0)
	  Pcoord.push_back(T_unsigned<modint,tdeg_t>(coeff,it->u));
      }
      if (!Pcoord.empty() && Pcoord.front().g!=1){
	smallmultmod(invmod(Pcoord.front().g,env),f4buchbergerv[permutation[i]],env);
	Pcoord.front().g=1;
      }
      if (freemem){
	vector<modint> tmp; tmp.swap(v);
      }
    }
    if (learning && f4buchberger_info){
#if 0
      f4buchberger_info->push_back(*info_ptr);
#else
      info_t<tdeg_t> tmp;
      f4buchberger_info->push_back(tmp);
      info_t<tdeg_t> & i=f4buchberger_info->back();
      swap(i.quo,info_ptr->quo);
      swap(i.permu,info_ptr->permu);
      swap(i.R.coord,info_ptr->R.coord);
      i.R.order=info_ptr->R.order;
      i.R.dim=info_ptr->R.dim;
#endif
    }
    return 1;
  }

  template<class tdeg_t>
  bool apply(vector<int> permu,vectpolymod<tdeg_t> & res){
    vectpolymod<tdeg_t> tmp;
    for (unsigned i=0;i<res.size();++i){
      tmp.push_back(polymod<tdeg_t>(res.front().order,res.front().dim));
      swap(tmp[i].coord,res[permu[i]].coord);
      tmp[i].sugar=res[permu[i]].sugar;
    }
    swap(tmp,res);
    return true;
  }

  template<class tdeg_t>
  int f4mod(vectpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,vector< paire > & smallposp,vectpolymod<tdeg_t> & f4buchbergerv,bool learning,unsigned & learned_position,vector< paire > * pairs_reducing_to_zero,info_t<tdeg_t> & information,vector< info_t<tdeg_t> >* f4buchberger_info,unsigned & f4buchberger_info_position,bool recomputeR, polymod<tdeg_t> & TMP1,polymod<tdeg_t> & TMP2){
    // Improve: we don't really need to compute the s-polys here
    // it's sufficient to do that at linalg step
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " Computing s-polys " << smallposp.size() << '\n';
    if (f4buchbergerv.size()<smallposp.size())
      f4buchbergerv.clear();
    f4buchbergerv.resize(smallposp.size());
    for (unsigned i=0;i<smallposp.size();++i){
      f4buchbergerv[i].dim=TMP1.dim; f4buchbergerv[i].order=TMP1.order;
      paire bk=smallposp[i];
      if (!learning && pairs_reducing_to_zero && f4buchberger_info && learned_position<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[learned_position]){
	if (debug_infolevel>2)
	  CERR << bk << " f4buchberger learned " << learned_position << '\n';
	++learned_position;
	f4buchbergerv[i].coord.clear();
	continue;
      }
      if (debug_infolevel>2)
	CERR << bk << " f4buchberger not learned " << learned_position << '\n';
      if (debug_infolevel>2 && (equalposcomp(G,bk.first)==0 || equalposcomp(G,bk.second)==0))
	CERR << CLOCK()*1e-6 << " mod reducing pair with 1 element not in basis " << bk << '\n';
      // polymod<tdeg_t> h(res.front().order,res.front().dim);
      spolymod<tdeg_t>(res[bk.first],res[bk.second],TMP1,TMP2,env);
      f4buchbergerv[i].coord.swap(TMP1.coord);
    }
    if (f4buchbergerv.empty())
      return 0;
    // reduce spolys in f4buchbergerv
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " base size " << G.size() << " reduce f4buchberger begin on " << f4buchbergerv.size() << " pairs" << '\n';
    if (!learning && f4buchberger_info && f4buchberger_info_position<f4buchberger_info->size()){
      info_t<tdeg_t> & info=(*f4buchberger_info)[f4buchberger_info_position];
      // apply(perminv(info.permu),f4buchbergerv);
      if (recomputeR){
	swap(information.permu,info.permu);
	reducemodf4buchberger(f4buchbergerv,res,G,-1,env,info);
	swap(information.permu,info.permu);
      }
      else
	rref_f4buchbergermod(f4buchbergerv,res,G,-1,info.quo,info.R,env,information.permu,false); // don't split
      // apply(info.permu,f4buchbergerv);
      // information.permu should be identity, otherwise the whole learning process failed
      for (unsigned j=0;j<information.permu.size();++j){
	if (information.permu[j]!=info.permu[j]){
	  CERR << "learning failed"<<'\n';
	  return -1;
	}
      }
      ++f4buchberger_info_position;
    }
    else {
      reducemodf4buchberger(f4buchbergerv,res,G,-1,env,information);
      if (learning && f4buchberger_info){
#if 0
	f4buchberger_info->push_back(information);
#else
	info_t<tdeg_t> tmp;
	f4buchberger_info->push_back(tmp);
	info_t<tdeg_t> & i=f4buchberger_info->back();
	swap(i.quo,information.quo);
	swap(i.permu,information.permu);
	swap(i.R.coord,information.R.coord);
	i.R.order=information.R.order;
	i.R.dim=information.R.dim;
#endif
      }
    }
    return 1;
  }

  template<class tdeg_t>
  bool in_gbasisf4buchbergermod(vectpolymod<tdeg_t> &res,unsigned ressize,vector<unsigned> & G,modint env,bool totdeg,vector< paire > * pairs_reducing_to_zero,vector< info_t<tdeg_t> > * f4buchberger_info,bool recomputeR){
    unsigned cleared=0;
    unsigned learned_position=0,f4buchberger_info_position=0;
    bool sugar=false,learning=pairs_reducing_to_zero && pairs_reducing_to_zero->empty();
    if (debug_infolevel>1000)
      res.dbgprint(); // instantiate dbgprint()
    int capa=512;
    if (f4buchberger_info)
      capa=int(f4buchberger_info->capacity());
    polymod<tdeg_t> TMP1(res.front().order,res.front().dim),TMP2(res.front().order,res.front().dim);
    vector< paire > B,BB;
    B.reserve(256); BB.reserve(256);
    vector<unsigned> smallposv;
    smallposv.reserve(256);
    info_t<tdeg_t> information;
    order_t order=res.front().order;
    if (order.o==_PLEX_ORDER) // if (order!=_REVLEX_ORDER && order!=_TDEG_ORDER)
      totdeg=false;
    vector<unsigned> oldG(G);
    for (unsigned l=0;l<ressize;++l){
#ifdef GIAC_REDUCEMODULO
      reducesmallmod(res[l],res,G,-1,env,TMP2,env!=0);
#endif
      gbasis_updatemod(G,B,res,l,TMP2,env,true,oldG);
    }
    for (int age=1;!B.empty() && !interrupted && !ctrl_c;++age){
      if (age+1>=capa)
	return false; // otherwise reallocation will make pointers invalid
      oldG=G;
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " begin new iteration mod, " << env << " number of pairs: " << B.size() << ", base size: " << G.size() << '\n';
      if (1){
	// mem clear: remove res[i] if i is not in G nor in B
	vector<bool> clean(G.back()+1,true);
	for (unsigned i=0;i<G.size();++i){
	  clean[G[i]]=false;
	}
	for (unsigned i=0;i<B.size();++i){
	  clean[B[i].first]=false;
	  clean[B[i].second]=false;
	}
	for (unsigned i=0;i<clean.size() && i<res.size();++i){
	  if (clean[i] && res[i].coord.capacity()>1 && !res[i].coord.empty()){
	    cleared += unsigned(res[i].coord.capacity()) - 1;
	    polymod<tdeg_t> clearer;
	    clearer.coord.push_back(res[i].coord.front());
	    clearer.coord.swap(res[i].coord);
	  }
	}
      }
      // find smallest lcm pair in B
      tdeg_t small0,cur;
      unsigned smallpos,smalltotdeg=0,curtotdeg=0,smallsugar=0,cursugar=0;
      smallposv.clear();
      for (smallpos=0;smallpos<B.size();++smallpos){
	if (!res[B[smallpos].first].coord.empty() && !res[B[smallpos].second].coord.empty())
	  break;
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return false;
      }
      index_lcm(res[B[smallpos].first].coord.front().u,res[B[smallpos].second].coord.front().u,small0,order);
      smallsugar=res[B[smallpos].first].sugar+(small0-res[B[smallpos].first].coord.front().u).total_degree(order);
      smalltotdeg=small0.total_degree(order);
      smallposv.push_back(smallpos);
      for (unsigned i=smallpos+1;i<B.size();++i){
#ifdef TIMEOUT
	control_c();
#endif
	if (interrupted || ctrl_c)
	  return false;
	if (res[B[i].first].coord.empty() || res[B[i].second].coord.empty())
	  continue;
	bool doswap=false;
	index_lcm(res[B[i].first].coord.front().u,res[B[i].second].coord.front().u,cur,order);
	cursugar=res[B[smallpos].first].sugar+(cur-res[B[smallpos].first].coord.front().u).total_degree(order);
	curtotdeg=cur.total_degree(order);
	if ( !totdeg || order.o==_PLEX_ORDER)
	  doswap=tdeg_t_strictly_greater(small0,cur,order);
	else {
	  if (sugar){
	    if (smallsugar!=cursugar)
	      doswap = smallsugar > cursugar;
	  }
	  else {
	    if (smalltotdeg!=curtotdeg)
	      doswap = smalltotdeg > curtotdeg;
	  }
	}
	if (doswap){
	  smallsugar=cursugar;
	  smalltotdeg=curtotdeg;
	  // CERR << "swap mod " << curtotdeg << " " << res[B[i].first].coord.front().u << " " << res[B[i].second].coord.front().u << '\n';
	  swap(small0,cur); // small=cur;
	  smallpos=i;
	  smallposv.clear();
	  smallposv.push_back(i);
	}
	else {
	  if (totdeg && curtotdeg==smalltotdeg && (!sugar || cursugar==smallsugar))
	    smallposv.push_back(i);
	}
      }
      if (smallposv.size()<=GBASISF4_BUCHBERGER){
	unsigned i=smallposv[0];
	paire bk=B[i];
	B.erase(B.begin()+i);
	if (!learning && pairs_reducing_to_zero && learned_position<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[learned_position]){
	  if (debug_infolevel>2)
	    CERR << bk << " learned " << learned_position << '\n';
	  ++learned_position;
	  continue;
	}
	if (debug_infolevel>2)
	  CERR << bk << " not learned " << learned_position << '\n';
	if (debug_infolevel>2 && (equalposcomp(G,bk.first)==0 || equalposcomp(G,bk.second)==0))
	  CERR << CLOCK()*1e-6 << " mod reducing pair with 1 element not in basis " << bk << '\n';
	// polymod<tdeg_t> h(res.front().order,res.front().dim);
	spolymod<tdeg_t>(res[bk.first],res[bk.second],TMP1,TMP2,env);
	if (debug_infolevel>1){
	  CERR << CLOCK()*1e-6 << " mod reduce begin, pair " << bk << " spoly size " << TMP1.coord.size() << " sugar degree " << TMP1.sugar << " totdeg deg " << TMP1.coord.front().u.total_degree(order) << " degree " << TMP1.coord.front().u << '\n';
	}
#if 0 // def GBASIS_HEAP
	heap_reducemod(TMP1,res,G,-1,information.quo,TMP2,env);
	swap(TMP1.coord,TMP2.coord);
#else
	reducemod(TMP1,res,G,-1,TMP1,env);
#endif
	if (debug_infolevel>1){
	  if (debug_infolevel>3){ CERR << TMP1 << '\n'; }
	  CERR << CLOCK()*1e-6 << " mod reduce end, remainder size " << TMP1.coord.size() << " begin gbasis update" << '\n';
	}
	if (!TMP1.coord.empty()){
	  increase(res);
	  if (ressize==res.size())
	    res.push_back(polymod<tdeg_t>(TMP1.order,TMP1.dim));
	  swap(res[ressize].coord,TMP1.coord);
	  ++ressize;
#if GBASIS_POSTF4BUCHBERGER==0
	  // this does not warrant full interreduced answer
	  // because at the final step we assume that each spoly
	  // is reduced by the previous spolys in res
	  // either by the first reduction or by inter-reduction
	  // here at most GBASISF4_BUCHBERGER spolys may not be reduced by the previous ones
	  // it happens for example for cyclic8, element no 97
	  gbasis_updatemod(G,B,res,ressize-1,TMP2,env,false,oldG);
#else
	  gbasis_updatemod(G,B,res,ressize-1,TMP2,env,true,oldG);
#endif
	  if (debug_infolevel>3)
	    CERR << CLOCK()*1e-6 << " mod basis indexes " << G << " pairs indexes " << B << '\n';
	}
	else {
	  if (learning && pairs_reducing_to_zero){
	    if (debug_infolevel>2)
	      CERR << "learning " << bk << '\n';
	    pairs_reducing_to_zero->push_back(bk);
	  }
	}
	continue;
      }
      vector< paire > smallposp;
      if (smallposv.size()==B.size()){
	swap(smallposp,B);
	B.clear();
      }
      else {
	for (unsigned i=0;i<smallposv.size();++i)
	  smallposp.push_back(B[smallposv[i]]);
	// remove pairs
	for (int i=int(smallposv.size())-1;i>=0;--i)
	  B.erase(B.begin()+smallposv[i]);
      }
      vectpolymod<tdeg_t> f4buchbergerv; // collect all spolys
      int f4res=-1;
      if (1 && env<(1<<30))
	f4res=f4mod(res,G,env,smallposp,f4buchbergerv,learning,learned_position,pairs_reducing_to_zero,f4buchberger_info,f4buchberger_info_position,recomputeR);
      else
	f4res=f4mod(res,G,env,smallposp,f4buchbergerv,learning,learned_position,pairs_reducing_to_zero,information,f4buchberger_info,f4buchberger_info_position,recomputeR,TMP1,TMP2);
      if (f4res==-1)
	return false;
      if (f4res==0)
	continue;
      // update gbasis and learning
      // requires that Gauss pivoting does the same permutation for other primes
      if (learning && pairs_reducing_to_zero){
	for (unsigned i=0;i<f4buchbergerv.size();++i){
	  if (f4buchbergerv[i].coord.empty()){
	    if (debug_infolevel>2)
	      CERR << "learning f4buchberger " << smallposp[i] << '\n';
	    pairs_reducing_to_zero->push_back(smallposp[i]);
	  }
	}
      }
      unsigned added=0;
      for (unsigned i=0;i<f4buchbergerv.size();++i){
	if (!f4buchbergerv[i].coord.empty())
	  ++added;
      }
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " reduce f4buchberger end on " << added << " from " << f4buchbergerv.size() << " pairs, gbasis update begin" << '\n';
      for (unsigned i=0;i<f4buchbergerv.size();++i){
	if (!f4buchbergerv[i].coord.empty()){
	  increase(res);
	  if (ressize==res.size())
	    res.push_back(polymod<tdeg_t>(TMP1.order,TMP1.dim));
	  swap(res[ressize].coord,f4buchbergerv[i].coord);
	  ++ressize;
#ifdef GBASIS_POSTF4BUCHBERGER
#if GBASIS_POSTF4BUCHBERGER==0
	  if (learning || !f4buchberger_info || f4buchberger_info_position-1>=f4buchberger_info->size())
	    gbasis_updatemod(G,B,res,ressize-1,TMP2,env,false,oldG);
#else
	  gbasis_updatemod(G,B,res,ressize-1,TMP2,env,added<=GBASISF4_BUCHBERGER,oldG);
#endif
#else
	  gbasis_updatemod(G,B,res,ressize-1,TMP2,env,true,oldG);
#endif
	}
	else {
	  // if (!learning && pairs_reducing_to_zero)  CERR << " error learning "<< '\n';
	}
      }
#if GBASIS_POSTF4BUCHBERGER==0
      if (!learning && f4buchberger_info && f4buchberger_info_position-1<f4buchberger_info->size()){
	B=(*f4buchberger_info)[f4buchberger_info_position-1].B;
	G=(*f4buchberger_info)[f4buchberger_info_position-1].G;
	continue;
      }
      if (learning && f4buchberger_info){
	f4buchberger_info->back().B=B;
	f4buchberger_info->back().G=G;
	f4buchberger_info->back().nonzero=added;
      }
#endif
	  unsigned debut = unsigned(G.size()) - added;
#if GBASIS_POSTF4BUCHBERGER>0
      if (added>GBASISF4_BUCHBERGER){
	// final interreduce
	vector<unsigned> G1(G.begin(),G.begin()+debut);
	vector<unsigned> G2(G.begin()+debut,G.end());
	vector<int> permu2;
	if (!learning && f4buchberger_info){
	  const info_t & info=(*f4buchberger_info)[f4buchberger_info_position-1];
	  rref_f4buchbergermod_nointerreduce(res,G1,res,G2,-1,info.quo2,info.R2,env,permu2);
	}
	else {
	  information.R2.order=TMP1.order;
	  information.R2.dim=TMP1.dim;
	  TMP1.coord.clear();
	  if (debug_infolevel>1)
	    CERR << CLOCK()*1e-6 << " collect monomials from old basis" << '\n';
	  collect(res,G1,TMP1); // collect all monomials in res[G[0..debut-1]]
	  // in_heap_reducemod(TMP1,res,G2,-1,info_tmp.quo2,TMP2,&info_tmp.R2,env);
	  in_heap_reducemod(TMP1,res,G2,-1,information.quo2,TMP2,&information.R2,env);
	  rref_f4buchbergermod_nointerreduce(res,G1,res,G2,-1,information.quo2,information.R2,env,permu2);
	  if (f4buchberger_info){
	    info_t & i=f4buchberger_info->back();
	    swap(i.quo2,information.quo2);
	    swap(i.R2.coord,information.R2.coord);
	    i.R2.order=TMP1.order;
	    i.R2.dim=TMP1.dim;
	  }
	}
      }
#endif
      // CERR << "finish loop G.size "<<G.size() << '\n';
      // CERR << added << '\n';
    }
#if GBASIS_POSTF4BUCHBERGER==0
    // final interreduce step2
    for (unsigned j=0; j<G.size();++j){
      reducesmallmod(res[G[j]],res,G,j,env,TMP2,true);
    }
#endif
    if (ressize<res.size())
      res.resize(ressize);
    if (debug_infolevel>1){
      unsigned t=0;
      for (unsigned i=0;i<res.size();++i)
	t += unsigned(res[i].coord.size());
      CERR << CLOCK()*1e-6 << " total number of monomials in res " << t << '\n';
      CERR << "Number of monomials cleared " << cleared << '\n';
    }
    // sort(res.begin(),res.end(),tripolymod<tdeg_t>);
    return true;
  }
#endif // GBASISF4_BUCHBERGER

  template<class tdeg_t>
  bool in_gbasisf4buchbergermod(vectpoly8<tdeg_t> & res8,vectpolymod<tdeg_t> &res,vector<unsigned> & G,modint env,bool totdeg,vector< paire > * pairs_reducing_to_zero,vector< info_t<tdeg_t> > * f4buchberger_info,bool recomputeR){
    convert(res8,res,env);
    unsigned ressize = unsigned(res8.size());
    bool b=in_gbasisf4buchbergermod(res,ressize,G,env,totdeg,pairs_reducing_to_zero,f4buchberger_info,recomputeR);
    convert(res,res8,env);
    return b;
  }

  // set P mod p*q to be chinese remainder of P mod p and Q mod q
  template<class tdeg_t>
  bool chinrem(poly8<tdeg_t> &P,const gen & pmod,poly8<tdeg_t> & Q,const gen & qmod,poly8<tdeg_t> & tmp){
    gen u,v,d,pqmod(pmod*qmod);
    egcd(pmod,qmod,u,v,d);
    if (u.type==_ZINT && qmod.type==_INT_)
      u=modulo(*u._ZINTptr,qmod.val);
    if (d==-1){ u=-u; v=-v; d=1; }
    if (d!=1)
      return false;
    int qmodval=0,U=0;
    mpz_t tmpz;
    mpz_init(tmpz);
    if (qmod.type==_INT_ && u.type==_INT_ && pmod.type==_ZINT){
      qmodval=qmod.val;
      U=u.val;
    }
    typename vector< T_unsigned<gen,tdeg_t> >::iterator it=P.coord.begin(),itend=P.coord.end(),jt=Q.coord.begin(),jtend=Q.coord.end();
    if (P.coord.size()==Q.coord.size()){
#ifndef USE_GMP_REPLACEMENTS
      if (qmodval){
	for (;it!=itend;++it,++jt){
	  if (it->u!=jt->u || jt->g.type!=_INT_)
	    break;
	}
	if (it==itend){
	  for (it=P.coord.begin(),jt=Q.coord.begin();it!=itend;++jt,++it){
	    if (it->g.type==_ZINT){
	      mpz_set_si(tmpz,jt->g.val);
	      mpz_sub(tmpz,tmpz,*it->g._ZINTptr);
	      mpz_mul_si(tmpz,*pmod._ZINTptr,(longlong(U)*modulo(tmpz,qmodval))%qmodval);
	      mpz_add(*it->g._ZINTptr,*it->g._ZINTptr,tmpz);
	    }
	    else {
	      mpz_mul_si(tmpz,*pmod._ZINTptr,(U*(longlong(jt->g.val)-it->g.val))%qmodval);
	      if (it->g.val>=0)
		mpz_add_ui(tmpz,tmpz,it->g.val);
	      else
		mpz_sub_ui(tmpz,tmpz,-it->g.val);
	      it->g=tmpz;
	    }
	  }
	  return true;
	}
	else {
	  if (debug_infolevel)
	    CERR << "warning chinrem: exponent mismatch " << it->u << "," << jt->u << '\n';
	}
      }
#endif
    }
    else {
      if (debug_infolevel)
	CERR << "warning chinrem: sizes differ " << P.coord.size() << "," << Q.coord.size() << '\n';
    }
    tmp.coord.clear(); tmp.dim=P.dim; tmp.order=P.order;
    tmp.coord.reserve(P.coord.size()+3); // allow 3 more terms in Q without realloc
    for (it=P.coord.begin(),jt=Q.coord.begin();it!=itend && jt!=jtend;){
      if (it->u==jt->u){
	gen g;
#ifndef USE_GMP_REPLACEMENTS
	if (qmodval && jt->g.type==_INT_){
	  if (it->g.type==_ZINT){
	    mpz_set_si(tmpz,jt->g.val);
	    mpz_sub(tmpz,tmpz,*it->g._ZINTptr);
	    mpz_mul_si(tmpz,*pmod._ZINTptr,(longlong(U)*modulo(tmpz,qmodval))%qmodval);
	    mpz_add(tmpz,tmpz,*it->g._ZINTptr);
	  }
	  else {
	    mpz_mul_si(tmpz,*pmod._ZINTptr,(U*(longlong(jt->g.val)-it->g.val))%qmodval);
	    if (it->g.val>=0)
	      mpz_add_ui(tmpz,tmpz,it->g.val);
	    else
	      mpz_sub_ui(tmpz,tmpz,-it->g.val);
	  }
	  g=tmpz;
	}
	else
#endif
	  g=it->g+u*(jt->g-it->g)*pmod;
	tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),it->u));
	++it; ++jt;
	continue;
      }
      if (tdeg_t_strictly_greater(it->u,jt->u,P.order)){
	if (debug_infolevel)
	  CERR << "chinrem: exponent mismatch using first " << '\n';
	gen g=it->g-u*(it->g)*pmod;
	tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),it->u));
	++it;
      }
      else {
	if (debug_infolevel)
	  CERR << "chinrem: exponent mismatch using second " << '\n';
	gen g=u*(jt->g)*pmod;
	tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),jt->u));
	++jt;
      }
    }
    if (it!=itend && debug_infolevel)
      CERR << "chinrem (gen): exponent mismatch at end using first, # " << itend-it << '\n';
    for (;it!=itend;++it){
      gen g=it->g-u*(it->g)*pmod;
      tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),it->u));
    }
    if (jt!=jtend && debug_infolevel)
      CERR << "chinrem (gen): exponent mismatch at end using second " << jtend-jt << '\n';
    for (;jt!=jtend;++jt){
      gen g=u*(jt->g)*pmod;
      tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),jt->u));
    }
    swap(P.coord,tmp.coord);
    mpz_clear(tmpz);
    return true;
  }

  // set P mod p*q to be chinese remainder of P mod p and Q mod q
  // P and Q must have same leading monomials,
  // otherwise returns 0 and leaves P unchanged
  template<class tdeg_t>
  int chinrem(vectpoly8<tdeg_t> &P,const gen & pmod,vectpoly8<tdeg_t> & Q,const gen & qmod,poly8<tdeg_t> & tmp){
    if (P.size()!=Q.size())
      return 0;
    for (unsigned i=0;i<P.size();++i){
      if (P[i].coord.front().u!=Q[i].coord.front().u)
	return 0;
    }
    // LP(P)==LP(Q), proceed to chinese remaindering
    for (unsigned i=0;i<P.size();++i){
      if (!chinrem(P[i],pmod,Q[i],qmod,tmp))
	return -1;
    }
    return 1;
  }


  // set P mod p*q to be chinese remainder of P mod p and Q mod q
  template<class tdeg_t>
  bool chinrem(poly8<tdeg_t> &P,const gen & pmod,const polymod<tdeg_t> & Q,int qmodval,poly8<tdeg_t> & tmp){
    gen u,v,d,pqmod(qmodval*pmod);
    egcd(pmod,qmodval,u,v,d);
    if (u.type==_ZINT)
      u=modulo(*u._ZINTptr,qmodval);
    if (d==-1){ u=-u; v=-v; d=1; }
    if (d!=1)
      return false;
    int U=u.val;
    mpz_t tmpz;
    mpz_init(tmpz);
    typename vector< T_unsigned<gen,tdeg_t> >::iterator it=P.coord.begin(),itend=P.coord.end();
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=Q.coord.begin(),jtend=Q.coord.end();
#ifndef USE_GMP_REPLACEMENTS
    if (P.coord.size()==Q.coord.size()){
      for (;it!=itend;++it,++jt){
	if (it->u!=jt->u)
	  break;
      }
      if (it==itend){
	for (it=P.coord.begin(),jt=Q.coord.begin();it!=itend;++jt,++it){
	  if (pmod.type!=_ZINT){
	    it->g=it->g+u*(jt->g-it->g)*pmod;
	    continue;
	  }
	  if (it->g.type==_ZINT){
#if 1
	    int amodq=modulo(*it->g._ZINTptr,qmodval);
	    if (amodq==jt->g)
	      continue;
	    mpz_mul_si(tmpz,*pmod._ZINTptr,(U*(jt->g-longlong(amodq)))%qmodval);
	    mpz_add(*it->g._ZINTptr,*it->g._ZINTptr,tmpz);
#else
	    mpz_set_si(tmpz,jt->g);
	    mpz_sub(tmpz,tmpz,*it->g._ZINTptr);
	    mpz_mul_si(tmpz,*pmod._ZINTptr,(longlong(U)*modulo(tmpz,qmodval))%qmodval);
	    mpz_add(*it->g._ZINTptr,*it->g._ZINTptr,tmpz);
#endif
	  }
	  else {
	    mpz_mul_si(tmpz,*pmod._ZINTptr,(U*(longlong(jt->g)-it->g.val))%qmodval);
	    if (it->g.val>=0)
	      mpz_add_ui(tmpz,tmpz,it->g.val);
	    else
	      mpz_sub_ui(tmpz,tmpz,-it->g.val);
	    it->g=tmpz;
	  }
	}
	mpz_clear(tmpz);
	return true;
      }
    }
#endif
    tmp.coord.clear(); tmp.dim=P.dim; tmp.order=P.order;
    tmp.coord.reserve(P.coord.size()+3); // allow 3 more terms in Q without realloc
    for (it=P.coord.begin(),jt=Q.coord.begin();it!=itend && jt!=jtend;){
      if (it->u==jt->u){
	gen g;
#ifndef USE_GMP_REPLACEMENTS
	if (pmod.type==_ZINT){
	  if (it->g.type==_ZINT){
	    mpz_set_si(tmpz,jt->g);
	    mpz_sub(tmpz,tmpz,*it->g._ZINTptr);
	    mpz_mul_si(tmpz,*pmod._ZINTptr,(longlong(U)*modulo(tmpz,qmodval))%qmodval);
	    mpz_add(tmpz,tmpz,*it->g._ZINTptr);
	  }
	  else {
	    mpz_mul_si(tmpz,*pmod._ZINTptr,(U*(longlong(jt->g)-it->g.val))%qmodval);
	    if (it->g.val>=0)
	      mpz_add_ui(tmpz,tmpz,it->g.val);
	    else
	      mpz_sub_ui(tmpz,tmpz,-it->g.val);
	  }
	  g=tmpz;
	}
	else
#endif
	  g=it->g+u*(jt->g-it->g)*pmod;
	tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),it->u));
	++it; ++jt;
	continue;
      }
      if (tdeg_t_strictly_greater(it->u,jt->u,P.order)){
	if (debug_infolevel)
	  CERR << "chinrem: exponent mismatch using first " << '\n';
	gen g=it->g-u*(it->g)*pmod;
	tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),it->u));
	++it;
      }
      else {
	if (debug_infolevel)
	  CERR << "chinrem: exponent mismatch using second " << '\n';
	gen g=u*((jt->g)*pmod);
	tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),jt->u));
	++jt;
      }
    }
    if (it!=itend && debug_infolevel)
      CERR << "chinrem (int): exponent mismatch at end using first #" << itend-it << '\n';
    for (;it!=itend;++it){
      gen g=it->g-u*(it->g)*pmod;
      tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),it->u));
    }
    if (jt!=jtend && debug_infolevel)
      CERR << "chinrem (int): exponent mismatch at end using second #" << jtend-jt << '\n';
    for (;jt!=jtend;++jt){
      gen g=u*((jt->g)*pmod);
      tmp.coord.push_back(T_unsigned<gen,tdeg_t>(smod(g,pqmod),jt->u));
    }
    swap(P.coord,tmp.coord);
    mpz_clear(tmpz);
    return true;
  }

  // set P mod p*q to be chinese remainder of P mod p and Q mod q
  // P and Q must have same leading monomials,
  // otherwise returns 0 and leaves P unchanged
  template<class tdeg_t>
  int chinrem(vectpoly8<tdeg_t> &P,const gen & pmod,const vectpolymod<tdeg_t> & Q,int qmod,poly8<tdeg_t> & tmp,int start=0){
    if (P.size()!=Q.size())
      return 0;
    for (unsigned i=start;i<P.size();++i){
      if (P[i].coord.empty() && Q[i].coord.empty())
	continue;
      if (P[i].coord.empty())
	return 0;
      if (Q[i].coord.empty())
	return 0;
      if (P[i].coord.front().u!=Q[i].coord.front().u)
	return 0;
    }
    // LP(P)==LP(Q), proceed to chinese remaindering
    for (unsigned i=start;i<P.size();++i){
      if (!chinrem(P[i],pmod,Q[i],qmod,tmp))
	return -1;
    }
    return 1;
  }

  // a mod b = r/u with n and d<sqrt(b)/2
  // a*u = r mod b -> a*u+b*v=r, Bezout with a and b
  bool fracmod(int a,int b,int & n,int & d){
    if (a<0){
      if (!fracmod(-a,b,n,d))
	return false;
      n=-n;
      return true;
    }
    int r=b,u=0; // v=1
    int r1=a,u1=1,r2,u2,q; // v1=0
    for (;double(2*r1)*r1>b;){
      q=r/r1;
      u2=u-q*u1;
      r2=r-q*r1;
      u=u1;
      u1=u2;
      r=r1;
      r1=r2;
    }
    if (double(2*u1)*u1>b)
      return false;
    if (u1<0){ u1=-u1; r1=-r1; }
    n=r1; d=u1;
    return true;
  }

  // search for d such that d*P mod p has small coefficients
  // call with d set to 1,
  template<class tdeg_t>
  bool findmultmod(const poly8<tdeg_t> & P,int p,int & d){
    int n,s=int(P.coord.size());
    for (int i=0;i<s;++i){
      int a=smod(longlong(P.coord[i].g.val)*d,p);
      if (double(2*a)*a<p)
	continue;
      int d1=1;
      if (!fracmod(a,p,n,d1) || double(2*d1)*d1>p){
	if (debug_infolevel)
	  COUT << "findmultmod failure " << a << " mod " << p << '\n';
	return false;
      }
      d=d*d1;
    }
    if (debug_infolevel){
      for (int i=0;i<s;++i){
	int a=smod(longlong(P.coord[i].g.val)*d,p);
	if (double(2*a)*a>=p){
	  COUT << "possible findmultmod failure " << P.coord[i].g.val << " " << d << " " << a << " " << p << '\n';
	  //return false;
	}
      }
    }
    return true;
  }

  template<class tdeg_t>
  bool fracmod(const poly8<tdeg_t> &P,const gen & p,
	       mpz_t & d,mpz_t & d1,mpz_t & absd1,mpz_t &u,mpz_t & u1,mpz_t & ur,mpz_t & q,mpz_t & r,mpz_t &sqrtm,mpz_t & tmp,
	       poly8<tdeg_t> & Q){
    Q.coord.clear();
    Q.coord.reserve(P.coord.size());
    Q.dim=P.dim;
    Q.order=P.order;
    Q.sugar=P.sugar;
    gen L=1;
    bool tryL=true;
    for (unsigned i=0;i<P.coord.size();++i){
      gen g=P.coord[i].g,num,den;
      if (g.type==_INT_)
	g.uncoerce();
      if ( (g.type!=_ZINT) || (p.type!=_ZINT) ){
	CERR << "bad type"<<'\n';
	return false;
      }
      if (tryL && L.type==_ZINT){
	num=smod(L*g,p);
	if (is_greater(p,4*num*num,context0)){
	  g=fraction(num,L);
	  Q.coord.push_back(T_unsigned<gen,tdeg_t>(g,P.coord[i].u));
	  continue;
	}
      }
      if (!in_fracmod(p,g,d,d1,absd1,u,u1,ur,q,r,sqrtm,tmp,num,den))
	return false;
      if (num.type==_ZINT && mpz_sizeinbase(*num._ZINTptr,2)<=30)
	num=int(mpz_get_si(*num._ZINTptr));
      if (den.type==_ZINT && mpz_sizeinbase(*den._ZINTptr,2)<=30)
	den=int(mpz_get_si(*den._ZINTptr));
      if (!is_positive(den,context0)){ // ok
	den=-den;
	num=-num;
      }
      g=fraction(num,den);
      if (tryL){
	L=lcm(L,den);
	tryL=is_greater(p,L*L,context0);
      }
      Q.coord.push_back(T_unsigned<gen,tdeg_t>(g,P.coord[i].u));
    }
    return true;
  }

  template<class tdeg_t>
  bool fracmod(const vectpoly8<tdeg_t> & P,const gen & p_,
		      mpz_t & d,mpz_t & d1,mpz_t & absd1,mpz_t &u,mpz_t & u1,mpz_t & ur,mpz_t & q,mpz_t & r,mpz_t &sqrtm,mpz_t & tmp,
		      vectpoly8<tdeg_t> & Q){
    Q.resize(P.size());
    gen p=p_;
    if (p.type==_INT_)
      p.uncoerce();
    bool ok=true;
    for (unsigned i=0;i<P.size();++i){
      if (!fracmod(P[i],p,d,d1,absd1,u,u1,ur,q,r,sqrtm,tmp,Q[i])){
	ok=false;
	break;
      }
    }
    return ok;
  }

  template <class tdeg_t>
  void cleardeno(poly8<tdeg_t> &P){
    gen g=1;
    for (unsigned i=0;i<P.coord.size();++i){
      if (P.coord[i].g.type==_FRAC)
	g=lcm(g,P.coord[i].g._FRACptr->den);
    }
    if (g!=1){
      for (unsigned i=0;i<P.coord.size();++i){
	P.coord[i].g=g*P.coord[i].g;
      }
    }
  }

  template<class tdeg_t>
  void cleardeno(vectpoly8<tdeg_t> & P){
    for (unsigned i=0;i<P.size();++i){
      cleardeno(P[i]);
      if (debug_infolevel){
	if (i%10==9){
	  COUT << "+";
	  COUT.flush();
	}
	if (i%500==499)
	  COUT << " " << CLOCK()*1e-6 << " remaining " << P.size()-i << '\n';
      }
    }
  }


  template<class tdeg_t>
  void collect(const vectpoly8<tdeg_t> & f4buchbergerv,polymod<tdeg_t> & allf4buchberger){
    typename vectpoly8<tdeg_t>::const_iterator it=f4buchbergerv.begin(),itend=f4buchbergerv.end();
    vector<heap_tt<tdeg_t> > H;
    H.reserve(itend-it);
    order_t keyorder={_REVLEX_ORDER,0};
    for (unsigned i=0;it!=itend;++i,++it){
      keyorder=it->order;
      if (!it->coord.empty())
	H.push_back(heap_tt<tdeg_t>(i,0,it->coord.front().u));
    }
    compare_heap_tt<tdeg_t> key(keyorder);
    make_heap(H.begin(),H.end(),key);
    while (!H.empty()){
      std::pop_heap(H.begin(),H.end(),key);
      // push root node of the heap in allf4buchberger
      heap_tt<tdeg_t> & current =H.back();
      if (allf4buchberger.coord.empty() || allf4buchberger.coord.back().u!=current.u)
	allf4buchberger.coord.push_back(T_unsigned<modint,tdeg_t>(1,current.u));
      ++current.polymodpos;
      if (current.polymodpos>=f4buchbergerv[current.f4buchbergervpos].coord.size()){
	H.pop_back();
	continue;
      }
      current.u=f4buchbergerv[current.f4buchbergervpos].coord[current.polymodpos].u;
      std::push_heap(H.begin(),H.end(),key);
    }
  }

  template<class tdeg_t>
  void makeline(const poly8<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vecteur & v){
    v=vecteur(R.coord.size(),0);
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=R.coord.begin(),jtbeg=jt,jtend=R.coord.end();
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v[jt-jtbeg]=it->g;
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v[jt-jtbeg]=it->g;
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  unsigned firstnonzero(const vecteur & v){
    for (unsigned i=0;i<v.size();++i){
      if (v[i]!=0)
	return i;
    }
    return unsigned(v.size());
  }

  struct sparse_gen {
    gen val;
    unsigned pos;
    sparse_gen(const gen & v,unsigned u):val(v),pos(u){};
    sparse_gen():val(0),pos(-1){};
  };

  template<class tdeg_t>
  void makeline(const poly8<tdeg_t> & p,const tdeg_t * shiftptr,const polymod<tdeg_t> & R,vector<sparse_gen> & v){
    typename std::vector< T_unsigned<gen,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=R.coord.begin(),jtend=R.coord.end();
    if (shiftptr){
      for (;it!=itend;++it){
	tdeg_t u=it->u+*shiftptr;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v.push_back(sparse_gen(it->g,int(jt-R.coord.begin())));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=it->u;
	for (;jt!=jtend;++jt){
	  if (jt->u==u){
	    v.push_back(sparse_gen(it->g,int(jt-R.coord.begin())));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

#ifdef x86_64
  bool checkreducef4buchberger_64(vector<modint> &v,vector<modint> & coeff,const vector< vector<sparse_element> > & M,modint env,vector<int128_t> & w){
    w.resize(v.size());
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    vector<int128_t>::iterator wt=w.begin();
    for (;vt!=vtend;++wt,++vt){
      *wt=*vt;
    }
    for (unsigned i=0;i<M.size();++i){
      const vector<sparse_element> & m=M[i];
      const sparse_element * it=&m.front(),*itend=it+m.size(),*it2;
      if (it==itend)
	continue;
      int128_t & ww=w[it->pos];
      if (ww==0){
	coeff[i]=0;
	continue;
      }
      modint c=coeff[i]=(modint2(invmod(it->val,env))*ww)%env;
      // CERR << "multiplier ok line " << i << " value " << c << " " << w << '\n';
      if (!c)
	continue;
      ww=0;
      ++it;
      it2=itend-8;
      for (;it<=it2;){
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
	w[it->pos] -= modint2(c)*(it->val);
	++it;
      }
      for (;it!=itend;++it){
	w[it->pos] -= modint2(c)*(it->val);
      }
    }
    for (vt=v.begin(),wt=w.begin();vt!=vtend;++wt,++vt){
      if (*wt && (*wt % env))
	return false;
    }
    return true;
  }

  bool checkreducef4buchbergersplit_64(vector<modint> &v,vector<modint> & coeff,const vector< vector<shifttype> > & M,vector<vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env,vector<int128_t> & w){
    w.resize(v.size());
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    vector<int128_t>::iterator wt=w.begin(),wtend=w.end();
    for (;vt!=vtend;++wt,++vt){
      *wt=*vt;
    }
    for (unsigned i=0;i<M.size();++i){
      const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
      vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end();
      if (jt==jtend)
	continue;
      const vector<shifttype> & mindex=M[i];
      const shifttype * it=&mindex.front();
      unsigned pos=0;
      next_index(pos,it);
      // if (pos>v.size()) CERR << "error" <<'\n';
      modint c=coeff[i]=(modint2(invmod(*jt,env))*w[pos])%env;
      w[pos]=0;
      if (!c)
	continue;
      for (++jt;jt!=jtend;++jt){
#ifdef GIAC_SHORTSHIFTTYPE
	next_index(pos,it);
	int128_t &x=w[pos];
	x -= modint2(c)*(*jt);
#else
	w[*it] -= modint2(c)*(*jt);
	++it;
#endif
      }
    }
    for (wt=w.begin();wt!=wtend;++wt){
      if (*wt % env)
	return false;
    }
    return true;
  }

#endif

  // return true if v reduces to 0
  // in addition to reducef4buchberger, compute the coeffs
  bool checkreducef4buchberger(vector<modint> &v,vector<modint> & coeff,const vector< vector<sparse_element> > & M,modint env){
    for (unsigned i=0;i<M.size();++i){
      const vector<sparse_element> & m=M[i];
      vector<sparse_element>::const_iterator it=m.begin(),itend=m.end(),it1=itend-8;
      if (it==itend)
	continue;
      modint c=coeff[i]=v[it->pos];
      if (!c)
	continue;
      c=coeff[i]=(modint2(invmod(it->val,env))*c)%env;
      v[it->pos]=0;
      for (++it;it<it1;){
	modint *x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
	x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
	x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
	x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
	x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
	x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
	x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
	x=&v[it->pos];
	*x=(*x-modint2(c)*(it->val))%env;
	++it;
      }
      for (;it!=itend;++it){
	modint &x=v[it->pos];
	x=(x-modint2(c)*(it->val))%env;
      }
    }
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return false;
    }
    return true;
  }

  // return true if v reduces to 0
  // in addition to reducef4buchberger, compute the coeffs
  bool checkreducef4buchbergersplit(vector<modint> &v,vector<modint> & coeff,const vector< vector<shifttype> > & M,vector<vector<modint> > & coeffs,vector<coeffindex_t> & coeffindex,modint env){
    for (unsigned i=0;i<M.size();++i){
      const vector<modint> & mcoeff=coeffs[coeffindex[i].u];
      vector<modint>::const_iterator jt=mcoeff.begin(),jtend=mcoeff.end();
      if (jt==jtend)
	continue;
      const vector<shifttype> & mindex=M[i];
      const shifttype * it=&mindex.front();
      unsigned pos=0;
      next_index(pos,it);
      // if (pos>v.size()) CERR << "error" <<'\n';
      modint c=coeff[i]=(modint2(invmod(*jt,env))*v[pos])%env;
      v[pos]=0;
      if (!c)
	continue;
      for (++jt;jt!=jtend;++jt){
#ifdef GIAC_SHORTSHIFTTYPE
	next_index(pos,it);
	modint &x=v[pos];
#else
	modint &x=v[*it];
	++it;
#endif
	x=(x-modint2(c)*(*jt))%env;
      }
    }
    vector<modint>::iterator vt=v.begin(),vtend=v.end();
    for (vt=v.begin();vt!=vtend;++vt){
      if (*vt)
	return false;
    }
    return true;
  }

  // Find x=a mod amod and =b mod bmod
  // We have x=a+A*amod=b+B*Bmod
  // hence A*amod-B*bmod=b-a
  // let u*amod+v*bmod=1
  // then A=(b-a)*u is a solution
  // hence x=a+(b-a)*u*amod mod (amod*bmod) is the solution
  // hence x=a+((b-a)*u mod bmod)*amod
  static bool ichinrem_inplace(matrice & a,const vector< vector<modint> > &b,const gen & amod, int bmod){
    gen U,v,d;
    egcd(amod,bmod,U,v,d);
    if (!is_one(d) || U.type!=_ZINT)
      return false;
    int u=mpz_get_si(*U._ZINTptr);
    longlong q;
    for (unsigned i=0;i<a.size();++i){
      gen * ai = &a[i]._VECTptr->front(), * aiend=ai+a[i]._VECTptr->size();
      const modint * bi = &b[i].front();
      for (;ai!=aiend;++bi,++ai){
	if (*bi==0 && ai->type==_INT_ && ai->val==0)
	  continue;
	q=longlong(*bi)-(ai->type==_INT_?ai->val:modulo(*ai->_ZINTptr,bmod));
	q=(q*u) % bmod;
	if (amod.type==_ZINT && ai->type==_ZINT){
	  if (q>=0)
	    mpz_addmul_ui(*ai->_ZINTptr,*amod._ZINTptr,int(q));
	  else
	    mpz_submul_ui(*ai->_ZINTptr,*amod._ZINTptr,-int(q));
	}
	else
	  *ai += int(q)*amod;
      }
    }
    return true;
  }

  template<class tdeg_t>
  gen linfnorm(const poly8<tdeg_t> & p,GIAC_CONTEXT){
    gen B=0;
    for (unsigned i=0;i<p.coord.size();++i){
      gen b=abs(p.coord[i].g,contextptr);
      if (is_strictly_greater(b,B,contextptr))
	B=b;
    }
    return B;
  }

  template<class tdeg_t>
  gen linfnorm(const vectpoly8<tdeg_t> & v,GIAC_CONTEXT){
    gen B=0;
    for (unsigned i=0;i<v.size();++i){
      gen b=linfnorm(v[i],contextptr);
      if (is_strictly_greater(b,B,contextptr))
	B=b;
    }
    return B;
  }

  bool chk_equal_mod(const gen & a,const vector<int> & p,int m){
    if (a.type!=_VECT || a._VECTptr->size()!=p.size())
      return false;
    const_iterateur it=a._VECTptr->begin(),itend=a._VECTptr->end();
    vector<int>::const_iterator jt=p.begin();
    for (;it!=itend;++jt,++it){
      if (it->type==_INT_ && it->val==*jt) continue;
      if (!chk_equal_mod(*it,*jt,m))
	return false;
    }
    return true;
  }

  bool chk_equal_mod(const vecteur & v,const vector< vector<int> >& p,int m){
    if (v.size()!=p.size())
      return false;
    for (unsigned i=0;i<p.size();++i){
      if (!chk_equal_mod(v[i],p[i],m))
	return false;
    }
    return true;
  }

  template<class tdeg_t>
  bool chk_equal_mod(const poly8<tdeg_t> & v,const polymod<tdeg_t> & p,int m){
    // FIXME: sizes may differ if a coeff of v is 0 mod m
    if (v.coord.size()!=p.coord.size())
      return false;
    if (p.coord.empty())
      return true;
	unsigned s = unsigned(p.coord.size());
    int lc=smod(v.coord[0].g,m).val;
    int lcp=p.coord[0].g;
    if (lcp!=1){
      for (unsigned i=0;i<s;++i){
	if (!chk_equal_mod(lcp*v.coord[i].g,(longlong(lc)*p.coord[i].g)%m,m))
	  return false;
      }
    }
    else {
      for (unsigned i=0;i<s;++i){
	if (!chk_equal_mod(v.coord[i].g,(longlong(lc)*p.coord[i].g)%m,m))
	  return false;
      }
    }
    return true;
  }

  template<class tdeg_t>
  bool chk_equal_mod(const vectpoly8<tdeg_t> & v,const vectpolymod<tdeg_t> & p,const vector<unsigned> & G,int m){
    if (v.size()!=G.size())
      return false;
    for (unsigned i=0;i<G.size();++i){
      if (!chk_equal_mod(v[i],p[G[i]],m))
	return false;
    }
    return true;
  }

  template<class tdeg_t>
  bool chk_equal_mod(const poly8<tdeg_t> & v,const poly8<tdeg_t> & p,int m){
    if (v.coord.size()!=p.coord.size())
      return false;
    unsigned s=unsigned(p.coord.size());
    int lc=smod(v.coord[0].g,m).val;
    for (unsigned i=0;i<s;++i){
      if (!chk_equal_mod(v.coord[i].g,(longlong(lc)*p.coord[i].g.val)%m,m))
	return false;
    }
    return true;
  }

  template<class tdeg_t>
  bool chk_equal_mod(const vectpoly8<tdeg_t> & v,const vectpoly8<tdeg_t> & p,const vector<unsigned> & G,int m){
    if (v.size()!=G.size())
      return false;
    for (unsigned i=0;i<G.size();++i){
      if (!chk_equal_mod(v[i],p[G[i]],m))
	return false;
    }
    return true;
  }

  // excluded==-1 and G==identity in calls
  // if eps>0 the check is probabilistic
  template<class tdeg_t>
  bool checkf4buchberger(vectpoly8<tdeg_t> & f4buchbergerv,const vectpoly8<tdeg_t> & res,vector<unsigned> & G,unsigned excluded,double eps){
    if (f4buchbergerv.empty())
      return true;
    polymod<tdeg_t> allf4buchberger(f4buchbergerv.front().order,f4buchbergerv.front().dim),rem(allf4buchberger);
    vectpolymod<tdeg_t> resmod,quo;
    convert(res,resmod,0);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " checkf4buchberger begin collect monomials on #polys " << f4buchbergerv.size() << '\n';
    // collect all terms in f4buchbergerv
    collect(f4buchbergerv,allf4buchberger);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " checkf4buchberger symbolic preprocess" << '\n';
    // find all monomials required to reduce allf4buchberger with res[G[.]]
    polymod<tdeg_t> R;
    in_heap_reducemod(allf4buchberger,resmod,G,excluded,quo,rem,&R,0);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " checkf4buchberger end symbolic preprocess" << '\n';
    // build a matrix with rows res[G[.]]*quo[.] in terms of monomials in allf4buchberger
    // sort the matrix
    // checking reduction to 0 is equivalent to
    // write a line from f4buchbergerv[]
    // as a linear combination of the lines of this matrix
    // we will do that modulo a list of primes
    // and keep track of the coefficients of the linear combination (the quotients)
    // we reconstruct the quotients in Q by fracmod
    // once they stabilize, we compute the lcm l of the denominators
    // we multiply by l to have an equality on Z
    // we compute bounds on the coefficients of the products res*quo
    // and on l*f4buchbergerv, and we check further the equality modulo additional
    // primes until the equality is proved
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin build M" << '\n';
    vector< vector<sparse_gen> > M;
    vector<sparse_element> atrier;
    unsigned N=unsigned(R.coord.size()),i,j=0,nterms=0;
    M.reserve(N); // actual size is at most N (difference is the remainder part size)
    for (i=0;i<res.size();++i){
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      for (;jt!=jtend;++j,++jt){
	M.push_back(vector<sparse_gen>(0));
	makeline(res[G[i]],&jt->u,R,M[j]);
	nterms += unsigned(M[j].size());
	atrier.push_back(sparse_element(M[j].front().pos,j));
      }
    }
    sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
    vector< vector<sparse_gen> > M1(atrier.size());
    for (i=0;i<atrier.size();++i){
      swap(M1[i],M[atrier[i].pos]);
    }
    swap(M,M1);
    // CERR << M << '\n';
    if (debug_infolevel>0)
      CERR << CLOCK()*1e-6 << " rows, columns, terms: " << M.size() << "x" << N << "=" << nterms << '\n';
    // PSEUDO_MOD is not interesting here since there is no inter-reduction
    gen p(int(longlong(1<<31)-1));
    gen pip(1);
    vectpolymod<tdeg_t> f4buchbergervmod;
    matrice coeffmat;
    vector< vector<modint> > coeffmatmodp(f4buchbergerv.size(),vector<modint>(M.size()));
    gen bres=linfnorm(res,context0);
    gen bf4buchberger=linfnorm(f4buchbergerv,context0);
    matrice prevmatq;
    bool stable=false;
    gen bound=0;
    for (int iter=0;;++iter){
      if (eps>0 && is_greater(eps*pip,1,context0))
	return true;
      p=prevprime(p-1);
      int env=p.val;
      // check that p does not divide a leading monomial in M
      unsigned j;
      for (j=0;j<M.size();++j){
	if (smod(M[j].front().val,p)==0)
	  break;
      }
      if (j<M.size())
	continue;
      // compute M mod p
      vector< vector<sparse_element> > Mp(M.size());
      for (unsigned i=0;i<M.size();++i){
	const vector<sparse_gen> & Mi=M[i];
	vector<sparse_element> Ni;
	Ni.reserve(Mi.size());
	for (unsigned j=0;j<Mi.size();++j){
	  const sparse_gen & Mij=Mi[j];
	  modint tmp= Mij.val.type==_ZINT?modulo(*Mij.val._ZINTptr,env):Mij.val.val%env;
	  Ni.push_back(sparse_element(tmp,Mij.pos));
	}
	swap(Mp[i],Ni);
      }
      // reduce f4buchbergerv and stores coefficients of quotients for f4buchbergerv[i] in coeffmat[i]
      convert(f4buchbergerv,f4buchbergervmod,env);
      if (debug_infolevel>0)
	CERR << CLOCK()*1e-6 << " checking mod " << p << '\n';
      vector<modint> v;
      unsigned countres=0;
#ifdef x86_64
      vector<int128_t> v128;
#endif
      for (unsigned i=0;i<f4buchbergervmod.size();++i){
	makeline<tdeg_t>(f4buchbergervmod[i],0,R,v);
#if 0 // def x86_64
	if (!checkreducef4buchberger_64(v,coeffmatmodp[i],Mp,env,v128))
	  return false;
#else
	if (!checkreducef4buchberger(v,coeffmatmodp[i],Mp,env))
	  return false;
#endif
	if (iter==0){
	  unsigned countrescur=0;
	  vector<modint> & coeffi=coeffmatmodp[i];
	  for (unsigned j=0;j<coeffi.size();++j){
	    if (coeffi[j])
	      ++countrescur;
	  }
	  if (countrescur>countres)
	    countres=countrescur;
	}
      }
      // if (iter==0) bound=pow(bres,int(countres),context0);
      if (stable){
	if (!chk_equal_mod(prevmatq,coeffmatmodp,env))
	  stable=false;
	// if stable compute bounds and compare with product of primes
	// if 2*bounds < product of primes recheck stabilization and return true
	if (is_strictly_greater(pip,bound,context0)){
	  if (debug_infolevel>0)
	    CERR << CLOCK()*1e-6 << " modular check finished " << '\n';
	  return true;
	}
      }
      // combine coeffmat with previous one by chinese remaindering
      if (debug_infolevel>0)
	CERR << CLOCK()*1e-6 << " chinrem mod " << p << '\n';
      if (iter)
	ichinrem_inplace(coeffmat,coeffmatmodp,pip,p.val);
      else
	vectvector_int2vecteur(coeffmatmodp,coeffmat);
      pip=pip*p;
      if (is_greater(bound,pip,context0))
	continue;
      if (!stable){
	// check stabilization
	matrice checkquo;
	checkquo.reserve(coeffmat.size());
	for (unsigned k=0;k<coeffmat.size();++k){
	  if (prevmatq.size()>k && chk_equal_mod(prevmatq[k],coeffmatmodp[k],env))
	    checkquo.push_back(prevmatq[k]);
	  else
	    checkquo.push_back(fracmod(coeffmat[k],pip));
	  if (prevmatq.size()>k && checkquo[k]!=prevmatq[k])
	    break;
	  if (k>(prevmatq.size()*3)/2+2)
	    break;
	}
	if (checkquo!=prevmatq){
	  swap(prevmatq,checkquo);
	  if (debug_infolevel>0)
	    CERR << CLOCK()*1e-6 << " unstable mod " << p << " reconstructed " << prevmatq.size() << '\n';
	  continue;
	}
	matrice coeffmatq=*_copy(checkquo,context0)._VECTptr;
	if (debug_infolevel>0)
	  CERR << CLOCK()*1e-6 << " full stable mod " << p << '\n';
	stable=true;
	gen lall=1; vecteur l(coeffmatq.size());
	for (unsigned i=0;i<coeffmatq.size();++i){
	  lcmdeno(*coeffmatq[i]._VECTptr,l[i],context0);
	  if (is_strictly_greater(l[i],lall,context0))
	    lall=l[i];
	}
	if (debug_infolevel>0)
	  CERR << CLOCK()*1e-6 << " lcmdeno ok/start bound " << p << '\n';
	gen ball=1,bi; // ball is the max bound of all coeff in coeffmatq
	for (unsigned i=0;i<coeffmatq.size();++i){
	  bi=linfnorm(coeffmatq[i],context0);
	  if (is_strictly_greater(bi,ball,context0))
	    ball=bi;
	}
	// bound for res and f4buchbergerv
	bound=bres*ball;
	gen bound2=lall*bf4buchberger;
	// lcm of deno and max of coeff
	if (is_strictly_greater(bound2,bound,context0))
	  bound=bound2;
      }
    }
    return true;
  }


  // excluded==-1 and G==identity in calls
  // if eps>0 the check is probabilistic
  template<class tdeg_t>
  bool checkf4buchbergersplit(vectpoly8<tdeg_t> & f4buchbergerv,const vectpoly8<tdeg_t> & res,vector<unsigned> & G,unsigned excluded,double eps){
    if (f4buchbergerv.empty())
      return true;
    polymod<tdeg_t> allf4buchberger(f4buchbergerv.front().order,f4buchbergerv.front().dim),rem(allf4buchberger);
    vectpolymod<tdeg_t> resmod,quo;
    convert(res,resmod,0);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " checkf4buchberger split begin collect monomials on #polys " << f4buchbergerv.size() << '\n';
    // collect all terms in f4buchbergerv
    collect(f4buchbergerv,allf4buchberger);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " checkf4buchberger split symbolic preprocess" << '\n';
    // find all monomials required to reduce allf4buchberger with res[G[.]]
    polymod<tdeg_t> R;
    in_heap_reducemod(allf4buchberger,resmod,G,excluded,quo,rem,&R,0);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " checkf4buchberger split end symbolic preprocess" << '\n';
    // build a matrix with rows res[G[.]]*quo[.] in terms of monomials in allf4buchberger
    // sort the matrix
    // checking reduction to 0 is equivalent to
    // write a line from f4buchbergerv[]
    // as a linear combination of the lines of this matrix
    // we will do that modulo a list of primes
    // and keep track of the coefficients of the linear combination (the quotients)
    // we reconstruct the quotients in Q by fracmod
    // once they stabilize, we compute the lcm l of the denominators
    // we multiply by l to have an equality on Z
    // we compute bounds on the coefficients of the products res*quo
    // and on l*f4buchbergerv, and we check further the equality modulo additional
    // primes until the equality is proved
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin build Mcoeff/Mindex" << '\n';
    vector< vector<gen> > Mcoeff(G.size());
    vector<vector<shifttype> > Mindex;
    vector<coeffindex_t> coeffindex;
    vector<sparse_element> atrier;
    unsigned N=unsigned(R.coord.size()),i,j=0,nterms=0;
    Mindex.reserve(N);
    atrier.reserve(N);
    coeffindex.reserve(N);
    for (i=0;i<G.size();++i){
      Mcoeff[i].reserve(res[G[i]].coord.size());
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      for (;jt!=jtend;++j,++jt){
	Mindex.push_back(vector<shifttype>(0));
#ifdef GIAC_SHORTSHIFTTYPE
	Mindex[j].reserve(1+int(1.1*res[G[i]].coord.size()));
#else
	Mindex[j].reserve(res[G[i]].coord.size());
#endif
      }
    }
    for (i=0,j=0;i<G.size();++i){
      // copy coeffs of res[G[i]] in Mcoeff
      copycoeff(res[G[i]],Mcoeff[i]);
      // for each monomial of quo[i], find indexes and put in Mindex
      typename std::vector< T_unsigned<modint,tdeg_t> >::const_iterator jt=quo[i].coord.begin(),jtend=quo[i].coord.end();
      for (;jt!=jtend;++j,++jt){
	coeffindex.push_back(coeffindex_t(N<0xffff,i));
	makelinesplit(res[G[i]],&jt->u,R,Mindex[j]);
	atrier.push_back(sparse_element(first_index(Mindex[j]),j));
      }
    }
    sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
    vector< vector<shifttype> > Mindex1(atrier.size());
    vector<coeffindex_t> coeffindex1(atrier.size());
    for (i=0;i<atrier.size();++i){
      swap(Mindex1[i],Mindex[atrier[i].pos]);
      swap(coeffindex1[i],coeffindex[atrier[i].pos]);
    }
    swap(Mindex,Mindex1);
    swap(coeffindex,coeffindex1);
    // CERR << M << '\n';
    if (debug_infolevel>0)
      CERR << CLOCK()*1e-6 << " rows, columns, terms: " << Mindex.size() << "x" << N << "=" << nterms << '\n';
    // PSEUDO_MOD is not interesting here since there is no inter-reduction
    gen p(int(longlong(1<<31)-1));
    gen pip(1);
    vectpolymod<tdeg_t> f4buchbergervmod;
    matrice coeffmat;
    vector< vector<modint> > coeffmatmodp(f4buchbergerv.size(),vector<modint>(Mindex.size()));
    gen bres=linfnorm(res,context0);
    gen bf4buchberger=linfnorm(f4buchbergerv,context0);
    matrice prevmatq;
    bool stable=false;
    gen bound=0;
    vector< vector<modint> > Mcoeffp(Mcoeff.size());
    for (int iter=0;;++iter){
      if (eps>0 && is_greater(eps*pip,1,context0))
	return true;
      p=prevprime(p-1);
      int env=p.val;
      // check that p does not divide a leading monomial in M
      unsigned j;
      for (j=0;j<Mcoeff.size();++j){
	if (smod(Mcoeff[j].front(),p)==0)
	  break;
      }
      if (j<Mcoeff.size())
	continue;
      // compute Mcoeff mod p
      for (unsigned i=0;i<Mcoeff.size();++i){
	const vector<gen> & Mi=Mcoeff[i];
	vector<modint> & Ni=Mcoeffp[i];
	Ni.clear();
	Ni.reserve(Mi.size());
	for (unsigned j=0;j<Mi.size();++j){
	  const gen & Mij=Mi[j];
	  modint tmp= Mij.type==_ZINT?modulo(*Mij._ZINTptr,env):Mij.val%env;
	  Ni.push_back(tmp);
	}
      }
      // reduce f4buchbergerv and stores coefficients of quotients for f4buchbergerv[i] in coeffmat[i]
      convert(f4buchbergerv,f4buchbergervmod,env);
      if (debug_infolevel>0)
	CERR << CLOCK()*1e-6 << " checking mod " << p << '\n';
      vector<modint> v;
      unsigned countres=0;
#ifdef x86_64
      vector<int128_t> v128;
#endif
      for (unsigned i=0;i<f4buchbergervmod.size();++i){
	makeline(f4buchbergervmod[i],0,R,v);
#ifdef x86_64
	if (!checkreducef4buchbergersplit_64(v,coeffmatmodp[i],Mindex,Mcoeffp,coeffindex,env,v128))
	  return false;
#else
	if (!checkreducef4buchbergersplit(v,coeffmatmodp[i],Mindex,Mcoeffp,coeffindex,env))
	  return false;
#endif
	if (iter==0){
	  unsigned countrescur=0;
	  vector<modint> & coeffi=coeffmatmodp[i];
	  for (unsigned j=0;j<coeffi.size();++j){
	    if (coeffi[j])
	      ++countrescur;
	  }
	  if (countrescur>countres)
	    countres=countrescur;
	}
      }
      // if (iter==0) bound=pow(bres,int(countres),context0);
      if (stable){
	if (!chk_equal_mod(prevmatq,coeffmatmodp,env))
	  stable=false;
	// if stable compute bounds and compare with product of primes
	// if 2*bounds < product of primes recheck stabilization and return true
	if (is_strictly_greater(pip,bound,context0)){
	  if (debug_infolevel>0)
	    CERR << CLOCK()*1e-6 << " modular check finished " << '\n';
	  return true;
	}
      }
      // combine coeffmat with previous one by chinese remaindering
      if (debug_infolevel>0)
	CERR << CLOCK()*1e-6 << " chinrem mod " << p << '\n';
      if (iter)
	ichinrem_inplace(coeffmat,coeffmatmodp,pip,p.val);
      else
	vectvector_int2vecteur(coeffmatmodp,coeffmat);
      pip=pip*p;
      if (is_greater(bound,pip,context0))
	continue;
      if (!stable){
	// check stabilization
	matrice checkquo;
	checkquo.reserve(coeffmat.size());
	for (unsigned k=0;k<coeffmat.size();++k){
	  if (prevmatq.size()>k && chk_equal_mod(prevmatq[k],coeffmatmodp[k],env))
	    checkquo.push_back(prevmatq[k]);
	  else
	    checkquo.push_back(fracmod(coeffmat[k],pip));
	  if (prevmatq.size()>k && checkquo[k]!=prevmatq[k])
	    break;
	  if (k>(prevmatq.size()*3)/2+2)
	    break;
	}
	if (checkquo!=prevmatq){
	  swap(prevmatq,checkquo);
	  if (debug_infolevel>0)
	    CERR << CLOCK()*1e-6 << " unstable mod " << p << " reconstructed " << prevmatq.size() << '\n';
	  continue;
	}
	matrice coeffmatq=*_copy(checkquo,context0)._VECTptr;
	if (debug_infolevel>0)
	  CERR << CLOCK()*1e-6 << " full stable mod " << p << '\n';
	stable=true;
	gen lall=1; vecteur l(coeffmatq.size());
	for (unsigned i=0;i<coeffmatq.size();++i){
	  lcmdeno(*coeffmatq[i]._VECTptr,l[i],context0);
	  if (is_strictly_greater(l[i],lall,context0))
	    lall=l[i];
	}
	if (debug_infolevel>0)
	  CERR << CLOCK()*1e-6 << " lcmdeno ok/start bound " << p << '\n';
	gen ball=1,bi; // ball is the max bound of all coeff in coeffmatq
	for (unsigned i=0;i<coeffmatq.size();++i){
	  bi=linfnorm(coeffmatq[i],context0);
	  if (is_strictly_greater(bi,ball,context0))
	    ball=bi;
	}
	// bound for res and f4buchbergerv
	bound=bres*ball;
	gen bound2=lall*bf4buchberger;
	// lcm of deno and max of coeff
	if (is_strictly_greater(bound2,bound,context0))
	  bound=bound2;
      }
    }
    return true;
  }

  /* ***************
     BEGIN ZPOLYMOD
     ***************  */
#if GIAC_SHORTSHIFTTYPE==16
  // Same algorithms compressing data
  // since all spolys reduced at the same time share the same exponents
  // we will keep the exponents only once in memory
  // sizeof(zmodint)=8 bytes, sizeof(T_unsigned<modint,tdeg_t>)=28 or 36
  typedef T_unsigned<modint,unsigned> zmodint;
  template <class tdeg_t>
  struct zpolymod {
    order_t order;
    short int dim;
    bool in_gbasis; // set to false in zgbasis_updatemod for "small" reductors that we still want to use for reduction
    short int age:15;
    vector<zmodint> coord;
    const vector<tdeg_t> * expo;
    tdeg_t ldeg;
    int maxtdeg;
    int fromleft,fromright;
    double logz;
    zpolymod():in_gbasis(true),dim(0),expo(0),ldeg(),age(0),fromleft(-1),fromright(-1),logz(1) {order.o=0; order.lex=0; order.dim=0; maxtdeg=-1;}
    zpolymod(order_t o,int d): in_gbasis(true),dim(d),expo(0),ldeg(),age(0),fromleft(-1),fromright(-1),logz(1) {order=o; order.dim=d; maxtdeg=-1;}
    zpolymod(order_t o,int d,const tdeg_t & l): in_gbasis(true),dim(d),expo(0),ldeg(l),age(0),fromleft(-1),fromright(-1),logz(1) {order=o; order.dim=d; maxtdeg=-1;}
    zpolymod(order_t o,int d,const vector<tdeg_t> * e,const tdeg_t & l): in_gbasis(true),dim(d),expo(e),ldeg(l),age(0),fromleft(-1),fromright(-1),logz(1) {order=o; order.dim=d; maxtdeg=-1;}
    void dbgprint() const;
    void compute_maxtdeg(){
      if (expo){
	std::vector< zmodint >::iterator pt=coord.begin(),ptend=coord.end();
	for (;pt!=ptend;++pt){
	  int tmp=(*expo)[pt->u].total_degree(order);
	  if (tmp>maxtdeg)
	    maxtdeg=tmp;
	}
      }
    }
  };

  template<class tdeg_t>
  struct zinfo_t {
    vector< vector<tdeg_t> > quo;
    vector<tdeg_t> R,rem;
    vector<int> permu;
    vector< paire > B;
    vector<unsigned> G,permuB;
    unsigned nonzero,Ksizes;
  };

  template<class tdeg_t>
  void zsmallmultmod(modint a,zpolymod<tdeg_t> & p,modint m){
    std::vector< zmodint >::iterator pt=p.coord.begin(),ptend=p.coord.end();
    if (a==1 || a==1-m){
      for (;pt!=ptend;++pt){
	modint tmp=pt->g;
	if (tmp<0) tmp += m;
	pt->g=tmp;
      }
      return;
    }
    for (;pt!=ptend;++pt){
      modint tmp=(longlong(pt->g)*a)%m;
      if (tmp<0) tmp += m;
      pt->g=tmp;
    }
  }

  template<class tdeg_t>
  bool operator == (const zpolymod<tdeg_t> & p,const zpolymod<tdeg_t> &q){
    if (p.coord.size()!=q.coord.size() || p.expo!=q.expo)
      return false;
    for (unsigned i=0;i<p.coord.size();++i){
      if (p.coord[i].u!=q.coord[i].u || p.coord[i].g!=q.coord[i].g)
	return false;
    }
    return true;
  }

#ifdef NSPIRE
  template<class T,class tdeg_t>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os, const zpolymod<tdeg_t> & p)
#else
  template<class tdeg_t>
  ostream & operator << (ostream & os, const zpolymod<tdeg_t> & p)
#endif
  {
    if (!p.expo)
      return os << "error, null pointer in expo " ;
    std::vector<zmodint>::const_iterator it=p.coord.begin(),itend=p.coord.end();
    int t2;
    os << "zpolymod(" << p.logz << "," << p.age << ":" << p.fromleft << "," << p.fromright << "): ";
    if (it==itend)
      return os << 0 ;
    for (;it!=itend;){
      os << it->g  ;
#ifndef GBASIS_NO_OUTPUT
      if ((*p.expo)[it->u].vars64()){
	if ((*p.expo)[it->u].tdeg%2){
	  degtype * i=(degtype *)((*p.expo)[it->u].ui+1);
	  for (int j=0;j<(*p.expo)[it->u].order_.dim;++j){
	    t2=i[j];
	    if (t2)
	      os << "*x"<< j << "^" << t2  ;
	  }
	  ++it;
	  if (it==itend)
	    break;
	  os << " + ";
	  continue;
	}
      }
#endif
      short tab[GROEBNER_VARS+1];
      (*p.expo)[it->u].get_tab(tab,p.order);
      switch (p.order.o){
      case _PLEX_ORDER:
	for (int i=0;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2)
	    os << "*x"<< i << "^" << t2  ;
	}
	break;
      case _TDEG_ORDER:
	for (int i=1;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  if (t2)
	    os << "*x"<< i-1 << "^" << t2  ;
	}
	break;
      case _REVLEX_ORDER:
	for (int i=1;i<=GROEBNER_VARS;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
#if GROEBNER_VARS==15
      case _3VAR_ORDER:
	for (int i=1;i<=3;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 3-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=5;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 7+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
      case _7VAR_ORDER:
	for (int i=1;i<=7;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 7-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=9;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 11+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
      case _11VAR_ORDER:
	for (int i=1;i<=11;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 11-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	for (int i=13;i<=15;++i){
	  t2 = tab[i];
	  if (t2==0)
	    continue;
	  os << "*x"<< 15+p.dim-i;
	  if (t2!=1)
	    os << "^" << t2;
	}
	break;
#endif
      }
      ++it;
      if (it==itend)
	break;
      os << " + ";
    }
    return os;
  }

  template<class tdeg_t>
  void zpolymod<tdeg_t>::dbgprint() const {
    CERR << *this << '\n';
  }

  template<class tdeg_t>
  class vectzpolymod:public vector<zpolymod<tdeg_t> >{
  public:
    void dbgprint() const { CERR << *this << '\n'; }
  };

  template<class tdeg_t>
  void zleftright(const vectzpolymod<tdeg_t> & res,const vector< paire > & B,vector<tdeg_t> & leftshift,vector<tdeg_t> & rightshift){
    tdeg_t l;
    for (unsigned i=0;i<B.size();++i){
      const zpolymod<tdeg_t> & p=res[B[i].first];
      const zpolymod<tdeg_t> & q=res[B[i].second];
      if (debug_infolevel>2)
	CERR << "zleftright " << p << "," << q << '\n';
      index_lcm_overwrite(p.ldeg,q.ldeg,l,p.order);
      leftshift[i]=l-p.ldeg;
      rightshift[i]=l-q.ldeg;
    }
  }

  // collect monomials from pairs of res (vector of polymod<tdeg_t>s), shifted by lcm
  // does not collect leading monomial (since they cancel)
  template<class tdeg_t>
  bool zcollect(const vectzpolymod<tdeg_t> & res,const vector< paire > & B,const vector<unsigned> & permuB,vector<tdeg_t> & allf4buchberger,vector<tdeg_t> & leftshift,vector<tdeg_t> & rightshift){
    int start=1,countdiscarded=0;
    vector<heap_tt<tdeg_t> > Ht;
    heap_tt<tdeg_t> heap_elem;
    vector<heap_tt_ptr<tdeg_t> > H;
    Ht.reserve(2*B.size()+1);
    H.reserve(2*B.size());
    unsigned s=0;
    order_t keyorder={_REVLEX_ORDER,0};
    for (unsigned i=0;i<B.size();++i){
      unsigned truei=permuB[i];
      const paire & Bi=B[truei];
      const zpolymod<tdeg_t> & p=res[Bi.first];
      const zpolymod<tdeg_t> & q=res[Bi.second];
      keyorder=p.order;
      bool eq=i>0 && Bi.second==B[permuB[i-1]].second && rightshift[truei]==rightshift[permuB[i-1]];
      if (int(p.coord.size())>start){
	s = giacmax(s, unsigned(p.coord.size()));
	Ht.push_back(heap_tt<tdeg_t>(true,truei,start,(*p.expo)[p.coord[start].u]+leftshift[truei]));
	H.push_back(heap_tt_ptr<tdeg_t>(&Ht.back()));
      }
      if (!eq && int(q.coord.size())>start){
	s = giacmax(s, unsigned(q.coord.size()));
	Ht.push_back(heap_tt<tdeg_t>(false,truei,start,(*q.expo)[q.coord[start].u]+rightshift[truei]));
	H.push_back(heap_tt_ptr<tdeg_t>(&Ht.back()));
      }
    }
    allf4buchberger.reserve(s); // int(s*std::log(1+H.size())));
    compare_heap_tt_ptr<tdeg_t> key(keyorder);
    make_heap(H.begin(),H.end(),key);
    while (!H.empty()){
      // push root node of the heap in allf4buchberger
      heap_tt<tdeg_t> & current = *H.front().ptr;
      if (int(current.u.total_degree(keyorder))>GBASISF4_MAX_TOTALDEG){
	CERR << "Error zcollect total degree too large" << current.u.total_degree(keyorder) << '\n';
	return false;
      }
      if (allf4buchberger.empty() || allf4buchberger.back()!=current.u)
	allf4buchberger.push_back(current.u);
      unsigned vpos;
      if (current.left)
	vpos=B[current.f4buchbergervpos].first;
      else
	vpos=B[current.f4buchbergervpos].second;
      ++current.polymodpos;
      const zpolymod<tdeg_t> & resvpos=res[vpos];
      for (int startheappos=1;current.polymodpos<resvpos.coord.size();++current.polymodpos){
	add((*resvpos.expo)[resvpos.coord[current.polymodpos].u],current.left?leftshift[current.f4buchbergervpos]:rightshift[current.f4buchbergervpos],current.u,keyorder.dim);
	// if (current.left) current.u=(*resvpos.expo)[resvpos.coord[current.polymodpos].u]+leftshift[current.f4buchbergervpos]; else current.u=(*resvpos.expo)[resvpos.coord[current.polymodpos].u]+rightshift[current.f4buchbergervpos];
	int newtdeg=current.u.total_degree(keyorder);
	// quick look in part of heap: if monomial is already there, increment current.polymodpos
	int heappos=startheappos,ntests=int(H.size()); // giacmin(8,H.size());
	for (;heappos<ntests;){
	  heap_tt<tdeg_t> & heapcurrent=*H[heappos].ptr;
	  int heapcurtdeg=heapcurrent.u.total_degree(keyorder);
	  if (heapcurtdeg<newtdeg){
	    heappos=ntests; break;
	  }
	  if (heapcurtdeg==newtdeg && heapcurrent.u==current.u)
	    break;
	  //if (heappos<8) ++heappos; else
	    heappos=2*heappos+1;
	}
	if (heappos<ntests){
	  ++countdiscarded;
	  startheappos=2*heappos+1;
	  continue;
	}
	break;
      }
      // push_back &current into heap so that pop_heap will bubble out the
      // modified root node (initialization will exchange two identical pointers)
      if (current.polymodpos<resvpos.coord.size())
	H.push_back(heap_tt_ptr<tdeg_t>(&current));
      std::pop_heap(H.begin(),H.end(),key);
      H.pop_back();
    }
    if (debug_infolevel>1)
      CERR << "pairs " << B.size() << ", discarded monomials " << countdiscarded << '\n';
    return true;
  }

  // returns heap actual size, 0 means no quotients (all elements of q empty())
  template<class tdeg_t>
  size_t zsymbolic_preprocess(const vector<tdeg_t> & f,const vectzpolymod<tdeg_t> & g,const vector<unsigned> & G,unsigned excluded,vector< vector<tdeg_t> > & q,vector<tdeg_t> & rem,vector<tdeg_t> & R){
    int countdiscarded=0;
    // divides f by g[G[0]] to g[G[G.size()-1]] except maybe g[G[excluded]]
    // CERR << f << "/" << g << '\n';
    // first implementation: use quotient heap for all quotient/divisor
    // do not use heap chain
    // ref Monaghan Pearce if g.size()==1
    // R is the list of all monomials
    if (f.empty() || G.empty())
      return 0;
    int dim=g[G.front()].dim;
    order_t order=g[G.front()].order;
#ifdef GIAC_GBASIS_PERMUTATION
    // First reorder G in order to use the "best" possible reductor
    // This is done using the ldegree of g[G[i]] (should be minmal)
    // and the number of terms (should be minimal)
    vector<zsymb_data<tdeg_t> > GG(G.size());
    for (unsigned i=0;i<G.size();++i){
      zsymb_data<tdeg_t> zz={i,g[G[i]].ldeg,g[G[i]].order,unsigned(g[G[i]].coord.size()),g[G[i]].age};
      GG[i]=zz;
    }
    sort(GG.begin(),GG.end());
#endif
    tdeg_t minldeg(g[G.front()].ldeg);
    R.clear();
    rem.clear();
    // if (G.size()>q.size()) q.clear();
    q.resize(G.size());
    unsigned guess=0;
    for (unsigned i=0;i<G.size();++i){
      q[i].clear();
      guess += unsigned(g[G[i]].coord.size());
      if (!tdeg_t_greater(g[G[i]].ldeg,minldeg,order))
	minldeg=g[G[i]].ldeg;
    }
    vector<heap_t<tdeg_t> > H_;
    vector<unsigned> H;
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " Heap reserve " << 3*f.size() << " * " << sizeof(heap_t<deg_t>)+sizeof(unsigned) << ", number of monomials in basis " << guess  << '\n';
    // H_.reserve(guess);
    // H.reserve(guess);
    H_.reserve(3*f.size());
    H.reserve(3*f.size());
    heap_t_compare<tdeg_t> key(H_,order);
    unsigned k=0,i; // k=position in f
    tdeg_t m;
    bool finish=false;
    while (!H.empty() || k<f.size()){
      // is highest remaining degree in f or heap?
      if (k<f.size() && (H.empty() || tdeg_t_greater(f[k],H_[H.front()].u,order)) ){
	// it's in f or both
	m=f[k];
	++k;
      }
      else {
	m=H_[H.front()].u;
      }
      //CERR << m << '\n';
      R.push_back(m);
      // extract from heap all terms having m as monomials, substract from c
      while (!H.empty() && H_[H.front()].u==m){
	heap_t<tdeg_t> & current=H_[H.front()]; // was root node of the heap
	const zpolymod<tdeg_t> & gcurrent = g[G[current.i]];
	++current.gj;
	for (int startheappos=1;current.gj<gcurrent.coord.size();++current.gj){
	  //current.u=q[current.i][current.qi]+(*gcurrent.expo)[gcurrent.coord[current.gj].u];
	  add(q[current.i][current.qi],(*gcurrent.expo)[gcurrent.coord[current.gj].u],current.u,dim);
	  int newtdeg=current.u.total_degree(order);
	  // quick look in part of heap: is monomial already there?
	  int heappos=startheappos,ntests=int(H.size()); // giacmin(8,H.size());
	  for (;heappos<ntests;){
	    heap_t<tdeg_t> & heapcurrent=H_[H[heappos]];
	    int heapcurtdeg=heapcurrent.u.total_degree(order);
	    if (heapcurtdeg<newtdeg){
	      heappos=ntests; break;
	    }
	    if (heapcurtdeg==newtdeg && heapcurrent.u==current.u)
	      break;
	    //if (heappos<8) ++heappos; else
	      heappos=2*heappos+1;
	  }
	  if (heappos<ntests){
	    ++countdiscarded;
	    startheappos=2*heappos+1;
	    continue;
	  }
	  break;
	}
	if (current.gj<gcurrent.coord.size()){
	  H.push_back(H.front());
	  std::pop_heap(H.begin(),H.end(),key);
	  H.pop_back();
	}
	else {
	  std::pop_heap(H.begin(),H.end(),key);
	  H.pop_back();
	}
      } // end while !H.empty()
      // divide (c,m) by one of the g if possible, otherwise push in remainder
      if (finish){
	rem.push_back(m); // add to remainder
	continue;
      }
#ifdef GIAC_DEG_FIRST
      int mtot=m.total_degree(order);
#endif
      unsigned ii;
      for (ii=0;ii<G.size();++ii){
#ifdef GIAC_GBASIS_PERMUTATION
	i=GG[ii].pos; // we can use any permutation of 0..G.size()-1 here
#else
	i=ii;
#endif
	if (i==excluded)
	  continue;
	const tdeg_t & deg=g[G[i]].ldeg;
#ifdef GIAC_DEG_FIRST
	if (deg.total_degree(order)>mtot){
	  ii=G.size(); break;
	}
#endif
	if (tdeg_t_all_greater(m,deg,order))
	  break;
      }
      if (ii==G.size()){
	rem.push_back(m); // add to remainder
	// no monomial divide m, check if m is greater than one of the monomial of G
	// if not we can push all remaining monomials in rem
	finish=!tdeg_t_greater(m,minldeg,order);
	continue;
      }
      // add m/leading monomial of g[G[i]] to q[i]
      const zpolymod<tdeg_t> & gGi=g[G[i]];
      tdeg_t monom=m-gGi.ldeg;
      q[i].push_back(monom);
      // CERR << i << " " << q[i] << '\n';
      // push in heap
      int startheappos=0;
      for (int pos=1;pos<gGi.coord.size();++pos){
#if 0 // not useful here, but would be above!
	tdeg_t newmonom=(*gGi.expo)[gGi.coord[pos].u] + monom;
	int newtdeg=newmonom.total_degree(order);
	// quick look in part of heap is monomial already there
	int heappos=startheappos,ntests=H.size(); // giacmin(8,H.size());
	for (;heappos<ntests;heappos=2*heappos+1){
	  heap_t<tdeg_t> & current=H_[H[heappos]];
	  int curtdeg=current.u.total_degree(order);
	  if (curtdeg<newtdeg){
	    heappos=ntests; break;
	  }
	  if (curtdeg==newtdeg && current.u==newmonom)
	    break;
	}
	if (heappos<ntests){
	  ++countdiscarded;
	  startheappos=2*heappos+1;
	  continue;
	}
	heap_t<tdeg_t> current = { i, unsigned(q[i].size()) - 1, pos, newmonom };
#else
	heap_t<tdeg_t> current = { i, unsigned(q[i].size()) - 1, unsigned(pos), (*gGi.expo)[gGi.coord[pos].u] + monom };
#endif
	H.push_back(hashgcd_U(H_.size()));
	H_.push_back(current);
	key.ptr=&H_.front();
	std::push_heap(H.begin(),H.end(),key);
	break;
      }
    } // end main heap pseudo-division loop
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " Heap actual size was " << H_.size() << " discarded monomials " << countdiscarded << '\n';
    return H_.size();
  }

  template<class tdeg_t,class modint_t>
  void zcopycoeff(const zpolymod<tdeg_t> & p,vector<modint_t> & v,modint env,int start){
    std::vector< zmodint >::const_iterator it=p.coord.begin()+start,itend=p.coord.end();
    v.clear();
    v.reserve(itend-it);
    for (;it!=itend;++it){
      modint g=it->g;
      if (g<0) g += env;
      v.push_back(g);
    }
  }

template<class tdeg_t,class modint_t>
  void zcopycoeff(const zpolymod<tdeg_t> & p,vector<modint_t> & v,int start){
    std::vector< zmodint >::const_iterator it=p.coord.begin()+start,itend=p.coord.end();
    v.clear();
    v.reserve(itend-it);
    for (;it!=itend;++it){
      modint g=it->g;
      v.push_back(g);
    }
  }

  // dichotomic seach for jt->u==u in [jt,jtend[
  template<class tdeg_t>
  bool dicho(typename std::vector<tdeg_t>::const_iterator & jt,typename std::vector<tdeg_t>::const_iterator jtend,const tdeg_t & u,order_t order){
    if (*jt==u) return true;
    if (jtend-jt<=6){ ++jt; return false; }// == test faster
    for (;;){
      int step=int((jtend-jt)/2);
      typename std::vector<tdeg_t>::const_iterator j=jt+step;
      if (j==jt)
	return *j==u;
      //PREFETCH(&*(j+step/2));
      //PREFETCH(&*(jt+step/2));
      if (int res=tdeg_t_greater(*j,u,order)){
	jt=j;
	if (res==2)
	  return true;
      }
      else
	jtend=j;
    }
  }

#if 1 // #if 0 for old versions of gcc
  template<>
  bool dicho(std::vector<tdeg_t64>::const_iterator & jt,std::vector<tdeg_t64>::const_iterator jtend,const tdeg_t64 & u,order_t order){
    if (*jt==u) return true;
    if (jtend-jt<=6){ ++jt; return false; }// == test faster
#ifdef GIAC_ELIM
    if (u.tab[0]%2){
      int utdeg=u.tab[0],utdeg2=u.tdeg2;
      ulonglong uelim=u.elim;
      for (;;){
	int step=(jtend-jt)/2;
	std::vector<tdeg_t64>::const_iterator j=jt+step;
	if (j==jt)
	  return *j==u;
	if (j->tab[0]!=utdeg){
	  if (j->tdeg>utdeg)
	    jt=j;
	  else
	    jtend=j;
	  continue;
	}
	if (j->tdeg2!=utdeg2){
	  if (j->tdeg2>utdeg2)
	    jt=j;
	  else
	    jtend=j;
	  continue;
	}
	if (j->elim!=uelim){
	  if (j->elim<uelim)
	    jt=j;
	  else
	    jtend=j;
	  continue;
	}
	if (int res=tdeg_t_greater(*j,u,order)){
	  jt=j;
	  if (res==2)
	    return true;
	}
	else
	  jtend=j;
      }
    }
#endif
    for (;;){
      int step=int((jtend-jt)/2);
      std::vector<tdeg_t64>::const_iterator j=jt+step;
      if (j==jt)
	return *j==u;
      if (int res=tdeg_t_greater(*j,u,order)){
	jt=j;
	if (res==2)
	  return true;
      }
      else
	jtend=j;
    }
  }
#endif

  template<class tdeg_t>
  void zmakelinesplit(const zpolymod<tdeg_t> & p,const tdeg_t * shiftptr,const vector<tdeg_t> & R,void * Rhashptr,const vector<int> & Rdegpos,vector<shifttype> & v,vector<shifttype> * prevline,int start=0){
    std::vector<zmodint>::const_iterator it=p.coord.begin()+start,itend=p.coord.end();
    typename std::vector<tdeg_t>::const_iterator Rbegin=R.begin(),jt=Rbegin,jtend=R.end();
    double nop1=double(R.size());
    double nop2=2*p.coord.size()*std::log(nop1)/std::log(2.0);
    bool dodicho=nop2<nop1;
    const vector<tdeg_t> & expo=*p.expo;
    unsigned pos=0,Rpos=0;
    if (shiftptr){
      tdeg_t u=*shiftptr+*shiftptr; // create a new memory slot
      const shifttype * st=prevline?&prevline->front():0;
      for (;it!=itend;++it){
	add(expo[it->u],*shiftptr,u,p.dim);
#ifdef GIAC_RHASH
	int hashi=u.hash_index(Rhashptr);
	if (hashi>=0){
	  pushsplit(v,pos,hashi);
	  ++jt;
	  continue;
	}
#endif
	if (dodicho){
	  typename std::vector<tdeg_t>::const_iterator end=jtend;
	  if (st){
	    next_index(Rpos,st);
	    end=Rbegin+Rpos;
	  }
#ifdef GIAC_RDEG
	  int a=Rdegpos[u.tdeg+1],b=Rdegpos[u.tdeg];
	  if (jt-Rbegin<a)
	    jt=Rbegin+a;
	  if (end-Rbegin>b)
	    end=Rbegin+b;
#endif
	  if (dicho(jt,end,u,p.order)){
	    pushsplit(v,pos,unsigned(jt-Rbegin));
	    ++jt;
	    continue;
	  }
	}
	for (;jt!=jtend;++jt){
	  if (*jt==u){
	    pushsplit(v,pos,int(jt-Rbegin));
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=expo[it->u];
#ifdef GIAC_RHASH
	int hashi=u.hash_index(Rhashptr);
	if (hashi>=0){
	  pushsplit(v,pos,hashi);
	  ++jt;
	  continue;
	}
#endif
#if 1
	if (dodicho && dicho(jt,jtend,u,p.order)){
	  pushsplit(v,pos,unsigned(jt-Rbegin));
	  ++jt;
	  continue;
	}
#endif
	for (;jt!=jtend;++jt){
	  if (*jt==u){
	    pushsplit(v,pos,int(jt-Rbegin));
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  template<class tdeg_t,class modint_t>
  void zmakeline(const zpolymod<tdeg_t> & p,const tdeg_t * shiftptr,const vector<tdeg_t> & R,vector<modint_t> & v,int start=0){
    int Rs=int(R.size());
    // if (v.size()!=Rs) v.resize(Rs);
    // v.assign(Rs,0);
    std::vector<zmodint>::const_iterator it=p.coord.begin()+start,itend=p.coord.end();
    typename std::vector<tdeg_t>::const_iterator jt=R.begin(),jtbeg=jt,jtend=R.end();
    double nop1=double(R.size());
    double nop2=2*p.coord.size()*std::log(nop1)/std::log(2.0);
    bool dodicho=nop2<nop1;
    const std::vector<tdeg_t> & expo=*p.expo;
    if (shiftptr){
      tdeg_t u=R.front()+R.front(); // create u with refcount 1
      for (;it!=itend;++it){
	add(expo[it->u],*shiftptr,u,p.dim);
	if (dodicho && dicho(jt,jtend,u,p.order)){
	  v[jt-jtbeg]=it->g;
	  ++jt;
	  continue;
	}
	for (;jt!=jtend;++jt){
	  if (*jt==u){
	    v[jt-jtbeg]=it->g;
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=expo[it->u];
	if (dodicho && dicho(jt,jtend,u,p.order)){
	  v[jt-jtbeg]=it->g;
	  ++jt;
	  continue;
	}
	for (;jt!=jtend;++jt){
	  if (*jt==u){
	    v[jt-jtbeg]=it->g;
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

  template<class tdeg_t,class modint_t>
  void zmakelinesub(const zpolymod<tdeg_t> & p,const tdeg_t * leftshift,const zpolymod<tdeg_t> & q,const tdeg_t * rightshift,const vector<tdeg_t> & R,vector<modint_t> & v,int start,modint env){
    std::vector< zmodint >::const_iterator pt=p.coord.begin()+start,ptend=p.coord.end();
    std::vector< zmodint >::const_iterator qt=q.coord.begin()+start,qtend=q.coord.end();
    typename std::vector<tdeg_t>::const_iterator jt=R.begin(),jtbeg=jt,jtend=R.end();
    const std::vector<tdeg_t> & pexpo=*p.expo;
    const std::vector<tdeg_t> & qexpo=*q.expo;
    double nop1=double(R.size());
    double nop2=2*p.coord.size()*std::log(nop1)/std::log(2.0);
    bool dodicho=nop2<nop1;
    int dim=p.dim;
    order_t order=p.order;
    if (leftshift && rightshift){
      tdeg_t ul=R.front()+R.front();
      tdeg_t ur=R.front()+R.front();
      bool updatep=true,updateq=true;
      for (;pt!=ptend && qt!=qtend;){
	if (updatep){
	  add(pexpo[pt->u],*leftshift,ul,dim);
	  updatep=false;
	}
	if (updateq){
	  add(qexpo[qt->u],*rightshift,ur,dim);
	  updateq=false;
	}
	if (tdeg_t_greater(ul,ur,order)){
	  if (dodicho && dicho(jt,jtend,ul,order)){
	    if (ul==ur){
	      v[jt-jtbeg] = (pt->g-longlong(qt->g)); // %env;
	      ++jt; ++pt; ++qt;
	      updatep=updateq=true;
	      continue;
	    }
	    v[jt-jtbeg] = pt->g;
	    ++jt; ++pt;
	    updatep=true;
	    continue;
	  }
	  for (;jt!=jtend;++jt){
	    if (*jt==ul){
	      if (ul==ur){
		v[jt-jtbeg] = (pt->g-longlong(qt->g)); // %env;
		++jt; ++pt; ++qt;
		updatep=updateq=true;
	      }
	      else {
		v[jt-jtbeg] = pt->g;
		++jt; ++pt;
		updatep=true;
	      }
	      break;
	    }
	  }
	  continue;
	} // end if ul>=ur
	if (dodicho && dicho(jt,jtend,ur,order)){
	  v[jt-jtbeg] = -qt->g;
	  ++jt; ++qt;
	  updateq=true;
	  continue;
	}
	for (;jt!=jtend;++jt){
	  if (*jt==ur){
	    v[jt-jtbeg] = -qt->g;
	    ++jt; ++qt;
	    updateq=true;
	    break;
	  }
	}
      } // for (pt!=ptend && qt!=qtend)
      for (;pt!=ptend;){
	add(pexpo[pt->u],*leftshift,ul,dim);
	if (dodicho && dicho(jt,jtend,ul,order)){
	  v[jt-jtbeg] = pt->g;
	  ++jt; ++pt;
	  continue;
	}
	for (;jt!=jtend;++jt){
	  if (*jt==ul){
	    v[jt-jtbeg] = pt->g;
	    ++jt; ++pt;
	    break;
	  }
	}
      }
      for (;qt!=qtend;){
	add(qexpo[qt->u],*rightshift,ur,dim);
	if (dodicho && dicho(jt,jtend,ur,order)){
	  v[jt-jtbeg] = -qt->g;
	  ++jt; ++qt;
	  continue;
	}
	for (;jt!=jtend;++jt){
	  if (*jt==ur){
	    v[jt-jtbeg] = -qt->g;
	    ++jt; ++qt;
	    break;
	  }
	}
      }
    }
  }

  template<class tdeg_t,class modint_t>
  void zmakelinesub(const zpolymod<tdeg_t> & p,const tdeg_t * shiftptr,const vector<tdeg_t> & R,vector<modint_t> & v,int start,modint env){
    std::vector< zmodint >::const_iterator it=p.coord.begin()+start,itend=p.coord.end();
    typename std::vector<tdeg_t>::const_iterator jt=R.begin(),jtbeg=jt,jtend=R.end();
    const std::vector<tdeg_t> & expo=*p.expo;
    double nop1=double(R.size());
    double nop2=2*p.coord.size()*std::log(nop1)/std::log(2.0);
    bool dodicho=nop2<nop1;
    if (shiftptr){
      tdeg_t u=R.front()+R.front();
      for (;it!=itend;++it){
	add(expo[it->u],*shiftptr,u,p.dim);
	if (dodicho && dicho(jt,jtend,u,p.order)){
#if 1
	  v[jt-jtbeg] -= it->g;
#else
	  modint_t & vv=v[jt-jtbeg];
	  if (vv)
	    vv = (vv-longlong(it->g))%env;
	  else
	    vv=-it->g;
#endif
	  ++jt;
	  continue;
	}
	for (;jt!=jtend;++jt){
	  if (*jt==u){
#if 1
	    v[jt-jtbeg] -= it->g;
#else
	    modint_t & vv=v[jt-jtbeg];
	    if (vv)
	      vv = (vv-longlong(it->g));//%env;
	    else
	      vv = -it->g;
#endif
	    ++jt;
	    break;
	  }
	}
      }
    }
    else {
      for (;it!=itend;++it){
	const tdeg_t & u=expo[it->u];
	if (dodicho && dicho(jt,jtend,u,p.order)){
#if 1
	  v[jt-jtbeg] -= it->g;
#else
	  modint_t & vv=v[jt-jtbeg];
	  vv = (vv-longlong(it->g))%env;
#endif
	  ++jt;
	  continue;
	}
	for (;jt!=jtend;++jt){
	  if (*jt==u){
#if 1
	    v[jt-jtbeg]-=it->g;
#else
	    modint_t & vv=v[jt-jtbeg];
	    vv = (vv-longlong(it->g))%env;
#endif
	    ++jt;
	    break;
	  }
	}
      }
    }
  }

template<class modint_t,class modint_u>
  void zsub(vector<modint_t> & v64,const vector<modint_u> & subcoeff,const vector<shifttype> & subindex){
    if (subcoeff.empty()) return;
    typename vector<modint_t>::iterator wt=v64.begin();
    const modint_u * jt=&subcoeff.front(),*jtend=jt+subcoeff.size(),*jt_=jtend-8;
    const shifttype * it=&subindex.front();
    // first shift
    unsigned pos=0; next_index(pos,it); wt += pos;
    *wt -= (*jt); ++jt;
    bool shortshifts=v64.size()<0xffff?true:checkshortshifts(subindex);
#ifdef GIAC_SHORTSHIFTTYPE
    if (shortshifts){
      for (;jt!=jtend;++jt){
	wt += *it; ++it;
	*wt -= (*jt);
      }
    }
    else {
      for (;jt!=jtend;++jt){
	next_index(wt,it);
	*wt -= (*jt);
      }
    }
#else // def GIAC_SHORTSHIFTTYPE
    for (;jt!=jtend;++jt){
      v64[*it] -= (*jt);
      ++it; ++jt;
    }
#endif // def GIAC_SHORTSHIFTTYPE
  }


  void zadd(vector<modint2> & v64,const vector<modint> & subcoeff,const vector<shifttype> & subindex){
    if (subcoeff.empty()) return;
    vector<modint2>::iterator wt=v64.begin();
    const modint * jt=&subcoeff.front(),*jtend=jt+subcoeff.size();
    const shifttype * it=&subindex.front();
    // first shift
    unsigned pos=0; next_index(pos,it); wt += pos;
    *wt += (*jt); ++jt;
    bool shortshifts=v64.size()<0xffff?true:checkshortshifts(subindex);
#ifdef GIAC_SHORTSHIFTTYPE
    if (shortshifts){
      for (;jt!=jtend;++jt){
	wt += *it; ++it;
	*wt += (*jt);
      }
    }
    else {
      for (;jt!=jtend;++jt){
	next_index(wt,it);
	*wt += (*jt);
      }
    }
#else // def GIAC_SHORTSHIFTTYPE
    for (;jt!=jtend;++jt){
      v64[*it] += (*jt);
      ++it; ++jt;
    }
#endif // def GIAC_SHORTSHIFTTYPE
  }

  inline modint makepositive(modint a,modint n){
    return a-(a>>31)*n; // return a<0?a+n:a;
  }

  template<class tdeg_t,class modint_t>
  void zadd(vector<modint_t> & v64,const zpolymod<tdeg_t> & subcoeff,const vector<shifttype> & subindex,int start,modint env){
    if (subcoeff.coord.size()<=start) return;
    typename vector<modint_t>::iterator wt=v64.begin();
    const zmodint * jt=&subcoeff.coord.front(),*jtend=jt+subcoeff.coord.size();
    jt += start;
    const shifttype * it=&subindex.front();
    // first shift
    unsigned pos=0; next_index(pos,it); wt += pos;
    *wt = makepositive(jt->g,env); ++jt;
    bool shortshifts=v64.size()<0xffff?true:checkshortshifts(subindex);
#ifdef GIAC_SHORTSHIFTTYPE
    if (shortshifts){
      for (;jt!=jtend;++jt){
	wt += *it; ++it;
	*wt = makepositive(jt->g,env);
      }
    }
    else {
      for (;jt!=jtend;++jt){
	next_index(wt,it);
	*wt = makepositive(jt->g,env);
      }
    }
#else // def GIAC_SHORTSHIFTTYPE
    for (;jt!=jtend;++jt){
      v64[*it] = makepositive(jt->g,env);
      ++it; ++jt;
    }
#endif // def GIAC_SHORTSHIFTTYPE
  }

  template<class tdeg_t>
  void zcollect_interreduce(const vectzpolymod<tdeg_t> & res,const vector< unsigned > & G,vector<tdeg_t> & allf4buchberger,int start){
    vector< heap_tt<tdeg_t> > Ht;
    heap_tt<tdeg_t> heap_elem;
    vector< heap_tt_ptr<tdeg_t> > H;
    Ht.reserve(2*G.size()+1);
    H.reserve(2*G.size());
    unsigned s=0;
    order_t keyorder={_REVLEX_ORDER,0};
    for (unsigned i=0;i<G.size();++i){
      const zpolymod<tdeg_t> & p=res[G[i]];
      keyorder=p.order;
      if (int(p.coord.size())>start){
	s = giacmax(s, unsigned(p.coord.size()));
	Ht.push_back(heap_tt<tdeg_t>(true,i,start,(*p.expo)[p.coord[start].u]));
	H.push_back(heap_tt_ptr<tdeg_t>(&Ht.back()));
      }
    }
    allf4buchberger.reserve(s); // int(s*std::log(1+H.size())));
    compare_heap_tt_ptr<tdeg_t> key(keyorder);
    make_heap(H.begin(),H.end(),key);
    while (!H.empty()){
      // push root node of the heap in allf4buchberger
      heap_tt<tdeg_t> & current = *H.front().ptr;
      if (allf4buchberger.empty() || allf4buchberger.back()!=current.u)
	allf4buchberger.push_back(current.u);
      ++current.polymodpos;
      unsigned vpos;
      vpos=G[current.f4buchbergervpos];
      if (current.polymodpos>=res[vpos].coord.size()){
	std::pop_heap(H.begin(),H.end(),key);
	H.pop_back();
	continue;
      }
      const zpolymod<tdeg_t> & resvpos=res[vpos];
      current.u=(*resvpos.expo)[resvpos.coord[current.polymodpos].u];
      // push_back &current into heap so that pop_heap will bubble out the
      // modified root node (initialization will exchange two identical pointers)
      H.push_back(heap_tt_ptr<tdeg_t>(&current));
      std::pop_heap(H.begin(),H.end(),key);
      H.pop_back();
    }
  }

  template<class tdeg_t>
  int zf4mod(vectzpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,const vector< paire > & B,const vector<unsigned> * & permuBptr,vectzpolymod<tdeg_t> & f4buchbergerv,bool learning,unsigned & learned_position,vector< paire > * pairs_reducing_to_zero,vector<zinfo_t<tdeg_t> > & f4buchberger_info,unsigned & f4buchberger_info_position,bool recomputeR,int age,bool multimodular,int parallel,bool interreduce);

  template<class tdeg_t>
  int zinterreduce_convert(vectzpolymod<tdeg_t> & res,vector< unsigned > & G,int env,bool learning,unsigned & learned_position,vector< paire > * pairs_reducing_to_zero,vector<zinfo_t<tdeg_t> > & f4buchberger_info,unsigned & f4buchberger_info_position,bool recomputeR,int age,bool multimodular,int parallel,vectpolymod<tdeg_t> & resmod,bool interred){
    if (!interred)
      return 12345;
    if (res.empty()){ resmod.clear(); return 0; }
    order_t order=res.front().order;
    int dim=res.front().dim;
    unsigned Gs=G.size();
    // if (parallel<2 || Gs<200 || !threads_allowed ) return -1; // or fix in computeK1 non parallel case
    vector<paire> B; // not used
    const vector<unsigned> * permuBptr=0; // not used
    vectzpolymod<tdeg_t> f4buchbergerv;
    int tmp=zf4mod(res,G,env,B,permuBptr,f4buchbergerv,learning,learned_position,pairs_reducing_to_zero,f4buchberger_info,f4buchberger_info_position,recomputeR,age,multimodular,parallel,true);
    //CERR << "interreduce " << tmp << '\n';
    if (tmp<0 || tmp==12345)
      return tmp;
    // build resmod from res leading monomial of res and f4buchbergerv
    ulonglong tot=0;
    for (unsigned i=0;i<Gs;++i)
      tot += f4buchbergerv[i].coord.size();
    // if (tot==0) return -1;
    for (unsigned i=0;i<Gs;++i){
      polymod<tdeg_t> & q=resmod[G[i]];
      zpolymod<tdeg_t> & p=f4buchbergerv[i];
      const vector<tdeg_t> & expo=*p.expo;
      q.dim=res[G[i]].dim;
      q.order=res[G[i]].order;
      q.fromleft=res[G[i]].fromleft;
      q.fromright=res[G[i]].fromright;
      q.age=res[G[i]].age;
      q.logz=res[G[i]].logz;
      q.coord.clear();
      q.coord.reserve(1+p.coord.size());
      if (res[G[i]].coord.empty())
	return -1;
      q.coord.push_back(T_unsigned<modint,tdeg_t>(res[G[i]].coord[0].g,(*res[G[i]].expo)[res[G[i]].coord[0].u]));
      for (unsigned j=0;j<p.coord.size();++j){
	modint g=p.coord[j].g;
	// g=(modint2(g)*coeff)%env;
	q.coord.push_back(T_unsigned<modint,tdeg_t>(g,expo[p.coord[j].u]));
      }
    }
    return 0;
  }

  template<class tdeg_t>
  void Rtorem(const vector<tdeg_t> & R,const vector<tdeg_t> & rem,vector<unsigned> & v){
    v.resize(R.size());
    typename vector<tdeg_t>::const_iterator it=R.begin(),itend=R.end(),jt=rem.begin(),jt0=jt,jtend=rem.end();
    vector<unsigned>::iterator vt=v.begin();
    for (;jt!=jtend;++jt){
      const tdeg_t & t=*jt;
      for (;it!=itend;++vt,++it){
	if (*it==t)
	  break;
      }
      *vt=hashgcd_U(jt-jt0);
    }
  }

  template <class tdeg_t>
  struct thread_buchberger_t {
    const vectzpolymod<tdeg_t> * resptr;
    vector< vector< modint> > * Kptr;
    const vector<unsigned> * G;
    const vector< paire > * Bptr;
    const vector<unsigned> * permuBptr;
    const vector<tdeg_t> *leftshiftptr,*rightshiftptr,*Rptr;
    void * Rhashptr;
    const vector<int> * Rdegposptr;
    int env,debut,fin,N,colonnes;
    const vector<unsigned> * firstposptr;
    const vector<vector<unsigned short> > * Mindexptr;
    const vector< vector<modint> > * Mcoeffptr;
    const vector<coeffindex_t> * coeffindexptr;
    vector< vector<shifttype> > * indexesptr;
    vector<used_t> * usedptr;
    unsigned * bitmap;
    bool displayinfo;
    bool learning;
    bool interreduce;
    const vector<paire> * pairs_reducing_to_zero; // read-only!
    int learned_position;
  };

  template <class tdeg_t>
  void * thread_buchberger(void * ptr_){
    thread_buchberger_t<tdeg_t> * ptr=(thread_buchberger_t<tdeg_t> *) ptr_;
    const vectzpolymod<tdeg_t> & res=*ptr->resptr;
    vector< vector<modint> > & K =*ptr->Kptr;
    const vector< paire > & B = *ptr->Bptr;
    const vector<unsigned> & G = *ptr->G;
    const vector<unsigned> & permuB = *ptr->permuBptr;
    const vector<tdeg_t> & leftshift=*ptr->leftshiftptr;
    const vector<tdeg_t> & rightshift=*ptr->rightshiftptr;
    const vector<tdeg_t> & R=*ptr->Rptr;
    void * Rhashptr=ptr->Rhashptr;
    const vector<int> & Rdegpos=*ptr->Rdegposptr;
    int env=ptr->env,debut=ptr->debut,fin=ptr->fin,N=ptr->N;
    const vector<unsigned> & firstpos=*ptr->firstposptr;
    int & colonnes=ptr->colonnes;
    const vector<vector<unsigned short> > &Mindex = *ptr->Mindexptr;
    const vector< vector<modint> > &Mcoeff = *ptr->Mcoeffptr;
    const vector<coeffindex_t> &coeffindex = *ptr->coeffindexptr;
    vector< vector<shifttype> > & indexes=*ptr->indexesptr;
    vector<used_t> & used = *ptr->usedptr;
    bool learning=ptr->learning;
    bool interreduce=ptr->interreduce;
    int pos=ptr->learned_position;
    const vector<paire> * pairs_reducing_to_zero=ptr->pairs_reducing_to_zero;
    bool displayinfo=ptr->displayinfo;
    unsigned * bitmap=ptr->bitmap+debut*((N>>5)+1);
    vector<modint2> v64(N);
    unsigned bk_prev=-1;
    const tdeg_t * rightshift_prev =0;
    vector<modint> subcoeff2;
    int effi=-1,Bs=int(B.size());
    if (interreduce){
      // tdeg_t nullshift(res[G[0]].dim);
      for (int i=debut;i<fin;++i){
	if (interrupted || ctrl_c)
	  return 0;
	zmakelinesplit(res[G[i]],(const tdeg_t *) 0,R,Rhashptr,Rdegpos,indexes[i],0,1);
	zadd(v64,res[G[i]],indexes[i],1,env);
	K[i].clear();
	int firstcol=indexes[i].empty()?0:indexes[i].front();
	colonnes=giacmin(colonnes,reducef4buchbergersplit(v64,Mindex,firstpos,firstcol,Mcoeff,coeffindex,K[i],bitmap,used,env));
	bitmap += (N>>5)+1;
      }
      return ptr_;
    }
    for (int i=debut;i<fin;++i){
      if (interrupted || ctrl_c)
	return 0;
      paire bk=B[permuB[i]];
      if (!learning && pairs_reducing_to_zero && pos<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[pos]){
	++pos;
	continue;
      }
      // no learning with parallel
      zmakelinesplit(res[bk.first],&leftshift[permuB[i]],R,Rhashptr,Rdegpos,indexes[i],0,1);
      if (bk_prev!=bk.second || !rightshift_prev || *rightshift_prev!=rightshift[permuB[i]]){
	zmakelinesplit(res[bk.second],&rightshift[permuB[i]],R,Rhashptr,Rdegpos,indexes[Bs+i],0,1);
	bk_prev=bk.second;
	rightshift_prev=&rightshift[permuB[i]];
      }
    }
    bk_prev=-1; rightshift_prev=0;
    pos=ptr->learned_position;
    for (int i=debut;i<fin;++i){
      if (interrupted || ctrl_c)
	return 0;
      if (displayinfo){
	if (i%10==9) {COUT << "+"; COUT.flush(); }
	if (i%500==499) COUT << " " << CLOCK()*1e-6 << " remaining " << fin-i << '\n';
      }
      paire bk=B[permuB[i]];
      if (!learning && pairs_reducing_to_zero && pos<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[pos]){
	++pos;
	unsigned tofill=(N>>5)+1;
	fill(bitmap,bitmap+tofill,0);
	bitmap += tofill;
	continue;
      }
      if (bk.second!=bk_prev || !rightshift_prev || *rightshift_prev!=rightshift[permuB[i]]){
	subcoeff2.clear();
	zcopycoeff(res[bk.second],subcoeff2,1);
	bk_prev=bk.second;
	rightshift_prev=&rightshift[permuB[i]];
      }
      // zcopycoeff(res[bk.first],subcoeff1,1);zadd(v64,subcoeff1,indexes[i]);
      zadd(v64,res[bk.first],indexes[i],1,env);
      effi=Bs+i;
      while (indexes[effi].empty() && effi)
	--effi;
      zsub(v64,subcoeff2,indexes[effi]);
      int firstcol=indexes[i].empty()?0:indexes[i].front();
      if (effi>=0 && !indexes[effi].empty())
	firstcol=giacmin(firstcol,indexes[effi].front());
      K[i].clear();
      colonnes=giacmin(colonnes,reducef4buchbergersplit(v64,Mindex,firstpos,firstcol,Mcoeff,coeffindex,K[i],bitmap,used,env));
      bitmap += (N>>5)+1;
    }
    return ptr_;
  }

  template<class tdeg_t>
  struct pair_compare {
    const vector< paire > * Bptr;
    const vectzpolymod<tdeg_t> * resptr ;
    const vector<tdeg_t> * leftshiftptr;
    const vector<tdeg_t> * rightshiftptr;
    order_t o;
    inline bool operator ()(unsigned a,unsigned b){
      unsigned Ba=(*Bptr)[a].second,Bb=(*Bptr)[b].second;
      const tdeg_t & adeg=(*resptr)[Ba].ldeg;
      const tdeg_t & bdeg=(*resptr)[Bb].ldeg;
      if (adeg!=bdeg)
	return tdeg_t_greater(bdeg,adeg,o)!=0; // return tdeg_t_greater(adeg,bdeg,o);
      const tdeg_t & aleft=(*rightshiftptr)[a];
      const tdeg_t & bleft=(*rightshiftptr)[b];
      return tdeg_t_strictly_greater(bleft,aleft,o);// return tdeg_t_strictly_greater(aleft,bleft,o);
    }
    pair_compare(const vector< paire > * Bptr_,
		 const vectzpolymod<tdeg_t> * resptr_ ,
		 const vector<tdeg_t> * leftshiftptr_,
		 const vector<tdeg_t> * rightshiftptr_,
		 const order_t & o_):Bptr(Bptr_),resptr(resptr_),rightshiftptr(rightshiftptr_),leftshiftptr(leftshiftptr_),o(o_){}
  };

  longlong memory_usage(){
#if defined HAVE_SYS_RESOURCE_H && !defined NSPIRE && !defined NSPIRE_NEWLIB
    struct rusage r_usage;
    getrusage(RUSAGE_SELF,&r_usage);
#ifdef __APPLE__
    return r_usage.ru_maxrss;
#else
    return r_usage.ru_maxrss*1000;
#endif
#endif
    return -1;
  }

   // #define GIAC_CACHE2ND 1; // cache 2nd pair reduction, slower

  template<class tdeg_t>
  int zf4computeK1(const unsigned N,const unsigned nrows,const double mem,const unsigned Bs,vectzpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,const vector< paire > & B,const vector<unsigned> & permuB,bool learning,unsigned & learned_position,vector< paire > * pairs_reducing_to_zero,const vector<tdeg_t> & leftshift,const vector<tdeg_t> & rightshift, const vector<tdeg_t> & R ,void * Rhashptr,const vector<int> & Rdegpos,const vector<unsigned> &firstpos,vector<vector<unsigned short> > & Mindex, const vector<coeffindex_t> & coeffindex,vector< vector<modint> > & Mcoeff,zinfo_t<tdeg_t> * info_ptr,vector<used_t> &used,unsigned & usedcount,unsigned * bitmap,vector< vector<modint> > & K,int parallel,bool interreduce){
    bool freemem=mem>4e7; // should depend on real memory available
    bool large=N>8000;
    // CERR << "after sort " << Mindex << '\n';
    // step3 reduce
    unsigned colonnes=N;
    vector<modint> v(N);
    vector<modint2> v64(N);
    vector<double> v64d(N);
#ifdef x86_64
    vector<int128_t> v128;
    if (!large)
      v128.resize(N);
#endif
    unsigned Kcols=N-nrows;
    unsigned Ksizes=Kcols;
    if (info_ptr && !learning)
      Ksizes=giacmin(info_ptr->Ksizes+3,Kcols);
    bool Kdone=false;
    int th=parallel-1; // giacmin(threads,64)-1;
#ifdef GIAC_CACHE2ND
    vector<modint2> subcoeff2;
#else
    vector<modint> subcoeff2;
#endif
    vector< vector<shifttype> > indexes(2*Bs);
#ifdef HAVE_LIBPTHREAD
    if (Bs>=200 && threads_allowed && parallel>1
	//&& (learning || !pairs_reducing_to_zero)
	/*parallel*/){
      vector<int> positions(1),learned_parallel(1);
      if (interreduce){
	for (unsigned i=0;i<Bs;++i){
	  indexes[i].reserve(res[G[i]].coord.size()+16);
	  K[i].reserve(Ksizes);
	}
	for (int i=1;i<parallel;++i){
	  positions.push_back((i*longlong(Bs))/parallel);
	}
	positions.push_back(Bs);
	learned_parallel=positions; // not used
      }
      else {
	// prepare memory and positions
	int pos=learned_position;
	for (unsigned i=0;i<Bs;++i){
	  if (pairs_reducing_to_zero && !learning && (*pairs_reducing_to_zero)[learned_position]==B[permuB[i]]){
	    ++learned_position;
	    continue;
	  }
	}
	// effective number of pairs to reduce is Bs-(learned_position-pos)
	int effBs=Bs-(learned_position-pos),effstep=effBs/parallel+1,effi=0,effend=effstep;
	learned_parallel[0]=pos;
	// scan again pairs to set end positions and learned_position
	for (unsigned i=0;i<Bs;++i){
	  if (pairs_reducing_to_zero && !learning && (*pairs_reducing_to_zero)[pos]==B[permuB[i]]){
	    ++pos;
	    continue;
	  }
	  indexes[i].reserve(res[B[permuB[i]].first].coord.size()+16);
	  indexes[Bs+i].reserve(res[B[permuB[i]].second].coord.size()+16);
	  K[i].reserve(Ksizes);
	  ++effi;
	  if (effi>=effend){
	    positions.push_back(i); // end position for this thread
	    learned_parallel.push_back(pos); // learned position for next thread
	    effend += effstep;
	  }
	}
	// fix last pair number
	while (positions.size()<parallel)
	  positions.push_back(Bs);
	while (learned_parallel.size()<parallel)
	  learned_parallel.push_back(pos);
	if (positions.size()<parallel || learned_parallel.size()<parallel){
	  COUT << "BUG " << parallel << " " << positions.size() << " " << learned_parallel.size() << '\n' << positions << '\n' << learned_parallel << '\n';
	}
	if (positions.size()==parallel)
	  positions.push_back(Bs);
	else
	  positions.back()=Bs;
      } // end else interreduce
      pthread_t tab[64];
      thread_buchberger_t<tdeg_t> buchberger_param[64];
      int colonnes=N;
      for (int j=0;j<=th;++j){
	thread_buchberger_t<tdeg_t> tmp={&res,&K,&G,&B,&permuB,&leftshift,&rightshift,&R,Rhashptr,&Rdegpos,env,positions[j],positions[j+1],int(N),int(Kcols),&firstpos,&Mindex,&Mcoeff,&coeffindex,&indexes,&used,bitmap,j==th && debug_infolevel>1,learning,interreduce,pairs_reducing_to_zero,learned_parallel[j]};
	buchberger_param[j]=tmp;
	bool res=true;
	// CERR << "write " << j << " " << p << '\n';
	if (j<th)
	  res=pthread_create(&tab[j],(pthread_attr_t *) NULL,thread_buchberger<tdeg_t>,(void *) &buchberger_param[j]);
	if (res)
	  thread_buchberger<tdeg_t>((void *)&buchberger_param[j]);
      }
      Kdone=true;
      colonnes=buchberger_param[th].colonnes;
      for (unsigned j=0;j<th;++j){
	void * ptr_=(void *)&th; // non-zero initialisation
	pthread_join(tab[j],&ptr_);
	if (!ptr_)
	  Kdone=false;
	thread_buchberger_t<tdeg_t> * ptr = (thread_buchberger_t<tdeg_t> *) ptr_;
	colonnes=giacmin(colonnes,ptr->colonnes);
      }
    } // end parallelization
#endif
    if (!Kdone){
      if (interreduce){
	for (unsigned i=0;i<Bs;++i){
	  indexes[i].reserve(res[G[i]].coord.size()+16);
	  K[i].reserve(Ksizes);
	}
	thread_buchberger_t<tdeg_t> tmp={&res,&K,&G,&B,&permuB,&leftshift,&rightshift,&R,Rhashptr,&Rdegpos,env,0,(int)Bs,int(N),int(Kcols),&firstpos,&Mindex,&Mcoeff,&coeffindex,&indexes,&used,bitmap,debug_infolevel>1,learning,interreduce,pairs_reducing_to_zero,0};
	thread_buchberger<tdeg_t>((void *)&tmp);
	return 0;
      }
      unsigned bk_prev=-1;
      const tdeg_t * rightshift_prev=0;
      int pos=learned_position;
      for (unsigned i=0;i<Bs;i++){
	if (interrupted || ctrl_c)
	  return -1;
	paire bk=B[permuB[i]];
	if (!learning && pairs_reducing_to_zero && pos<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[pos]){
	  ++pos;
	  continue;
	}
	bool done=false;
	const tdeg_t & curleft=leftshift[permuB[i]];
#ifdef GIAC_MAKELINECACHE // does not seem faster
	pair<int,int> ij=zmakelinecache[bk.first];
	if (ij.first!=-1){
	  // look in quo[ij.first] if leftshift[permuB[i]] is there, if true copy from Mindex
	  // except first index
	  typename std::vector<tdeg_t>::const_iterator cache_it=quo[ij.first].begin(),cache_end=quo[ij.first].end();
	  if (cache_it<cache_end && !dicho(cache_it,cache_end,curleft,order)){
	    if (cache_end-cache_it>5)
	      cache_it=cache_end;
	    else {
	      for (;cache_it<cache_end;++cache_it){
		if (*cache_it==curleft)
		  break;
	      }
	    }
	  }
	  if (cache_it<cache_end){
	    if (debug_infolevel>2)
	      CERR << "cached " << ij << '\n';
	    int pos=ij.second+cache_it-quo[ij.first].begin();
	    pos=permuM[pos];
	    vector<shifttype> & source=Mindex[pos];
	    vector<shifttype> & target=indexes[i];
	    target.reserve(source.size());
	    unsigned sourcepos=0,targetpos=0;
	    const shifttype * sourceptr=&source.front(),*sourceend=sourceptr+source.size();
	    next_index(sourcepos,sourceptr); // skip position 0
	    next_index(sourcepos,sourceptr);
	    pushsplit(target,targetpos,sourcepos);
	    for (;sourceptr<sourceend;++sourceptr){
	      target.push_back(*sourceptr);
	    }
	    //CERR << Mindex << '\n' << target << '\n';
	    done=true;
	  }
	}
#endif // GIAC_MAKELINECACHE
	if (!done){
	  indexes[i].reserve(res[bk.first].coord.size()+16);
	  zmakelinesplit(res[bk.first],&curleft,R,Rhashptr,Rdegpos,indexes[i],0,1);
	  //CERR << indexes[i] << '\n';
	}
	if (bk_prev!=bk.second || !rightshift_prev || *rightshift_prev!=rightshift[permuB[i]]){
	  indexes[Bs+i].reserve(res[bk.second].coord.size()+16);
	  zmakelinesplit(res[bk.second],&rightshift[permuB[i]],R,Rhashptr,Rdegpos,indexes[Bs+i],0,1);
	  bk_prev=bk.second;
	  rightshift_prev=&rightshift[permuB[i]];
	}
      } // end for (unsigned i=0;i<Bs;++i)
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " pairs indexes computed over " << R.size() << " monomials"<<'\n';
      bk_prev=-1; rightshift_prev=0;
      vector<modint> Ki; Ki.reserve(Ksizes);
      int effi=-1;
      for (unsigned i=0;i<Bs;++i){
	if (interrupted || ctrl_c)
	  return -1;
	if (debug_infolevel>1){
	  if (i%10==9) {COUT << "+"; COUT.flush(); }
	  if (i%500==499) COUT << " " << CLOCK()*1e-6 << " remaining " << Bs-i << '\n';
	}
	paire bk=B[permuB[i]];
	if (!learning && pairs_reducing_to_zero && learned_position<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[learned_position]){
	  if (debug_infolevel>2)
	    CERR << bk << " f4buchberger learned " << learned_position << '\n';
	  ++learned_position;
	  unsigned tofill=(N>>5)+1;
	  fill(bitmap,bitmap+tofill,0);
	  bitmap += tofill;
	  continue;
	}
	// zmakelinesub(res[bk.first],&leftshift[i],res[bk.second],&rightshift[i],R,v,1,env);
	// CERR << bk.first << " " << leftshift[i] << '\n';
	// v64.assign(N,0); // + reset v64 to 0, already done by zconvert_
	if (bk.second!=bk_prev || !rightshift_prev || *rightshift_prev!=rightshift[permuB[i]]){
	  subcoeff2.clear();
#ifdef GIAC_CACHE2ND
	  subcoeff2.resize(N);
	  zadd(subcoeff2,res[bk.second],indexes[i+Bs],1,env);
	  reducef4buchbergersplit(subcoeff2,Mindex,firstpos,0,Mcoeff,coeffindex,Ki,0 /* no bitmap, answer in subcoeff2 */,used,env);
#else
	  zcopycoeff(res[bk.second],subcoeff2,1);
#endif
	  bk_prev=bk.second;
	  rightshift_prev=&rightshift[permuB[i]];
	  if (effi>=0)
	    indexes[effi].clear();
	  effi=i+Bs;
	}
	int firstcol=indexes[i].empty()?0:indexes[i].front();
	if (effi>=0 && !indexes[effi].empty())
	  firstcol=giacmin(firstcol,indexes[effi].front());
	// zcopycoeff(res[bk.first],subcoeff1,1);zadd(v64,subcoeff1,indexes[i]);
	if (
#ifdef EMCC
	    env>(1<<24) && env<=94906249
#else
	    0
#endif
	    ){
	  // using doubles instead of 64 bits integer not supported in JS
	  zadd(v64d,res[bk.first],indexes[i],1,env);
	  indexes[i].clear();
#ifdef GIAC_CACHE2ND
	  sub(v64d,subcoeff2);
#else
	  zsub(v64d,subcoeff2,indexes[effi]);
#endif
	  Ki.clear();
	  colonnes=giacmin(colonnes,reducef4buchbergersplitdouble(v64d,Mindex,firstpos,firstcol,Mcoeff,coeffindex,Ki,bitmap,used,env));
	}
	else {
	  zadd(v64,res[bk.first],indexes[i],1,env);
	  indexes[i].clear();
#ifdef GIAC_CACHE2ND
	  sub(v64,subcoeff2);
#else
	  zsub(v64,subcoeff2,indexes[effi]);
#endif
	  Ki.clear();
	  colonnes=giacmin(colonnes,reducef4buchbergersplit(v64,Mindex,firstpos,firstcol,Mcoeff,coeffindex,Ki,bitmap,used,env));
	}
	bitmap += (N>>5)+1;
	if (Ksizes<Kcols){
	  K[i].swap(Ki);
	  Ki.reserve(Ksizes);
	  continue;
	}
	size_t Kis=Ki.size();
	if (Kis>Ki.capacity()*.8){
	  K[i].swap(Ki);
	  Ki.reserve(giacmin(Kcols,int(Kis*1.1)));
	}
	else {
#if 0
	  vector<modint> & target=K[i];
	  target.reserve(giacmin(Kcols,int(Kis*1.1)));
	  vector<modint>::const_iterator kit=Ki.begin(),kitend=Ki.end();
	  for (;kit!=kitend;++kit)
	    target.push_back(*kit);
#else
	  K[i]=Ki;
#endif
	}
	//CERR << v << '\n' << SK[i] << '\n';
      } // end for (i=0;i<B.size();++i)
    } // end if (!Kdone)
    // CERR << K << '\n';
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv split reduced " << Bs << " polynoms over " << N << " monomials, start at " << colonnes << '\n';
    return 0;
  }

#if 0
  void convert(const std::vector<modint2> & source,std::vector<modint> & target,modint env){
    target.clear();
    std::vector<modint2>::const_iterator it=source.begin(),itend=source.end();
    for (;it!=itend;++it){
      if (*it)
	target.push_back((*it)%env);
      else
	target.push_back(0);
    }
  }

  template<class tdeg_t>
  bool operator <(const pair<unsigned,const tdeg_t > & a,
		  const pair<unsigned,const tdeg_t > & b){
    if (a.first!=b.first)
      return a.first<b.first;
    order_t tmp={_REVLEX_ORDER,0};
    return !tdeg_t_greater(b.second,a.second,tmp);
  }

  template<class tdeg_t>
  int zf4computeK2(const unsigned N,const unsigned nrows,const double mem,const unsigned Bs,vectzpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,const vector< paire > & B,const vector<unsigned> & permuB,bool learning,unsigned & learned_position,vector< paire > * pairs_reducing_to_zero,const vector<tdeg_t> & leftshift,const vector<tdeg_t> & rightshift, const vector<tdeg_t> & R ,void * Rhashptr,const vector<int> & Rdegpos,const vector<unsigned> &firstpos,vector<vector<unsigned short> > & Mindex, const vector<coeffindex_t> & coeffindex,vector< vector<modint> > & Mcoeff,zinfo_t<tdeg_t> * info_ptr,vector<used_t> &used,unsigned & usedcount,unsigned * bitmap,vector< vector<modint> > & K,int parallel){
    map< pair<unsigned,const tdeg_t >,int > cache;
    vector< vector<modint> > cachecoeffs;
    cachecoeffs.reserve(2*Bs);
    vector<int> leftpos(Bs,-1),rightpos(Bs,-1);
    unsigned n=0;
    for (unsigned i=0;i<Bs;++i){
      paire bk=B[permuB[i]];
      pair<unsigned,const tdeg_t > pleft(bk.first,leftshift[permuB[i]]);
      typename map< pair<unsigned,const tdeg_t >,int >::const_iterator it=cache.find(pleft),itend=cache.end();
      if (it!=itend)
	leftpos[i]=it->second;
      else {
	leftpos[i]=n;
	++n;
	cachecoeffs.push_back(vector<modint>());
	cache[pleft]=n;
      }
      pair<unsigned,const tdeg_t > pright(bk.second,rightshift[permuB[i]]);
      it=cache.find(pright);
      itend=cache.end();
      if (it!=itend)
	rightpos[i]=it->second;
      else {
	rightpos[i]=n;
	++n;
	cachecoeffs.push_back(vector<modint>());
	cache[pright]=n;
      }
    }
    // compare n to 2*Bs, if caching is efficient compute cache
    CERR << "Cache relative occupation " << n/(2.*Bs) << " pairs " << Bs << '\n';
    return zf4computeK1(N,nrows,mem,Bs,res,G,env, B,permuB,learning,learned_position,pairs_reducing_to_zero,leftshift,rightshift,  R ,Rhashptr,Rdegpos,firstpos,Mindex, coeffindex,Mcoeff,info_ptr,used,usedcount,bitmap,K,parallel,interreduce);
    vector<shifttype> index;
    vector<modint> Ki;
    vector<modint2> v64(N);
    for (unsigned i=0;i<Bs;++i){
      if (interrupted || ctrl_c)
	return 0;
      paire bk=B[permuB[i]];
      pair<unsigned,const tdeg_t > pleft(bk.first,leftshift[permuB[i]]);
      unsigned j=leftpos[i];
      if (cachecoeffs[j].empty()){
	std::fill(v64.begin(),v64.end(),0);
	index.clear();
	Ki.clear();
	zmakelinesplit(res[bk.first],&leftshift[permuB[i]],R,Rhashptr,Rdegpos,index,0,1);
	int firstcol=0; // index.empty()?0:index.front();
	zadd(v64,res[bk.first],index,1,env);
	reducef4buchbergersplit(v64,Mindex,firstpos,firstcol,Mcoeff,coeffindex,Ki,0 /* bitmap set to 0: result in v64*/,used,env);
	// v64->32 bits reduction and push in cache
	convert(v64,cachecoeffs[j],env);
      }
      vector<modint> vleft(cachecoeffs[j]);
      pair<unsigned,const tdeg_t> pright(bk.second,rightshift[permuB[i]]);
      j=rightpos[i];
      if (cachecoeffs[j].empty()){
	std::fill(v64.begin(),v64.end(),0);
	index.clear();
	Ki.clear();
	zmakelinesplit(res[bk.second],&rightshift[permuB[i]],R,Rhashptr,Rdegpos,index,0,1);
	int firstcol=index.empty()?0:index.front();
	zadd(v64,res[bk.second],index,1,env);
	reducef4buchbergersplit(v64,Mindex,firstpos,firstcol,Mcoeff,coeffindex,Ki,0 /* bitmap set to 0: result in v64*/,used,env);
	// v64->32 bits reduction and push in cache
	cachecoeffs.push_back(vector<modint>());
	convert(v64,cachecoeffs[j],env);
      }
      sub(vleft,cachecoeffs[j],env);
      // v64=vleft
      for (unsigned j=0;j<N;++j){
	v64[j]=vleft[j];
      }
      K.push_back(vector<modint>());
      K.back().reserve(N-nrows);
      store_coeffs(v64,0,K.back(),bitmap,used,env);
      bitmap += (N>>5)+1;
    }
    return 0;
  }
#endif

  template<class tdeg_t>
  struct zbuildM_t {
    const vectzpolymod<tdeg_t> * res;
    const vector<unsigned> * G;
    modint env;
    bool multimodular;
    const vector< vector<tdeg_t> > * quo;
    const vector<tdeg_t> * R;
    const vector<int> * Rdegpos;
    void * Rhashptr;
    vector<coeffindex_t> * coeffindex;
    unsigned N;
    vector<vector<unsigned short> > * Mindex;
    vector< vector<modint> > * Mcoeff;
    vector<sparse_element> * atrier;
    int i,iend,j;
  };

  template<class tdeg_t>
  void do_zbuildM(const vectzpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,bool multimodular,const vector< vector<tdeg_t> > & quo,const vector<tdeg_t> & R,const vector<int> & Rdegpos,void * Rhashptr,vector<coeffindex_t> & coeffindex,unsigned N,vector<vector<unsigned short> > & Mindex,vector< vector<modint> > & Mcoeff,vector<sparse_element> & atrier,int i,int iend,int j){
    for (;i<iend;++i){
      // copy coeffs of res[G[i]] in Mcoeff
      if (!quo[i].empty())
	zcopycoeff(res[G[i]],Mcoeff[i],0);
      // for each monomial of quo[i], find indexes and put in Mindex
      // reverse order traversing quo[i]
      // In zmakelinesplit locate res[G[i]].coord.u+*jt by dichotomoy
      // between position just calculated before and
      // and same position in previous Mindex
      typename std::vector< tdeg_t >::const_iterator jt=quo[i].end()-1;
      int quos=int(quo[i].size());
      int Gi=G[i];
      for (int k=quos-1;k>=0;--k,--jt){
	zmakelinesplit(res[Gi],&*jt,R,Rhashptr,Rdegpos,Mindex[j+k],k==quos-1?0:&Mindex[j+k+1],0);
      }
      for (int k=0;k<quos;++j,++k){
	coeffindex[j]=coeffindex_t(N<=0xffff,i);
	if (!coeffindex[j].b)
	  coeffindex[j].b=checkshortshifts(Mindex[j]);
	atrier[j]=sparse_element(first_index(Mindex[j]),j);
      }
    }
  }

  template<class tdeg_t>
  void * zbuildM_(void * ptr_){
    zbuildM_t<tdeg_t> * ptr=(zbuildM_t<tdeg_t> *) ptr_;
    do_zbuildM<tdeg_t>(*ptr->res,*ptr->G,ptr->env,ptr->multimodular,*ptr->quo,*ptr->R,*ptr->Rdegpos,ptr->Rhashptr,*ptr->coeffindex,ptr->N,*ptr->Mindex,*ptr->Mcoeff,*ptr->atrier,ptr->i,ptr->iend,ptr->j);
    return ptr_;
  }

  template<class tdeg_t>
  void zbuildM(const vectzpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,bool multimodular,int parallel,const vector< vector<tdeg_t> > & quo,const vector<tdeg_t> & R,const vector<int> & Rdegpos,void * & Rhashptr,vector<coeffindex_t> & coeffindex,unsigned N,vector<vector<unsigned short> > & Mindex,vector< vector<modint> > & Mcoeff,vector<sparse_element> & atrier,int nrows){
#ifdef HAVE_LIBPTHREAD
    if (nrows<16)
      parallel=1;
    pthread_t tab[parallel];
    zbuildM_t<tdeg_t> zbuildM_param[parallel];
    int istart=0,iend=0,jstart=0,jend=0;
    for (int j=0;j<parallel;++j){
      if (j==parallel-1){
	iend=G.size();
      }
      else {
	for (iend=istart;iend<G.size();++iend){
	  jend += quo[iend].size();
	  if (jend>((j+1)*nrows/parallel)){
	    ++iend;
	    break;
	  }
	}
      }
      zbuildM_t<tdeg_t> tmp={&res,&G,env,multimodular,&quo,&R,&Rdegpos,Rhashptr,&coeffindex,N,&Mindex,&Mcoeff,&atrier,istart,iend,jstart};
      zbuildM_param[j]=tmp;
      bool res=true;
      if (j<parallel-1)
	res=pthread_create(&tab[j],(pthread_attr_t *) NULL,zbuildM_<tdeg_t>,(void *) &zbuildM_param[j]);
      if (res)
	zbuildM_<tdeg_t>((void *)&zbuildM_param[j]);
      istart=iend;
      jstart=jend;
    }
    for (unsigned j=0;j<parallel-1;++j){
      void * ptr_=(void *)&parallel; // non-zero initialisation
      pthread_join(tab[j],&ptr_);
      if (!ptr_)
	CERR << "Error building M" << '\n';
    }
#else
    zbuildM_t<tdeg_t> tmp={&res,&G,env,multimodular,&quo,&R,&Rdegpos,Rhashptr,&coeffindex,N,&Mindex,&Mcoeff,&atrier,0,int(G.size()),0};
    zbuildM_<tdeg_t>((void *)&tmp);
#endif
  } // end parallelization

  template<class tdeg_t>
  int zf4mod(vectzpolymod<tdeg_t> & res,const vector<unsigned> & G,modint env,const vector< paire > & B,const vector<unsigned> * & permuBptr,vectzpolymod<tdeg_t> & f4buchbergerv,bool learning,unsigned & learned_position,vector< paire > * pairs_reducing_to_zero,vector<zinfo_t<tdeg_t> > & f4buchberger_info,unsigned & f4buchberger_info_position,bool recomputeR,int age,bool multimodular,int parallel,bool interreduce){
    unsigned Bs=unsigned(interreduce?G.size():B.size());
    if (!Bs)
      return 0;
    int dim=res.front().dim;
    order_t order=res.front().order;
    vector<tdeg_t> leftshift(Bs);
    vector<tdeg_t> rightshift(Bs);
    if (!interreduce)
      zleftright(res,B,leftshift,rightshift);
    // IMPROVEMENT: sort pairs in B according to right term of the pair
    // If several pairs share the same right term,
    // reduce the right term without leading monomial once
    // reduce corresponding left terms without leading monomial
    // substract
    f4buchbergerv.resize(Bs);
    zinfo_t<tdeg_t> info_tmp;
    unsigned nonzero = unsigned(Bs);
    zinfo_t<tdeg_t> * info_ptr=0;
    if (!learning && f4buchberger_info_position<f4buchberger_info.size()){
      info_ptr=&f4buchberger_info[f4buchberger_info_position];
      ++f4buchberger_info_position;
      nonzero=info_ptr->nonzero;
      if (nonzero==0 && !interreduce){
	for (int i=0;i<f4buchbergerv.size();++i){
	  // CERR << v << '\n';
	  f4buchbergerv[i].expo=&info_ptr->rem;
	  f4buchbergerv[i].order=order;
	  f4buchbergerv[i].dim=dim;
	  vector< zmodint > & Pcoord=f4buchbergerv[i].coord;
	  Pcoord.clear();
	}
	return 1;
      }
    }
    else {
      vector<tdeg_t> all;
      vector<unsigned> permuB(Bs);
      for (unsigned i=0;i<Bs;++i)
	permuB[i]=i;
#if 1 // not required for one modular gbasis, but kept for multi-modular
      if (!interreduce){
	pair_compare<tdeg_t> trieur(&B,&res,&leftshift,&rightshift,order);
	sort(permuB.begin(),permuB.end(),trieur);
	if (debug_infolevel>2){
	  unsigned egales=0;
	  for (unsigned i=1;i<Bs;++i){
	    if (B[permuB[i-1]].second!=B[permuB[i]].second)
	      continue;
	    if (rightshift[permuB[i-1]]==rightshift[permuB[i]]){
	      egales++;
	      CERR << B[permuB[i-1]] << "=" << B[permuB[i]] << '\n';
	    }
	  }
	  CERR << egales << " right pair elements are the same" << '\n';
	  CERR << CLOCK()*1e-6 << " zf4buchberger begin collect monomials on #polys " << f4buchbergerv.size() << '\n';
	}
      }
#endif
      if (interreduce){
	zcollect_interreduce(res,G,all,1); // all monomials after leading one
      }
      else {
	if (!zcollect(res,B,permuB,all,leftshift,rightshift))
	  return -1;
      }
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " zf4buchberger symbolic preprocess" << '\n';
      zsymbolic_preprocess(all,res,G,-1,info_tmp.quo,info_tmp.rem,info_tmp.R);
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " zend symbolic preprocess" << '\n';
#if 0
      f4buchberger_info->push_back(*info_ptr);
#else
      zinfo_t<tdeg_t> tmp;
      f4buchberger_info.push_back(tmp);
      zinfo_t<tdeg_t> & i=f4buchberger_info.back();
      swap(i.quo,info_tmp.quo);
      swap(i.R,info_tmp.R);
      swap(i.rem,info_tmp.rem);
      swap(i.permuB,permuB);
      info_ptr=&f4buchberger_info.back();
#endif
    }
    const vector<unsigned> & permuB =info_ptr->permuB ;
    permuBptr=&permuB;
    const vector<tdeg_t> & R = info_ptr->R;
    vector<unsigned> Rtoremv;
    Rtorem(R,info_ptr->rem,Rtoremv); // positions of R degrees in rem
    const vector< vector<tdeg_t> > & quo = info_ptr->quo;
    //CERR << quo << '\n';
    unsigned N = unsigned(R.size()), i, j = 0;
    if (N==0) return 1;
    void * Rhashptr=0;
#ifdef GIAC_RHASH // default disabled
    tdeg_t64_hash_t Rhash;
    if (R.front().vars64()){
      Rhashptr=&Rhash;
      //Rhash.reserve(N);
      for (unsigned i=0;i<N;++i){
	R[i].add_to_hash(Rhashptr,i);
      }
    }
#endif
    int Rcurdeg=R.front().tdeg;
    vector<int> Rdegpos(Rcurdeg+2);
#ifdef GIAC_RDEG // default enabled
    for (unsigned i=0;i<N;++i){
      int tmp=R[i].tdeg;
      if (tmp==Rcurdeg)
	continue;
      for (;Rcurdeg>tmp;--Rcurdeg){
	Rdegpos[Rcurdeg]=i;
      }
    }
    for (;Rcurdeg>=0;--Rcurdeg){
      Rdegpos[Rcurdeg]=N;
    }
#endif
    unsigned nrows=0;
    for (i=0;i<G.size();++i){
      nrows += unsigned(quo[i].size());
    }
    if (nrows==0 && interreduce){
      // allready interreduced, nothing to do...
      info_ptr->nonzero=G.size();
      return 12345; // special code, already interreduced
    }
    double sknon0=0;
    unsigned usedcount=0,zerolines=0;
    vector< vector<modint> > K(Bs);
    vector<vector<unsigned short> > Mindex;
    vector< vector<modint> > Mcoeff(G.size());
    vector<coeffindex_t> coeffindex(nrows);
    Mindex.reserve(nrows);
    vector<sparse_element> atrier(nrows);
    // atrier.reserve(nrows);
    for (i=0;i<G.size();++i){
      typename std::vector<tdeg_t>::const_iterator jt=quo[i].begin(),jtend=quo[i].end();
      if (jt!=jtend)
	Mcoeff[i].reserve(res[G[i]].coord.size());
      for (;jt!=jtend;++j,++jt){
	Mindex.push_back(vector<unsigned short>(0));
	Mindex[j].reserve(int(1.1*res[G[i]].coord.size()));
      }
    }
#ifndef GIAC_MAKELINECACHE
    zbuildM(res,G,env,multimodular,parallel,quo,R,Rdegpos,Rhashptr,coeffindex,N,Mindex,Mcoeff,atrier,nrows);
#else // ZBUILDM
#ifdef GIAC_MAKELINECACHE
    vector< pair<int,int> > zmakelinecache(res.size(),pair<int,int>(-1,-1)); // -1 if res[k] is not in G, (i,j) if k==G[i] where j is the first index in Mindex of the part corresponding to res[G[i]]
#endif
    for (i=0,j=0;i<G.size();++i){
      // copy coeffs of res[G[i]] in Mcoeff
      if (1 || env<(1<<24)){
	if (!quo[i].empty())
	  zcopycoeff(res[G[i]],Mcoeff[i],0);
      }
      else
	zcopycoeff(res[G[i]],Mcoeff[i],env,0);
      // if (!Mcoeff[i].empty()) Mcoeff[i].front()=invmod(Mcoeff[i].front(),env);
      // for each monomial of quo[i], find indexes and put in Mindex
      // Improvement idea: reverse order traversing quo[i]
      // In zmakelinesplit locate res[G[i]].coord.u+*jt by dichotomoy
      // between position just calculated before and
      // and same position in previous Mindex
#if 1
      typename std::vector< tdeg_t >::const_iterator jt=quo[i].end()-1;
      int quos=int(quo[i].size());
      int Gi=G[i];
#ifdef GIAC_MAKELINECACHE
      zmakelinecache[Gi]=pair<int,int>(i,j);
#endif
      for (int k=quos-1;k>=0;--k,--jt){
	zmakelinesplit(res[Gi],&*jt,R,Rhashptr,Rdegpos,Mindex[j+k],k==quos-1?0:&Mindex[j+k+1],0);
      }
      for (int k=0;k<quos;++j,++k){
	coeffindex[j]=coeffindex_t(N<=0xffff,i);
	if (!coeffindex[j].b)
	  coeffindex[j].b=checkshortshifts(Mindex[j]);
	// atrier.push_back(sparse_element(first_index(Mindex[j]),j));
	atrier[j]=sparse_element(first_index(Mindex[j]),j);
      }
#else
      typename std::vector< tdeg_t >::const_iterator jt=quo[i].begin(),jtend=quo[i].end();
      for (;jt!=jtend;++j,++jt){
	coeffindex[j]=coeffindex_t(N<=0xffff,i);
	zmakelinesplit(res[G[i]],&*jt,R,Rhashptr,Rdegpos,Mindex[j],0,0);
	if (!coeffindex[j].b)
	  coeffindex[j].b=checkshortshifts(Mindex[j]);
	// atrier.push_back(sparse_element(first_index(Mindex[j]),j));
	atrier[j]=sparse_element(first_index(Mindex[j]),j);
      }
#endif
    }
#endif // ZBUILM
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " end build Mindex/Mcoeff zf4mod" << '\n';
    // should not sort but compare res[G[i]]*quo[i] monomials to build M already sorted
    // CERR << "before sort " << Mindex << '\n';
    sort_vector_sparse_element(atrier.begin(),atrier.end()); // sort(atrier.begin(),atrier.end(),tri1);
    vector<coeffindex_t> coeffindex1(atrier.size());
    double mem=0; // mem*4=number of bytes allocated for M1
    vector< vector<unsigned short> > Mindex1(atrier.size());
#ifdef GIAC_MAKELINECACHE
    vector<int> permuM(atrier.size());
#endif
    for (i=0;i<atrier.size();++i){
      swap(Mindex1[i],Mindex[atrier[i].pos]);
      mem += Mindex1[i].size();
      swap(coeffindex1[i],coeffindex[atrier[i].pos]);
#ifdef GIAC_MAKELINECACHE
      permuM[atrier[i].pos]=i;
#endif
    }
    swap(Mindex,Mindex1);
    nrows = unsigned(Mindex.size());
    swap(coeffindex,coeffindex1);
    vector<unsigned> firstpos(atrier.size());
    for (i=0;i < atrier.size();++i){
      firstpos[i]=atrier[i].val;
    }
    double ratio=(mem/nrows)/N;
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " Mindex sorted, rows " << nrows << " columns " << N << " terms " << mem << " ratio " << ratio <<'\n';
    if (N<nrows){
      CERR << "Error " << N << "," << nrows << '\n';
      return -1;
    }
    // ((N>>5)+1)*Bs should not exceed 2e9 otherwise this will segfault
    if (double(Bs)*(N>>5)>2e9){
      CERR << "Error, problem too large. Try again after running gbasis_max_pairs(n) with n<" << 2e9/(N>>5) << '\n';
      return -1;
    }
    vector<used_t> used(N,0);
    vector<unsigned> lebitmap(((N>>5)+1)*Bs);
    unsigned * bitmap=&lebitmap.front();
    int zres=zf4computeK1(N,nrows,mem,Bs,res,G,env, B,permuB,learning,learned_position,pairs_reducing_to_zero,leftshift,rightshift,  R ,Rhashptr,Rdegpos,firstpos,Mindex, coeffindex,Mcoeff,info_ptr,used,usedcount,bitmap,K,parallel,interreduce);
    if (zres!=0)
      return zres;
    if (debug_infolevel>1){
      CERR << '\n' << CLOCK()*1e-6 << " Memory usage: " << memory_usage()*1e-6 << "M" << '\n';
    }
    size_t Mindexsize=Mindex.size();
    Mindex.clear();
    Mcoeff.clear();
    {
      vector<vector<unsigned short> > Mindexclear;
      vector< vector<modint> > Mcoeffclear;
      Mindex.swap(Mindexclear);
      Mcoeff.swap(Mcoeffclear);
    }
    if (!pairs_reducing_to_zero){
      vector<tdeg_t> clearer;
      info_ptr->R.swap(clearer);
    }
    for (unsigned i=0;i<N;++i)
      usedcount += (used[i]>0);
    if (learning && info_ptr)
      info_ptr->Ksizes=usedcount;
    if (debug_infolevel>1){
      CERR << CLOCK()*1e-6 << " number of non-zero columns " << usedcount << " over " << N-Mindexsize << " (N " << N  << ", Mindex size " << Mindexsize << ")" << '\n'; // usedcount should be approx N-M.size()=number of cols of M-number of rows
      if (debug_infolevel>3)
	CERR << " column split used " << used << '\n';
    }
    // create dense matrix K
    bitmap=&lebitmap.front();
    create_matrix(bitmap,(N>>5)+1,used,K);
    // clear memory required for lescoeffs
    //vector<modint> tmp; lescoeffs.swap(tmp);
    { vector<unsigned> tmp1; lebitmap.swap(tmp1); }
    if (debug_infolevel>1){
      CERR << CLOCK()*1e-6 << " rref " << K.size() << "x" << usedcount << '\n';
      double nz=0;
      for (unsigned i=0;i<K.size();++i){
	vector<int> & Ki=K[i];
	for (unsigned j=0;j<Ki.size();++j){
	  if (Ki[j]) ++nz;
	}
      }
      CERR << "non-0 percentage " << (nz/K.size())/K.front().size() << '\n';
    }
    vecteur pivots; vector<int> permutation,maxrankcols; longlong idet;
    if (0 && !learning && info_ptr->permu.size()==Bs){
      permutation=info_ptr->permu;
      vector< vector<modint> > K1(Bs);
      for (unsigned i=0;i<Bs;++i){
	swap(K1[i],K[permutation[i]]);
      }
      swap(K1,K);
    }
    int th=parallel-1; // giacmin(threads,64)-1;
    if (interreduce){ // interreduce==true means final interreduction
      ;
    }
    else {
#if 0
      // vector< vector<modint> > Kcopy(K);
      smallmodrref(parallel,K,pivots,permutation,maxrankcols,idet,0,int(K.size()),0,usedcount,1/* fullreduction*/,0/*dontswapbelow*/,env,0/* rrefordetorlu*/,permutation.empty(),0,!multimodular,0,-1); // disable rref optimization in multi-modular mode otherwise cyclic92 fails
#else
      smallmodrref(parallel,K,pivots,permutation,maxrankcols,idet,0,int(K.size()),0,usedcount,0/* lower reduction*/,0/*dontswapbelow*/,env,0/* rrefordetorlu*/,permutation.empty()/* reset */,0,!multimodular,-1);
      if (1){
	if (debug_infolevel>1)
	  CERR << CLOCK()*1e-6 << " rref_upper " << '\n';
	int Ksize=int(K.size());
	if (//1
	    usedcount<=2*Ksize
	    || parallel==1 || Ksize<50
	    )
	  smallmodrref_upper(K,0,Ksize,0,usedcount,env);
	else {
	  thread_smallmodrref_upper(K,0,Ksize,0,usedcount,env,parallel);
	}
      }
    } // end if !interreduce
#endif
    unsigned Kcols=N-nrows;
    free_null_lines(K,0,Bs,0,Kcols);
    unsigned first0 = unsigned(pivots.size());
    if (!interreduce && first0<K.size() && learning){
      vector<modint> & tmpv=K[first0];
      for (i=0;i<tmpv.size();++i){
	if (tmpv[i])
	  break;
      }
      if (i==tmpv.size()){
	unsigned Ksize = unsigned(K.size());
	K.resize(first0);
	K.resize(Ksize);
      }
    }
    //CERR << permutation << K << '\n';
    if (!learning){
      // check that permutation is the same as learned permutation
      bool copy=false;
      for (unsigned j=0;j<permutation.size();++j){
	if (permutation[j]!=info_ptr->permu[j]){
	  copy=true;
	  if (K[permutation[j]].empty() && K[info_ptr->permu[j]].empty())
	    continue;
	  CERR << "learning failed"<<'\n';
	  return -1;
	}
      }
      if (copy)
	permutation=info_ptr->permu;
    }
    if (learning)
      info_ptr->permu=permutation;
    // CERR << K << "," << permutation << '\n';
    // vector<int> permu=perminv(permutation);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv interreduced" << '\n';
    for (i=0;i<f4buchbergerv.size();++i){
      // CERR << v << '\n';
      int pi=interreduce?i:permutation[i];
      f4buchbergerv[pi].expo=&info_ptr->rem;
      f4buchbergerv[pi].order=order;
      f4buchbergerv[pi].dim=dim;
      f4buchbergerv[pi].age=age;
      vector< zmodint > & Pcoord=f4buchbergerv[pi].coord;
      Pcoord.clear();
      vector<modint> & v =K[i];
      if (v.empty()){
	continue;
      }
      unsigned vcount=0;
      vector<modint>::const_iterator vt=v.begin(),vtend=v.end();
      for (;vt!=vtend;++vt){
	if (*vt)
	  ++vcount;
      }
      Pcoord.reserve(vcount);
      vector<used_t>::const_iterator ut=used.begin();
      unsigned pos=0;
      for (vt=v.begin();pos<N;++ut,++pos){
	if (!*ut)
	  continue;
	modint coeff=*vt;
	++vt;
	if (coeff!=0)
	  Pcoord.push_back(zmodint(coeff,Rtoremv[pos]));
      }
      if (!Pcoord.empty())
	f4buchbergerv[pi].ldeg=(*f4buchbergerv[pi].expo)[Pcoord.front().u];
      if (!interreduce && !Pcoord.empty() && ( (env > (1<< 24)) || Pcoord.front().g!=1) ){
	zsmallmultmod(invmod(Pcoord.front().g,env),f4buchbergerv[pi],env);
	Pcoord.front().g=1;
      }
      bool freemem=mem>4e7; // should depend on real memory available
      if (freemem){
	vector<modint> tmp; tmp.swap(v);
      }
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " f4buchbergerv stored" << '\n';
    return 1;
  }

  /*
Let {f1, ..., fr} be a set of polynomials. The Gebauer-Moller Criteria are as follows:
1. Criterion M holds for a pair (fi, fk) if ∃ j < k, such that
   LCM{LM(fj),LM(fk)} properly divides LCM{LM(fi),LM(fk)}
2. Criterion F holds for a pair (fi, fk) if ∃ j < i, such that
   LCM{LM(fj),LM(fk)} = LCM{LM(fi),LM(fk)}.
3. Criterion Bk holds for a pair (fi, fj) if ∃ j < k and
   LM(fk) | LCM{LM(fi),LM(fj)},
   LCM{LM(fi),LM(fk)} != LCM{LM(fi),LM(fj)}, and
   LCM{LM(fi),LM(fj)} != LCM{LM(fj, fk)
  */

  // oldG is the Gbasis before the first line of f4buchbergerv is added
  // otherwise we might miss some new pairs to be added
  // f:=(1387482169552326*s*t1*t2^2-25694114250969*s*t1*t2+240071563017*s*t1+579168836143704*t1*t2^2-10725348817476*t1*t2+100212766488*t1):;fb:=(-7035747399*s*t1^2*t2^2+118865637*s*t1^2*t2-793881*s*t1^2+118865637*s*t1*t2^2-1167858*s*t1*t2+1944*s*t1-1089126*s*t2^2+1944*s*t2+18*s-2936742966*t1^2*t2^2+49601160*t1^2*t2-328050*t1^2+49601160*t1*t2^2-485514*t1*t2+972*t1-446148*t2^2+972*t2+36):;rmp:=s^2+10*s+4:;gbasis([f,fb,rmp],[s,t1,t2],revlex);
  template<class tdeg_t>
  void zgbasis_updatemod(vector<unsigned> & G,vector< paire > & B,vectzpolymod<tdeg_t> & res,unsigned pos,vector<unsigned> & oldG,bool multimodular){
    if (debug_infolevel>2)
      CERR << CLOCK()*1e-6 << " zmod begin gbasis update " << G.size() << '\n';
    if (debug_infolevel>3)
      CERR << "G=" << G << "B=" << B << '\n';
    const zpolymod<tdeg_t> & h = res[pos];
    order_t order=h.order;
    short dim=h.dim;
    vector<unsigned> C,Ccancel;
    C.reserve(G.size()+1);
    const tdeg_t & h0=h.ldeg;
    for (unsigned i=0;i<oldG.size();++i){
      if (tdeg_t_all_greater(h0,res[oldG[i]].ldeg,order))
	return;
    }
    tdeg_t tmp1,tmp2;
    // C is used to construct new pairs
    // create pairs with h and elements g of G, then remove
    // -> if g leading monomial is prime with h, remove the pair
    // -> if g leading monomial is not disjoint from h leading monomial
    //    keep it only if lcm of leading monomial is not divisible by another one
#if 1
    unsigned tmpsize = unsigned(G.size());
    vector<tdeg_t> tmp(tmpsize);
    for (unsigned i=0;i<tmpsize;++i){
      if (
#ifdef GIAC_GBASIS_REDUCTOR_MAXSIZE
	  res[G[i]].in_gbasis
#else
	  1
#endif
	  )
	index_lcm(h0,res[G[i]].ldeg,tmp[i],order);
      else
	tmp[i].tab[0]=-1;
    }
#else
    // this would be faster but it does not work for
    // gbasis([25*y^2*x^6-10*y^2*x^5+59*y^2*x^4-20*y^2*x^3+43*y^2*x^2-10*y^2*x+9*y^2-80*y*x^6+136*y*x^5+56*y*x^4-240*y*x^3+104*y*x^2+64*x^6-192*x^5+192*x^4-64*x^3,25*y^2*6*x^5-10*y^2*5*x^4+59*y^2*4*x^3-20*y^2*3*x^2+43*y^2*2*x-10*y^2-80*y*6*x^5+136*y*5*x^4+56*y*4*x^3-240*y*3*x^2+104*y*2*x+64*6*x^5-192*5*x^4+192*4*x^3-64*3*x^2,25*2*y*x^6-10*2*y*x^5+59*2*y*x^4-20*2*y*x^3+43*2*y*x^2-10*2*y*x+9*2*y-80*x^6+136*x^5+56*x^4-240*x^3+104*x^2],[x,y],revlex);
    // pair <4,3> is not generated
    unsigned tmpsize=G.empty()?0:G.back()+1;
    vector<tdeg_t> tmp(tmpsize);
    for (unsigned i=0;i<tmpsize;++i){
      index_lcm(h0,res[i].ldeg,tmp[i],order);
    }
#endif
    vector <tdeg_t> cancellables;
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
#ifdef GIAC_GBASIS_REDUCTOR_MAXSIZE
      if (!res[G[i]].in_gbasis)
	continue;
#endif
      if (interrupted || ctrl_c)
	return;
      if (disjoint(h0,res[G[i]].ldeg,order,dim)){
#ifdef GIAC_GBASIS_DELAYPAIRS
	Ccancel.push_back(G[i]);
#endif
	//cancellables.push_back(tmp[i]);
	continue;
      }
      //if (equalposcomp(cancellables,tmp[i])){ CERR << "cancelled!" << '\n'; continue; }
      // h0 and G[i] leading monomial not prime together
#if 1
      tdeg_t * tmp1=&tmp[i];
#else
      tdeg_t * tmp1=&tmp[G[i]];
#endif
      tdeg_t * tmp2=&tmp[0],*tmpend=tmp2+tmpsize;
      for (;tmp2!=tmp1;++tmp2){
	if (tmp2->tab[0]==-1)
	  continue;
	if (tdeg_t_all_greater(*tmp1,*tmp2,order))
	  break; // found another pair, keep the smallest, or the first if equal
      }
      if (tmp2!=tmp1){
	tmp1->tab[0]=-1; // desactivate tmp1 since it is >=tmp2
	continue;
      }
      for (++tmp2;tmp2<tmpend;++tmp2){
	if (tmp2->tab[0]==-1)
	  continue;
	if (tdeg_t_all_greater(*tmp1,*tmp2,order) && *tmp1!=*tmp2){
	  tmp1->tab[0]=-1; // desactivate tmp1 since it is >tmp2
	  break;
	}
      }
      if (tmp2==tmpend)
	C.push_back(G[i]);
    }
    vector< paire > B1;
    B1.reserve(B.size()+C.size());
    for (unsigned i=0;i<B.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
      if (res[B[i].first].coord.empty() || res[B[i].second].coord.empty())
	continue;
      index_lcm_overwrite(res[B[i].first].ldeg,res[B[i].second].ldeg,tmp1,order);
#ifdef GIAC_GBASIS_DELAYPAIRS
      if (!multimodular){
	// look for the pair B[i].second,pos compared to B[i].first/B[i].second
	// and delay pair (seems slower in multimodular mode)
	// this speeds up problems with symmetries like cyclic*
	// comparisons below seem reversed but they increase speed probably
	// because they detect symmetries
	tdeg_t tmpBi2;
	index_lcm(h0,res[B[i].second].ldeg,tmpBi2,order);
	if (int p=equalposcomp(C,B[i].second)){
	  if (
	      0 && tmp1==tmpBi2
	      // tdeg_t_all_greater(tmp1,tmpBi2,order)
	      ){
	    C[p-1]=C[p-1]|0x80000000;
	    B1.push_back(B[i]);
	    continue;
	  }
	  if (
	      //tmp1==tmpBi2
	      tdeg_t_all_greater(tmpBi2,tmp1,order)
	      //tdeg_t_all_greater(tmp1,tmpBi2,order)
	      ){
	    // set B[i] for later reduction
	    B[i].live=false;
	    B1.push_back(B[i]);
	    continue;
	  }
	}
#if 1
	if (int p=equalposcomp(Ccancel,B[i].second)){
	  if (
	      //tmp1==tmpBi2
	      tdeg_t_all_greater(tmpBi2,tmp1,order)
	      //tdeg_t_all_greater(tmp1,tmpBi2,order)
             ){
	    B[i].live=false;
	    B1.push_back(B[i]);
	    continue;
	  }
	}
#endif
      }
#endif // GIAC_GBASIS_DELAYPAIRS
      if (!tdeg_t_all_greater(tmp1,h0,order)){
	B1.push_back(B[i]);
	continue;
      }
      index_lcm_overwrite(res[B[i].first].ldeg,h0,tmp2,order);
      if (tmp2==tmp1){
	B1.push_back(B[i]);
	continue;
      }
      index_lcm_overwrite(res[B[i].second].ldeg,h0,tmp2,order);
      if (tmp2==tmp1){
	B1.push_back(B[i]);
	continue;
      }
    }
    // B <- B1 union pairs(h,g) with g in C
    for (unsigned i=0;i<C.size();++i){
      B1.push_back(paire(pos,C[i]&0x7fffffff));
      if (C[i] & 0x80000000)
	B1.back().live=false;
    }
    swap(B1,B);
    // Update G by removing elements with leading monomial >= leading monomial of h
    if (debug_infolevel>2){
      CERR << CLOCK()*1e-6 << " end, pairs:"<< '\n';
      if (debug_infolevel>3)
	CERR << B << '\n';
      CERR << "mod begin Groebner interreduce " << '\n';
    }
    C.clear();
    C.reserve(G.size()+1);
    // bool pos_pushed=false;
    for (unsigned i=0;i<G.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c)
	return;
#ifdef GIAC_GBASIS_REDUCTOR_MAXSIZE
      if (!res[G[i]].in_gbasis)
	continue;
#endif
      if (!res[G[i]].coord.empty() && !tdeg_t_all_greater(res[G[i]].ldeg,h0,order)){
	C.push_back(G[i]);
	continue;
      }
#ifdef GIAC_GBASIS_REDUCTOR_MAXSIZE
      if (res[G[i]].coord.size()<=GIAC_GBASIS_REDUCTOR_MAXSIZE){
	res[G[i]].in_gbasis=false;
	C.push_back(G[i]);
      }
#endif
      // NB: removing all pairs containing i in it does not work
    }
    if (debug_infolevel>2)
      CERR << CLOCK()*1e-6 << " zmod end gbasis update " << '\n';
    for (unsigned i=0;i<C.size();++i){
      if (!res[C[i]].coord.empty() && tdeg_t_all_greater(h0,res[C[i]].ldeg,order)){
	swap(C,G); return;
      }
    }
    C.push_back(pos);
    swap(C,G);
  }

  template<class tdeg_t>
  void convert(const polymod<tdeg_t> & p,zpolymod<tdeg_t> & q,const vector<tdeg_t> & R){
    q.order=p.order;
    q.dim=p.dim;
    q.coord.clear();
    q.coord.reserve(p.coord.size());
    typename vector< T_unsigned<modint,tdeg_t> >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    typename vector<tdeg_t>::const_iterator jt=R.begin(),jt0=jt,jtend=R.end();
    for (;it!=itend;++it){
      const tdeg_t & u=it->u;
      for (;jt!=jtend;++jt){
	if (*jt==u)
	  break;
      }
      if (jt!=jtend){
	q.coord.push_back(zmodint(it->g,int(jt-jt0)));
	++jt;
      }
      else
	COUT << "not found" << '\n';
    }
    q.expo=&R;
    if (!q.coord.empty())
      q.ldeg=R[q.coord.front().u];
    q.fromleft=p.fromleft;
    q.fromright=p.fromright;
    q.age=p.age;
    q.logz=p.logz;
  }

  template<class tdeg_t>
  void convert(const zpolymod<tdeg_t> & p,polymod<tdeg_t> & q){
    q.dim=p.dim;
    q.order=p.order;
    q.coord.clear();
    q.coord.reserve(p.coord.size());
    vector< zmodint >::const_iterator it=p.coord.begin(),itend=p.coord.end();
    const vector<tdeg_t> & expo=*p.expo;
    for (;it!=itend;++it){
      q.coord.push_back(T_unsigned<modint,tdeg_t>(it->g,expo[it->u]));
    }
    q.fromleft=p.fromleft;
    q.fromright=p.fromright;
    q.age=p.age;
    q.logz=p.logz;
  }

  template<class tdeg_t>
  void zincrease(vector<zpolymod<tdeg_t> > &v){
    if (v.size()!=v.capacity())
      return;
    vector<zpolymod<tdeg_t> > w;
    w.reserve(v.size()*2);
    for (unsigned i=0;i<v.size();++i){
      w.push_back(zpolymod<tdeg_t>(v[i].order,v[i].dim,v[i].expo,v[i].ldeg));
      w[i].coord.swap(v[i].coord);
      w[i].age=v[i].age;
      w[i].fromleft=v[i].fromleft;
      w[i].fromright=v[i].fromright;
      w[i].age=v[i].age;
      w[i].logz=v[i].logz;
    }
    v.swap(w);
  }

  template<class tdeg_t>
  void smod(polymod<tdeg_t> & resmod,modint env){
    typename std::vector< T_unsigned<modint,tdeg_t> >::iterator it=resmod.coord.begin(),itend=resmod.coord.end();
    for (;it!=itend;++it){
      modint n=it->g;
#if 0
      n %= env;
#endif
      if (n*2LL>env)
	it->g -= env;
      else {
	if (n*2LL<=-env)
	  it->g += env;
      }
    }
  }

  template<class tdeg_t>
  void smod(vectpolymod<tdeg_t> & resmod,modint env){
    for (unsigned i=0;i<resmod.size();++i)
      smod(resmod[i],env);
  }

#if 0 // def HAVE_LIBPTHREAD // too slow because of mem locks
  template<class tdeg_t>
  struct zfinal_interreduce_t {
    vectpolymod<tdeg_t> * res,*resmod;
    vector<unsigned> * G;
    modint env;
    int start,end;
  };

  template<class tdeg_t>
  void * do_zfinal_interreduce(void * ptr_){
    zfinal_interreduce_t<tdeg_t> * ptr=(zfinal_interreduce_t<tdeg_t> *) ptr_;
    vectpolymod<tdeg_t> &res=*ptr->res;
    const vectpolymod<tdeg_t> &resmod=*ptr->resmod;
    vector<unsigned> & G=*ptr->G;
    int end=giacmin(G.size(),ptr->end);
    polymod<tdeg_t> TMP1(res[G[0]].order,res[G[0]].dim);
    for (int j=ptr->start; j<end;++j){
      reducesmallmod(res[G[j]],resmod,G,j,ptr->env,TMP1,true,0);
    }
    return ptr;
  }

  template<class tdeg_t>
  void zfinal_interreduce(vectpolymod<tdeg_t> &resmod,vector<unsigned> & G,modint env,int parallel){
    if (resmod.empty() || G.empty())
      return;
    vectpolymod<tdeg_t> res(resmod);
    pthread_t tab[parallel];
    zfinal_interreduce_t<tdeg_t> interreduceparam[parallel];
    int kstep=int(std::ceil(G.size()/double(parallel))),start=0,end=kstep;
    for (int j=0;j<parallel;++j){
      zfinal_interreduce_t<tdeg_t> tmp={&res,&resmod,&G,env,start,end};
      start=end;
      end += kstep;
      interreduceparam[j]=tmp;
      bool res=true;
      if (j<parallel-1)
	res=pthread_create(&tab[j],(pthread_attr_t *) NULL,do_zfinal_interreduce<tdeg_t>,(void *) &interreduceparam[j]);
      if (res)
	do_zfinal_interreduce<tdeg_t>((void *)&interreduceparam[j]);
    }
    for (int j=0;j<parallel;++j){
      void * ptr=(void *)&parallel; // non-zero initialisation
      if (j<parallel-1)
	pthread_join(tab[j],&ptr);
    }
    res.swap(resmod);
  }
#endif

  template<class tdeg_t>
  bool in_zgbasis(vectpolymod<tdeg_t> &resmod,unsigned ressize,vector<unsigned> & G,modint env,bool totdeg,vector< paire > * pairs_reducing_to_zero,vector< zinfo_t<tdeg_t> > & f4buchberger_info,bool recomputeR,bool eliminate_flag,bool multimodular,int parallel,bool interred){
    unsigned generators=ressize;
    bool seldeg=true; int sel1=0;
    ulonglong cleared=0;
    unsigned learned_position=0,f4buchberger_info_position=0;
    bool learning=f4buchberger_info.empty();
    unsigned capa = unsigned(f4buchberger_info.capacity());
    order_t order=resmod.front().order;
    short dim=resmod.front().dim;
    // if (order.dim-order.o==1) seldeg=false;
    polymod<tdeg_t> TMP2(order,dim);
    vector< paire > B,BB;
    B.reserve(256); BB.reserve(256);
    vector<unsigned> smallposv;
    smallposv.reserve(256);
    info_t<tdeg_t> information;
    if (order.o!=_REVLEX_ORDER && order.o!=_TDEG_ORDER)
      totdeg=false;
    vector<unsigned> oldG(G);
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " initial reduction: " << ressize << " memory " << memory_usage()*1e-6 << '\n';
    for (unsigned l=0;l<ressize;++l){
#ifdef GIAC_REDUCEMODULO
      reducesmallmod(resmod[l],resmod,G,-1,env,TMP2,env!=0);
#endif
      gbasis_updatemod(G,B,resmod,l,TMP2,env,true,oldG);
    }
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " initial collect, pairs " << B.size() << '\n';
    // init zpolymod<tdeg_t> before main loop
    collect(resmod,TMP2);
    // R0 stores monomials for the initial basis
    vector<tdeg_t> R0(TMP2.coord.size());
    for (unsigned l=0;l<TMP2.coord.size();++l)
      R0[l]=TMP2.coord[l].u;
    // Rbuchberger stores monomials for pairs that are not handled by f4
    vector< vector<tdeg_t> > Rbuchberger;
    const int maxage=65535;
    Rbuchberger.reserve(maxage+1);
    vectzpolymod<tdeg_t> res;
    res.resize(ressize);
    for (unsigned l=0;l<ressize;++l){
      convert(resmod[l],res[l],R0);
      zsmallmultmod(1,res[l],env);
    }
    resmod.clear();
    if (debug_infolevel>1000){
      res.dbgprint(); res[0].dbgprint(); // instantiate
    }
    double timebeg=CLOCK(),autodebug=5e8;
    vector<int> start_index_v;
    if (debug_infolevel>1)
      CERR << CLOCK()*1e-6 << " begin loop, mem " << memory_usage()*1e-6 << '\n';
    for (int age=1;!B.empty() && !interrupted && !ctrl_c;++age){
      if (f4buchberger_info.size()>=capa-2 || age>maxage){
	CERR << "Error zgbasis too many iterations" << '\n';
	return false; // otherwise reallocation will make pointers invalid
      }
      if (debug_infolevel<2 && (CLOCK()-timebeg)>autodebug)
	debug_infolevel=multimodular?1:2;
      start_index_v.push_back(int(res.size())); // store size for final interreduction
#ifdef TIMEOUT
      control_c();
#endif
      if (f4buchberger_info_position>=capa-1){
	CERR << "Error f4 info exhausted" << '\n';
	return false;
      }
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " begin new iteration " << age << " zmod, " << env << " number of pairs: " << B.size() << ", base size: " << G.size() << '\n';
      vector<bool> clean(res.size(),true);
      for (unsigned i=0;i<int(G.size());++i){
	clean[G[i]]=false;
      }
      if (!totdeg){
	for (unsigned i=0;i<int(res.size());++i){
	  if (res[i].maxtdeg==-1)
	    res[i].compute_maxtdeg();
	}
      }
      vector<tdeg_t> Blcm(B.size());
      vector<int> Blcmdeg(B.size());
      vector<unsigned> nterms(B.size());
      for (unsigned i=0;i<B.size();++i){
	clean[B[i].first]=false;
	clean[B[i].second]=false;
	index_lcm(res[B[i].first].ldeg,res[B[i].second].ldeg,Blcm[i],order);
	if (!totdeg)
	  Blcmdeg[i]=giacmax(res[B[i].first].maxtdeg+(Blcm[i]-res[B[i].first].ldeg).total_degree(order),res[B[i].second].maxtdeg+(Blcm[i]-res[B[i].second].ldeg).total_degree(order));
	nterms[i]=unsigned(res[B[i].first].coord.size()+res[B[i].second].coord.size());
      }
#if 1
      for (unsigned i=0;i<clean.size();++i){
	if (clean[i] && res[i].coord.capacity()>1){
	  cleared += int(res[i].coord.capacity())-1;
	  zpolymod<tdeg_t> clearer;
	  clearer.coord.swap(res[i].coord);
	}
      }
#endif
      vector< paire > smallposp;
      vector<unsigned> smallposv;
      if (!totdeg){
	// find smallest lcm pair in B
	// could also take nterms[i] in account
	unsigned firstdeg=RAND_MAX-1;
	for (unsigned i=0;i<B.size();++i){
	  if (!B[i].live) continue;
	  unsigned f=seldeg?Blcmdeg[i]:Blcm[i].selection_degree(order);
	  //unsigned f=Blcmdeg[i];
	  if (f>firstdeg)
	    continue;
	  if (f<firstdeg){
	    firstdeg=f;
	    continue;
	  }
	}
	for (unsigned i=0;i<B.size();++i){
	  if (!B[i].live) continue;
	  if (
	      (seldeg?Blcmdeg[i]:Blcm[i].selection_degree(order))==firstdeg
	      //Blcmdeg[i]==firstdeg
	      ){
	    smallposv.push_back(i);
	  }
	}
	if (smallposv.empty()) smallposv.resize(B.size());
	if (debug_infolevel>1)
	  CERR << CLOCK()*1e-6 << " zpairs min " << (seldeg?"total degree":"elimination degree ") << firstdeg << " #pairs " << smallposv.size() << '\n';
	if ( seldeg && (smallposv.size()<giacmin(order.o,3))
	    ){
	  ++sel1;
	  if (sel1%5==0){
	    seldeg=!seldeg;
	    if (debug_infolevel)
	      CERR << "Switching pair selection strategy to " << (seldeg?"total degree":"elimination degree") << '\n';
	  }
	}
	else
	  sel1=0;
	if (int(firstdeg)>GBASISF4_MAX_TOTALDEG){
	  CERR << "Error zgbasis degree too large" << '\n';
	  return false;
	}
      }
      else {
	// find smallest lcm pair in B
	unsigned smallnterms=RAND_MAX,firstdeg=RAND_MAX-1;
	for (unsigned i=0;i<B.size();++i){
	  if (!B[i].live) continue;
	  unsigned f=Blcm[i].total_degree(order);
	  if (f>firstdeg)
	    continue;
	  if (f<firstdeg){
	    firstdeg=f;
	    smallnterms=nterms[i];
	    continue;
	  }
	  if (nterms[i]<smallnterms)
	    smallnterms=nterms[i];
	}
	smallnterms *= 5;
	for (unsigned i=0;i<B.size();++i){
	  if (!B[i].live) continue;
	  if (
	      //nterms[i]<=smallnterms &&
	      Blcm[i].total_degree(order)==firstdeg){
	    smallposv.push_back(i);
	  }
	}
	if (smallposv.empty()) smallposv.resize(B.size());
	if (debug_infolevel>1)
	  CERR << CLOCK()*1e-6 << " zpairs min total degrees, nterms " << firstdeg << "," << smallnterms << " #pairs " << smallposv.size() << '\n';
      }
      if (debug_infolevel>3)
	CERR << "pairs reduced " << B << " indices " << smallposv << '\n';
      if (order.o!=_REVLEX_ORDER && smallposv.size()<=GBASISF4_BUCHBERGER){
	// pairs not handled by f4
	int modsize=int(resmod.size());
	if (modsize<res.size())
	  resmod.resize(res.size());
	for (int i=0;i<int(G.size());++i){
	  int Gi=G[i];
	  if (resmod[Gi].coord.empty())
	    convert(res[Gi],resmod[Gi]);
	}
	polymod<tdeg_t> TMP1(order,dim),TMP2(order,dim);
	zpolymod<tdeg_t> TMP;
	paire bk=B[smallposv.back()];
	B.erase(B.begin()+smallposv.back());
	if (!learning && pairs_reducing_to_zero && learned_position<pairs_reducing_to_zero->size() && bk==(*pairs_reducing_to_zero)[learned_position]){
	  if (debug_infolevel>2)
	    CERR << bk << " learned " << learned_position << '\n';
	  ++learned_position;
	  continue;
	}
	if (debug_infolevel>2)
	  CERR << bk << " not learned " << learned_position << '\n';
	if (resmod[bk.first].coord.empty())
	  convert(res[bk.first],resmod[bk.first]);
	if (resmod[bk.second].coord.empty())
	  convert(res[bk.second],resmod[bk.second]);
	spolymod<tdeg_t>(resmod[bk.first],resmod[bk.second],TMP1,TMP2,env);
	if (debug_infolevel>1){
	  CERR << CLOCK()*1e-6 << " mod reduce begin, pair " << bk << " spoly size " << TMP1.coord.size() << " totdeg deg " << TMP1.coord.front().u.total_degree(order) << " degree " << TMP1.coord.front().u << ", pair degree " << resmod[bk.first].coord.front().u << resmod[bk.second].coord.front().u << '\n';
	}
#if 1
	reducesmallmod(TMP1,resmod,G,-1,env,TMP2,true,0,true);
	// insure that new basis element has positive coord, required by zf4mod
	typename vector< T_unsigned<modint,tdeg_t> >::iterator it=TMP1.coord.begin(),itend=TMP1.coord.end();
	for (;it!=itend;++it){
	  if (it->g<0)
	    it->g += env;
	}
	// reducemod(TMP1,resmod,G,-1,TMP1,env,true);
#else
	polymod<tdeg_t> TMP3(TMP1);
	reducemod(TMP1,resmod,G,-1,TMP1,env,true);
	reducesmallmod(TMP3,resmod,G,-1,env,TMP2,true,0,true);
	typename vector< T_unsigned<modint,tdeg_t> >::iterator it=TMP3.coord.begin(),itend=TMP3.coord.end();
	for (;it!=itend;++it){
	  if (it->g<0)
	    it->g += env;
	}
	if (TMP3.coord!=TMP1.coord){
	  CERR << "Bug" << '\n';
	}
#endif
	if (debug_infolevel>1){
	  if (debug_infolevel>3){ CERR << TMP1 << '\n'; }
	  CERR << CLOCK()*1e-6 << " mod reduce end, remainder degree " << TMP1.coord.front().u << " size " << TMP1.coord.size() << " begin gbasis update" << '\n';
	}
	if (!TMP1.coord.empty()){
	  resmod.push_back(TMP1);
	  Rbuchberger.push_back(vector<tdeg_t>(TMP1.coord.size()));
	  vector<tdeg_t> & R0=Rbuchberger.back();
	  for (unsigned l=0;l<unsigned(TMP1.coord.size());++l)
	    R0[l]=TMP1.coord[l].u;
	  convert(TMP1,TMP,R0);
	  zincrease(res);
	  if (ressize==res.size())
	    res.push_back(zpolymod<tdeg_t>(order,dim,TMP.ldeg));
	  res[ressize].expo=TMP.expo;
	  swap(res[ressize].coord,TMP.coord);
	  ++ressize;
	  zgbasis_updatemod(G,B,res,ressize-1,oldG,multimodular);
	  if (debug_infolevel>3)
	    CERR << CLOCK()*1e-6 << " mod basis indexes " << G << " pairs indexes " << B << '\n';
	}
	else {
	  if (learning && pairs_reducing_to_zero){
	    if (debug_infolevel>2)
	      CERR << "learning " << bk << '\n';
	    pairs_reducing_to_zero->push_back(bk);
	  }
	}
	continue;
      } // end if smallposp.size() small
      unsigned np=smallposv.size();
      if (np==B.size() && np<=max_pairs_by_iteration){
	swap(smallposp,B);
	B.clear();
      }
      else {
	// multiply by parallel?
	if (//!pairs_reducing_to_zero &&
	    np>max_pairs_by_iteration)
	  np=max_pairs_by_iteration;
	for (unsigned i=0;i<np;++i)
	  smallposp.push_back(B[smallposv[i]]);
	// remove pairs
	for (int i=int(np)-1;i>=0;--i)
	  B.erase(B.begin()+smallposv[i]);
      }
      vectzpolymod<tdeg_t> f4buchbergerv; // collect all spolys
      int f4res=-1;
      const vector<unsigned> * permuBptr=0;
#if 0
      unsigned Galls=G.back();
      vector<unsigned> Gall;
      Gall.reserve(Galls);
      for (unsigned i=0;i<Galls;++i){
	if (!clean[i])
	  Gall.push_back(i);
      }
      f4res=zf4mod(res,Gall,env,smallposp,permuBptr,f4buchbergerv,learning,learned_position,pairs_reducing_to_zero,f4buchberger_info,f4buchberger_info_position,recomputeR,age,multimodular,parallel,false);
#else
      f4res=zf4mod(res,G,env,smallposp,permuBptr,f4buchbergerv,learning,learned_position,pairs_reducing_to_zero,f4buchberger_info,f4buchberger_info_position,recomputeR,age,multimodular,parallel,false);
#endif
      if (f4res==-1)
	return false;
      if (f4res==0)
	continue;
      if (!permuBptr && !learning && f4buchberger_info_position-1<f4buchberger_info.size())
	permuBptr=&f4buchberger_info[f4buchberger_info_position-1].permuB;
      if (permuBptr){
	for (unsigned i=0;i<f4buchbergerv.size();++i){
	  f4buchbergerv[i].fromleft=smallposp[(*permuBptr)[i]].first;
	  f4buchbergerv[i].fromright=smallposp[(*permuBptr)[i]].second;
	  f4buchbergerv[i].logz=res[f4buchbergerv[i].fromleft].logz+res[f4buchbergerv[i].fromright].logz;
	}
      }
      // update gbasis and learning
      // requires that Gauss pivoting does the same permutation for other primes
      if (multimodular && learning && pairs_reducing_to_zero){
	for (unsigned i=0;i<f4buchbergerv.size();++i){
	  if (f4buchbergerv[i].coord.empty()){
	    if (debug_infolevel>2)
	      CERR << "learning f4buchberger " << smallposp[(*permuBptr)[i]] << '\n';
	    pairs_reducing_to_zero->push_back(smallposp[(*permuBptr)[i]]);
	  }
	}
      }
      unsigned added=0;
      for (unsigned i=0;i<f4buchbergerv.size();++i){
	if (!f4buchbergerv[i].coord.empty())
	  ++added;
      }
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " reduce f4buchberger end on " << added << " from " << f4buchbergerv.size() << " pairs, gbasis update begin" << '\n';
      vector<unsigned> oldG(G);
      for (int i=0;i<f4buchbergerv.size();++i){
	// for (int i=f4buchbergerv.size()-1;i>=0;--i){
	if (!f4buchbergerv[i].coord.empty()){
	  zincrease(res);
	  if (debug_infolevel>2)
	    CERR << CLOCK()*1e-6 << " adding to basis leading degree " << f4buchbergerv[i].ldeg << '\n';
	  if (ressize==res.size())
	    res.push_back(zpolymod<tdeg_t>(order,dim,f4buchbergerv[i].ldeg));
	  res[ressize].expo=f4buchbergerv[i].expo;
	  swap(res[ressize].coord,f4buchbergerv[i].coord);
	  res[ressize].age=f4buchbergerv[i].age;
	  res[ressize].fromleft=f4buchbergerv[i].fromleft;
	  res[ressize].fromright=f4buchbergerv[i].fromright;
	  res[ressize].logz=f4buchbergerv[i].logz;
	  ++ressize;
	  if (!multimodular || learning || f4buchberger_info_position-1>=f4buchberger_info.size())
	    zgbasis_updatemod(G,B,res,ressize-1,oldG,multimodular);
	}
	else {
	  // if (!learning && pairs_reducing_to_zero)  CERR << " error learning "<< '\n';
	}
      }
      if (!multimodular) continue;
      if (!learning && f4buchberger_info_position-1<f4buchberger_info.size()){
	B=f4buchberger_info[f4buchberger_info_position-1].B;
	G=f4buchberger_info[f4buchberger_info_position-1].G;
	continue;
      }
      if (learning){
	f4buchberger_info.back().B=B;
	f4buchberger_info.back().G=G;
	f4buchberger_info.back().nonzero=added;
      }
      //unsigned debut=G.size()-added;
      // CERR << "finish loop G.size "<<G.size() << '\n';
      // CERR << added << '\n';
    } // end main loop
#ifdef GIAC_GBASIS_REDUCTOR_MAXSIZE
    // remove small size reductors that were kept despite not being in gbasis
    vector<unsigned> G1;
    for (unsigned i=0;i<G.size();++i){
      if (res[G[i]].in_gbasis)
	G1.push_back(G[i]);
    }
    G.swap(G1);
#endif
    // convert back zpolymod<tdeg_t> to polymod<tdeg_t>
    // if eliminate_flag is true, keep only basis element that do not depend
    // on variables to eliminate
    if (eliminate_flag && (order.o==_3VAR_ORDER || order.o>=_7VAR_ORDER)){
      resmod.clear();
      resmod.reserve(res.size());
      for (unsigned l=0;l<res.size();++l){
	tdeg_t d=res[l].ldeg;
#ifndef GBASIS_NO_OUTPUT
	if (d.vars64()){
	  if (d.tab[0]%2){
	    if (d.tab[0]/2)
	      continue;
	  }
	  else {
	    if (d.tdeg)
	      continue;
	  }
	}
	else {
	  if (d.tab[0])
	    continue;
	}
#endif
	resmod.resize(resmod.size()+1);
	convert(res[l],resmod.back());
      }
      res.clear();
      G.resize(resmod.size());
      for (unsigned j=0; j<resmod.size();++j)
	G[j]=j;
      // final interreduce step2
      polymod<tdeg_t> TMP1(order,dim);
      for (unsigned j=0; j<resmod.size();++j){
	reducesmallmod(resmod[j],resmod,G,j,env,TMP1,true);
      }
    }
    else {
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " zfinal interreduction begin " << G.size() << '\n';
      resmod.resize(res.size());
      for (unsigned l=0;l<res.size();++l){
	resmod[l].coord.clear();
      }
      int val=-1;
      if (1//generators<100
	  //parallel>1 && threads_allowed && G.size()>=200
	  ){
	val=zinterreduce_convert(res,G,env,learning,learned_position,pairs_reducing_to_zero,f4buchberger_info,f4buchberger_info_position,recomputeR,-1/* age*/,multimodular,parallel,resmod,interred);
	if (debug_infolevel && val<0)
	  CERR << "zinterreduce failure" << '\n';
	// zfinal_interreduce(resmod,G,env,parallel); // res->resmod must be done. discarded because too slow mem locks
      }
      if (val<0 || val==12345){
	for (unsigned l=0;l<G.size();++l){
	  convert(res[G[l]],resmod[G[l]]);
	}
	res.clear();
      }
      if (val<0){
	// final interreduce step2
	// by construction resmod[G[j]] is already interreduced by resmod[G[k]] for k<j
	polymod<tdeg_t> TMP1(order,dim);
	for (int j=int(G.size())-1; j>=0;--j){
	  if (debug_infolevel>1){
	    if (j%10==0){ CERR << "+"; CERR.flush();}
	    if (j%500==0){ CERR << CLOCK()*1e-6 << " remaining " << j << '\n';}
	  }
	  if (!start_index_v.empty() && G[j]<start_index_v.back())
	    start_index_v.pop_back();
	  if (0 && order.o==_REVLEX_ORDER) // this optimization did not work for cyclic10mod (at least when max_pairs=4096)
	    reducesmallmod(resmod[G[j]],resmod,G,j,env,TMP1,true,start_index_v.empty()?0:start_index_v.back());
	  else // full interreduction since Buchberger steps do not interreduce
	    reducesmallmod(resmod[G[j]],resmod,G,j,env,TMP1,true,0);
	}
      }
      if (debug_infolevel>1)
	CERR << CLOCK()*1e-6 << " zfinal interreduction end " << G.size() << '\n';
    }
    if (ressize<resmod.size())
      res.resize(ressize);
    if (debug_infolevel){ // was debug_infolevel>1
      unsigned t=0;
      for (unsigned i=0;i<resmod.size();++i)
	t += unsigned(resmod[i].coord.size());
      CERR << CLOCK()*1e-6 << " total number of monomials in res " << t << " basis size " << G.size() << '\n';
      CERR << "Number of monomials cleared " << cleared << '\n';
    }
    smod(resmod,env);
    return true;
  }

  template<class tdeg_t>
  bool zgbasis(vectpoly8<tdeg_t> & res8,vectpolymod<tdeg_t> &resmod,vector<unsigned> & G,modint env,bool totdeg,vector< paire > * pairs_reducing_to_zero,vector< zinfo_t<tdeg_t> > & f4buchberger_info,bool recomputeR,bool convertpoly8,bool eliminate_flag,bool multimodular,int parallel,bool interred){
    for (unsigned i=0;i<resmod.size();++i)
      resmod[i].coord.clear();
    convert(res8,resmod,env);
    unsigned ressize = unsigned(res8.size());
    bool b=in_zgbasis(resmod,ressize,G,env,totdeg,pairs_reducing_to_zero,f4buchberger_info,recomputeR,eliminate_flag,multimodular,parallel,interred);
    if (convertpoly8)
      convert(resmod,res8,env);
    return b;
  }

#endif // GIAC_SHORTSHIFTTYPE==16
  /* *************
     END ZPOLYMOD
     ************* */
  template<class tdeg_t>
  bool is_gbasis(const vectpoly8<tdeg_t> & res,double eps,bool modularcheck){
    if (res.empty())
      return false;
    if (debug_infolevel>0)
      CERR << "basis size " << res.size() << '\n';
    // build possible pairs (i,j) with i<j
    vector< vector<tdeg_t> > lcmpairs(res.size());
    vector<unsigned> G(res.size());
    for (unsigned i=0;i<res.size();++i)
      G[i]=i;
    vectpoly8<tdeg_t> vtmp,tocheck;
    vector< paire > tocheckpairs;
    if (eps>0 && eps<2e-9)
      modularcheck=true;
    if (modularcheck)
      tocheck.reserve(res.size()*10); // wild guess
    else
      tocheckpairs.reserve(res.size()*10);
    order_t order=res.front().order;
    int dim=res.front().dim;
    poly8<tdeg_t> TMP1(order,res.front().dim),TMP2(TMP1),
      spol(TMP1),spolred(TMP1);
    polymod<tdeg_t> spolmod(order,dim),TMP1mod(order,dim);
    vectpolymod<tdeg_t> resmod;
    for (unsigned i=0;i<res.size();++i){
      const poly8<tdeg_t> & h = res[i];
      const tdeg_t & h0=h.coord.front().u;
      vector<tdeg_t> tmp(res.size());
      for (unsigned j=i+1;j<res.size();++j){
	index_lcm(h0,res[j].coord.front().u,tmp[j],h.order);
      }
      swap(lcmpairs[i],tmp);
    }
    for (unsigned i=0;i<res.size();++i){
      if (debug_infolevel>1)
	CERR << "checking pairs for i="<<i<<", j=";
      const poly8<tdeg_t> & resi = res[i];
      const tdeg_t & resi0=resi.coord.front().u;
      for (unsigned j=i+1;j<res.size();++j){
	if (disjoint(resi0,res[j].coord.front().u,order,dim))
	  continue;
	// criterion M, F
	unsigned J=0;
	tdeg_t & lcmij=lcmpairs[i][j];
	for (;J<i;++J){
	  if (tdeg_t_all_greater(lcmij,lcmpairs[J][j],order))
	    break;
	}
	if (J<i)
	  continue;
	for (++J;J<j;++J){
	  tdeg_t & lcmJj=lcmpairs[J][j];
	  if (tdeg_t_all_greater(lcmij,lcmJj,order) && lcmij!=lcmJj)
	    break;
	}
	if (J<j)
	  continue;
	// last criterion
	unsigned k;
	for (k=j+1;k<res.size();++k){
	  if (lcmpairs[i][k]!=lcmij && lcmpairs[j][k]!=lcmij
	      && tdeg_t_all_greater(lcmij,res[k].coord.front().u,order))
	    break;
	}
	if (k<res.size())
	  continue;
	// compute and reduce s-poly
	if (debug_infolevel>1)
	  CERR <<  j << ",";
	if (modularcheck){
	  spoly(resi,res[j],spol,TMP1,0);
	  tocheck.push_back(poly8<tdeg_t>(order,dim));
	  swap(tocheck.back(),spol);
	}
	else
	  tocheckpairs.push_back(paire(i,j));
      } // end j loop
      if (debug_infolevel>1)
	CERR << '\n';
    }
    if (debug_infolevel>0)
      CERR << "Number of critical pairs to check " << (modularcheck?tocheck.size():tocheckpairs.size()) << '\n';
    if (modularcheck) // modular check is sometimes slow
      return checkf4buchberger(tocheck,res,G,-1,eps); // split version is slower!
    // integer check or modular check for one modulus (!= from first prime already used)
    modint p=(prevprime((1<<29)-30000000)).val;
    if (eps>0)
      convert(res,resmod,p);
    // FIXME should be parallelized
    for (unsigned i=0;i<tocheckpairs.size();++i){
#ifdef TIMEOUT
      control_c();
#endif
      if (interrupted || ctrl_c){
	CERR << "Check interrupted, assuming Groebner basis. Press Ctrl-C again to interrupt computation" << '\n';
	interrupted=ctrl_c=false;
	return true;
      }
      if (eps>0){
	spolymod<tdeg_t>(resmod[tocheckpairs[i].first],resmod[tocheckpairs[i].second],spolmod,TMP1mod,p);
	reducemod(spolmod,resmod,G,-1,TMP1mod,p);
	// gen den; heap_reduce(spol,res,G,-1,vtmp,spolred,TMP1,den,0);
	if (!TMP1mod.coord.empty())
	  return false;
      }
      else {
	spoly(res[tocheckpairs[i].first],res[tocheckpairs[i].second],spol,TMP1,0);
	reduce(spol,res,G,-1,vtmp,spolred,TMP1,TMP2,0);
	// gen den; heap_reduce(spol,res,G,-1,vtmp,spolred,TMP1,den,0);
	if (!spolred.coord.empty())
	  return false;
      }
      if (debug_infolevel>0){
	CERR << "+";
	if (i%512==511)
	  CERR << tocheckpairs.size()-i << " remaining" << '\n';
      }
    }
    if (debug_infolevel)
      CERR << '\n' << "Successful check of " << tocheckpairs.size() << " critical pairs" << '\n';
    return true;
  }


  /* *************
     RUR UTILITIES (rational univariate representation for 0 dimension ideals)
     ************* */
  int rur_dim(int dim,order_t order){
    if (order.o==_3VAR_ORDER) return 3;
    if (order.o==_7VAR_ORDER) return 7;
    if (order.o==_11VAR_ORDER) return 11;
    if (order.o==_16VAR_ORDER) return 16;
    if (order.o==_32VAR_ORDER) return 32;
    if (order.o==_48VAR_ORDER) return 48;
    if (order.o==_64VAR_ORDER) return 64;
    return dim;
  }

  // returns -1 if not 0 dimensional, -RAND_MAX if overflow
  // otherwise returns dimension of quotient and sets lm to the list of
  // leading monomials generating the quotient ideal
  template<class tdeg_t>
  int rur_quotient_ideal_dimension(const vectpolymod<tdeg_t> & gbmod,polymod<tdeg_t> & lm){
    if (gbmod.empty())
      return -1;
    order_t order=gbmod.front().order;
    int dim=gbmod.front().dim;
    lm.order=order; lm.dim=dim; lm.coord.clear();
    polymod<tdeg_t> gblm(order,dim);
    unsigned S = unsigned(gbmod.size());
    for (unsigned i=0;i<S;++i){
      if (gbmod[i].coord.empty())
	continue;
      gblm.coord.push_back(gbmod[i].coord.front());
    }
    // for 3var, 7var, 11 var search in the first 3 var, 7 var or 11 var
    // for revlex search for all variables
    // we must find a leading monomial in gbmod that contains only this variable
    int d=rur_dim(dim,order);
    vector<int> v(d);
    for (unsigned i=0;i<S;++i){
      index_t l;
      get_index(gblm.coord[i].u,l,order,dim);
      unsigned j,k;
      for (j=0;int(j)<d;++j){
	if (l[j]){
	  for (k=j+1;int(k)<d;++k){
	    if (l[k])
	      break;
	  }
	  if (k==d)
	    v[j]=l[j];
	  break;
	}
      }
    }
    // now all indices of v must be non 0
    double M=1;
    for (unsigned i=0;i<v.size();++i){
      if (v[i]==0)
	return -1;
      M *= v[i];
    }
    if (M>1e6)
      return -RAND_MAX; // overflow
    // the ideal is finite dimension, now we will compute the exact dimension
    // a monomial degree is associated to an integer with
    // [l_0,l_1,...,l_{d-1}] -> ((l_0*v1+l_1)*v2+...+l_{d-1} < v0*v1*...
    // perhaps a sieve would be faster, but it's harder to implement
    // and we won't consider too high order anyway...
    index_t cur(d);
    for (int I=0;I<M;++I){
      int i=I;
      // i-> cur -> tdeg_t
      for (int j=int(v.size())-1;j>=0;--j){
	cur[j]=i%v[j];
	i/=v[j];
      }
      tdeg_t curu(cur,order);
      // then search if > to one of the leading monomials for all indices
      unsigned j=-1;
      if (order.o==_3VAR_ORDER){
	for (j=0;j<S;++j){
	  tdeg_t u=gblm.coord[j].u;
	  if (curu.tab[1]>=u.tab[1] && curu.tab[2]>=u.tab[2] && curu.tab[3]>=u.tab[3])
	    break;
	}
      }
      if (order.o==_7VAR_ORDER){
      }
      if (order.o==_11VAR_ORDER){
      }
      if (order.o==_REVLEX_ORDER){
	for (j=0;j<S;++j){
	  if (tdeg_t_all_greater(curu,gblm.coord[j].u,order))
	    break;
	}
      }
      // if found continue, else add cur to the list of monomials
      if (j==gbmod.size())
	lm.coord.push_back(T_unsigned<modint,tdeg_t>(1,curu));
    }
    sort(lm.coord.begin(),lm.coord.end(),tdeg_t_sort_t<tdeg_t>(order));
    return unsigned(lm.coord.size());
  }

  // multiply a by b mod p in res
  // b is supposed to have small length
  template<class tdeg_t>
  void rur_mult(const polymod<tdeg_t> & a,const polymod<tdeg_t> & b,modint p,polymod<tdeg_t> & res){
    res.coord.clear();
    polymod<tdeg_t> tmp(b.order,b.dim);
    for (unsigned i=0;i<b.coord.size();++i){
      smallmultsubmodshift(res,0,(-b.coord[i].g) % p,a,b.coord[i].u,tmp,p);
      tmp.coord.swap(res.coord);
    }
  }

  // coordinates of cur w.r.t. lm
  template<class tdeg_t>
  void rur_coordinates(const polymod<tdeg_t> & cur,const polymod<tdeg_t> & lm,vecteur & tmp){
    unsigned k=0,j=0;
    for (;j<lm.coord.size() && k<cur.coord.size();++j){
      if (lm.coord[j].u!=cur.coord[k].u)
	tmp[j]=0;
      else {
	tmp[j]=cur.coord[k].g;
	++k;
      }
    }
    for (;j<lm.coord.size();++j){
      tmp[j]=0;
    }
  }

  // s*coordinates reduced as a linear combination of the lines of M
  template<class tdeg_t>
  bool rur_linsolve(const vectpolymod<tdeg_t> & gbmod,const polymod<tdeg_t> & lm,const polymod<tdeg_t> & s,const matrice & M,modint p,matrice & res){
    int S=int(lm.coord.size()),dim=lm.dim;
    order_t order=lm.order;
    polymod<tdeg_t> TMP1(order,dim);
    vector<unsigned> G(gbmod.size());
    for (unsigned i=0;i<G.size();++i)
      G[i]=i;
    matrice N(M);
    polymod<tdeg_t> si(order,dim);
    int d=rur_dim(dim,order);
    vecteur tmp(lm.coord.size());
    for (unsigned i=0;int(i)<d;++i){
      index_t l(dim);
      l[i]=1;
      smallshift(s.coord,tdeg_t(l,order),si.coord);
      reducesmallmod(si,gbmod,G,-1,p,TMP1,false);
      // get coordinates of cur in tmp (make them mod p)
      rur_coordinates(si,lm,tmp);
      N.push_back(tmp);
    }
    N=mtran(N);
    if (debug_infolevel)
      CERR << CLOCK()*1e-6 << " rur rref" << '\n';
    gen n=_rref(N,context0);
    if (!ckmatrix(n))
      return false;
    N=mtran(*n._VECTptr);
    // check that first line are idn
    for (unsigned i=0;i<lm.coord.size();++i){
      vecteur & Ni=*N[i]._VECTptr;
      for (unsigned j=0;j<lm.coord.size();++j){
	if (i==j && !is_one(Ni[j]))
	  return false;
	if (i!=j && !is_zero(Ni[j]))
	  return false;
      }
    }
    N=vecteur(N.begin()+lm.coord.size(),N.end());
    for (unsigned i=0;i<N.size();++i){
      vecteur & m =*N[i]._VECTptr;
      for (unsigned j=0;j<m.size();++j){
	if (m[j].type==_MOD)
	  m[j]=*m[j]._MODptr;
      }
      reverse(m.begin(),m.end());
      m=trim(m,0);
    }
    res=N;
    return true;
  }

  // Compute minimal polynomial of s
  template<class tdeg_t>
  bool rur_minpoly(const vectpolymod<tdeg_t> & gbmod,const polymod<tdeg_t> & lm,const polymod<tdeg_t> & s,modint p,vecteur & m,matrice & M){
    int S=int(lm.coord.size()),dim=lm.dim;
    order_t order=lm.order;
    polymod<tdeg_t> TMP1(order,dim);
    vector<unsigned> G(gbmod.size());
    for (unsigned i=0;i<G.size();++i)
      G[i]=i;
    M.clear();
    // set th i-th row of M with coordinates of s^i reduced/gbmod in terms of lm
    vecteur tmp(S);
    tmp[0]=makemod(0,p);
    tmp[S-1]=1;
    M.push_back(tmp);
    polymod<tdeg_t> cur(s);
    for (unsigned i=1;i<=lm.coord.size();++i){
      reducesmallmod(cur,gbmod,G,-1,p,TMP1,false);
      // get coordinates of cur in tmp (make them mod p)
      rur_coordinates(cur,lm,tmp);
      M.push_back(tmp);
      // multiply cur and s
      rur_mult(cur,s,p,TMP1);
      cur.coord.swap(TMP1.coord);
    }
    matrice N(M);
    M.pop_back(); // remove the last one (for further computations, assuming max rank)
    if (!N.empty() && !N.front()._VECTptr->empty()) N=mtran(N);
    vecteur K;
    if (debug_infolevel)
      CERR << CLOCK()*1e-6 << " begin rur ker" << '\n';
    if (!mker(N,K,1,context0) || K.empty() || K.front().type!=_VECT)
      return false;
    if (debug_infolevel)
      CERR << CLOCK()*1e-6 << " end rur ker" << '\n';
    m=*K.front()._VECTptr;
    for (unsigned i=0;i<m.size();++i){
      if (m[i].type==_MOD)
	m[i]=*m[i]._MODptr;
    }
    reverse(m.begin(),m.end());
    m=trim(m,0);
    if (debug_infolevel>1)
      CERR << "Minpoly for " << s << ":" << m << '\n';
    return true;
  }

  template<class tdeg_t>
  void rur_convert_univariate(const vecteur & v,int varno,polymod<tdeg_t> & tmp){
    int vs=int(v.size());
    order_t order=tmp.order;
    tmp.coord.clear();
    index_t l(tmp.dim);
    for (unsigned j=0;int(j)<vs;++j){
      l[varno]=vs-1-j;
      if (v[j].val)
	tmp.coord.push_back(T_unsigned<modint,tdeg_t>(v[j].val,tdeg_t(index_m(l),order)));
    }
  }

  // if radical==-1, shrink the ideal to radical part
  // if radical==1, the ideal is already radical
  // if radical==0, also tries with radical ideal
  // find a separating element, given the groebner basis and the list of leading
  // monomials of a basis of the quotient ideal
  // if true, then separating element is s
  // and s has m as minimal polynomial,
  // M is the list of rows coordinates of powers of s in lm
  // This will not work if the ideal is not a radical ideal
  // In that case, if we get a minimal pol of degree M > lm.size()/2 for one coord.
  // we search for each coordinate a relation polynomial1*coordinate-polynomial2=0
  // where degree(polynomial2)<M and degree(polynomial1) <= lm.size()-M
  // then we must consider particular values of t that cancel gcd(polynomial1,minpoly)
  template<class tdeg_t>
  bool rur_separate(vectpolymod<tdeg_t> & gbmod,polymod<tdeg_t> & lm,modint p,polymod<tdeg_t> & s,vecteur & m,matrice & M,int radical){
    order_t order=lm.order;
    int dim=lm.dim,d=rur_dim(dim,order);
    s.order=order; s.dim=dim;
    // first try coordinates
    vecteur minp(d);
    for (int i=d-1;i>=0;--i){
      s.coord.clear(); m.clear(); M.clear();
      index_t l(dim);
      l[i]=1;
      s.coord.push_back(T_unsigned<modint,tdeg_t>(1,tdeg_t(l,order)));
      if (!rur_minpoly(gbmod,lm,s,p,m,M))
	return false;
      if (m.size()==lm.coord.size()+1)
	return true;
      // keep m in order to shrink to the radical ideal if separation fails
      if (radical<=0)
	minp[i]=m;
    }
    // now try a random small integer linear combination
    if (radical!=-1){
      // 6 july 2020: # of tries 40->100 for
      // eqs:=[c^2 - 3, -a^14 + 85/6*a^12 - 13465/144*a^10 + 54523/144*a^8 - 20819/18*a^6 + 8831/3*a^4 - 7384*a^2 + 10800, -b^14 + 59/4*b^12 - 459/4*b^10 + 8159/16*b^8 - 11777/8*b^6 + 40395/16*b^4 - 10971/4*b^2 + 3267, -1/599040*(53136*b^14 - 692236*b^12 + 4886796*b^10 - 20593959*b^8 + 57747314*b^6 - 116274195*b^4 + 186055404*b^2 - 162887472)*(1008*a^13 - 12240*a^11 + 67325*a^9 - 218092*a^7 + 499656*a^5 - 847776*a^3 + 1063296*a) - 1/112320*(111024*a^14 - 1310760*a^12 + 7843395*a^10 - 35101603*a^8 + 158038072*a^6 - 630801328*a^4 + 1561088256*a^2 - 1489593600)*(56*b^13 - 708*b^11 + 4590*b^9 - 16318*b^7 + 35331*b^5 - 40395*b^3 + 21942*b)];
      // pour j de 1 jusque 100 faire gb:=gbasis(eqs,[c,a,b],rur);fpour;
      // 40 was insufficient for the 11th gbasis rur computation
      for (unsigned essai=0;essai<100;++essai){
	s.coord.clear(); m.clear(); M.clear();
	int n=(3+essai/5);
	int r=int(std_rand()*std::pow(double(n),double(d))/RAND_MAX),r1;
	for (unsigned i=0;int(i)<d;++i){
	  index_t l(dim);
	  l[i]=1;
	  r1=(r%n)-n/2;
	  r/=n;
	  if (r1)
	    s.coord.push_back(T_unsigned<modint,tdeg_t>(r1,tdeg_t(l,order)));
	}
	if (!rur_minpoly(gbmod,lm,s,p,m,M))
	  return false;
	if (m.size()==lm.coord.size()+1)
	  return true;
      }
      if (radical==1)
	return false;
    }
    // shrink ideal and try again
    bool shrinkit=false;
    environment env;
    env.modulo=p;
    env.moduloon=true;
    for (unsigned i=0;int(i)<d;++i){
      if (minp[i].type!=_VECT)
	continue;
      m=*minp[i]._VECTptr;
      if (m.empty())
	continue;
      vecteur m1=derivative(m,&env);
      m1=gcd(m,m1,&env);
      if (m1.size()>1){
	if (debug_infolevel)
	  CERR << "Adding sqrfree part " << m1 << " coordinate " << i << '\n';
	m1=operator_div(m,m1,&env); // m1 is the square free part
	polymod<tdeg_t> m1mod(order,dim);
	rur_convert_univariate(m1,i,m1mod);
	unsigned j;
	for (j=0;j<gbmod.size();++j){
	  if (tdeg_t_greater(m1mod.coord.front().u,gbmod[j].coord.front().u,order))
	    break;
	}
	gbmod.insert(gbmod.begin()+j,m1mod);
	shrinkit=true;
      }
    }
    if (!shrinkit)
      return false;
    vector<unsigned> G;
    if (!in_gbasisf4buchbergermod<tdeg_t>(gbmod,unsigned(gbmod.size()),G,p,/* totdeg */ true,0,0,true))
      return false;
    vectpolymod<tdeg_t> newgb;
    for (unsigned i=0;i<G.size();++i)
      newgb.push_back(gbmod[G[i]]);
    newgb.swap(gbmod);
    lm.coord.clear();
    if (rur_quotient_ideal_dimension(gbmod,lm)<0)
      return false;
    if (radical==-1)
      return true;
    return rur_separate(gbmod,lm,p,s,m,M,1);
  }

  template<class tdeg_t>
  bool rur_convert(const vecteur & v,const polymod<tdeg_t> & lm,polymod<tdeg_t> & res){
    res.coord.clear();
    res.order=lm.order; res.dim=lm.dim;
    if (v.size()>lm.coord.size())
      return false;
    for (unsigned i=0;i<v.size();++i){
      gen coeff=v[i];
      if (!is_zero(coeff))
	res.coord.push_back(T_unsigned<modint,tdeg_t>(coeff.val,lm.coord[i].u));
    }
    return true;
  }

  // set rur to be the list of s,
  // m the minimal polynomial of s as a polymod<tdeg_t> wrt the 1st var
  // and for each coordinate (sqrfree part of m) * coordinate
  // expressed as a polynomial in s (stored in a polymod<tdeg_t> wrt 1st var)
  template<class tdeg_t>
  bool rur_compute(vectpolymod<tdeg_t> & gbmod,polymod<tdeg_t> & lm,polymod<tdeg_t> & lmmodradical,int p,polymod<tdeg_t> & s,vectpolymod<tdeg_t> & rur){
    vecteur m,M,res;
    int dim=lm.dim;
    order_t order=lm.order;
    if (s.coord.empty()){
      // find separating element
      if (!rur_separate(gbmod,lm,p,s,m,M,0))
	return false;
    }
    else {
      // if lm!=lmmodradical, ideal is not radical, recompute radical part
      if (!(lm==lmmodradical)){
	polymod<tdeg_t> s1(s.order,s.dim);
	if (!rur_separate(gbmod,lm,p,s1,m,M,-1))
	  return false;
      }
      // separating element is already known
      if (!rur_minpoly(gbmod,lm,s,p,m,M) || m.size()!=lm.coord.size()+1)
	return false;
    }
    // find the square-free part of m, express it as a polymod<tdeg_t> using M
    environment env;
    env.modulo=p;
    env.moduloon=true;
    vecteur m1=derivative(m,&env);
    m1=gcd(m,m1,&env);
    if (debug_infolevel && m1.size()>1)
      CERR << CLOCK()*1e-6 << " sqrfree mod " << p << ":" << m1 << '\n';
    m1=operator_div(m,m1,&env); // m1 is the square free part
    vecteur m2=derivative(m1,&env); // m2 is the derivative, prime with m1
    // make the "product" with M (rows of M are powers of t)
    gen m3;
    for (unsigned i=0;i<m2.size();++i){
      gen coeff=m2[m2.size()-1-i];
      m3 += smod(coeff*M[i],p);
    }
    m3=smod(m3,p);
    polymod<tdeg_t> mprime(order,dim);
    if (m3.type==_VECT && m3._VECTptr->size()<=lm.coord.size())
      rur_convert(*m3._VECTptr,lm,mprime);
    else
      return false;
    if (debug_infolevel)
      CERR << CLOCK()*1e-6 << " rur linsolve" << '\n';
    if (!rur_linsolve(gbmod,lm,mprime,M,p,res))
      return false;
    // rur=[separating element,sqrfree part of minpoly,derivative of sqrfree part,
    // derivative of sqrfree part*other coordinates]
    rur.clear();
    rur.push_back(s);
    polymod<tdeg_t> tmp(order,dim);
    rur_convert_univariate(m1,0,tmp);
    rur.push_back(tmp);
    rur_convert_univariate(m2,0,tmp);
    rur.push_back(tmp);
    // convert res to rur
    for (unsigned i=0;i<res.size();++i){
      index_t l(dim);
      vecteur & v = *res[i]._VECTptr;
      rur_convert_univariate(v,0,tmp);
      rur.push_back(tmp);
    }
    return true;
  }

  // returns -1 if lm1 is not contained in lm2 and lm2 is not contained in lm1
  // returns 0 if lm1==lm2
  // returns 1 if lm1 contains lm2
  // returns 2 if lm2 contains lm1
  template<class tdeg_t>
  int rur_compare(polymod<tdeg_t> & lm1,polymod<tdeg_t> & lm2){
    unsigned s1=unsigned(lm1.coord.size()),s2=unsigned(lm2.coord.size());
    if (s1==s2){
      if (lm1==lm2)
	return 0;
      return -1;
    }
    if (s1>s2){
      unsigned i=0;
      for (unsigned j=0;j<s2;++i,++j){
	for (;i<s1;++i){
	  if (lm1.coord[i].u==lm2.coord[j].u)
	    break;
	}
	if (i==s1)
	  return -1;
      }
      return 1;
    }
    unsigned j=0;
    for (unsigned i=0;i<s1;++i,++j){
      for (;j<s2;++j){
	if (lm1.coord[i].u==lm2.coord[j].u)
	  break;
      }
      if (j==s2)
	return -1;
    }
    return 2;
  }

  /* ******************
     END RUR UTILITIES
     ****************** */

#ifdef HAVE_LIBPTHREAD
  template<class tdeg_t>
  struct thread_gbasis_t {
    vectpoly8<tdeg_t> current;
    vectpolymod<tdeg_t> resmod;
    vector<unsigned> G;
    int p;
    vector< paire > * reduceto0;
    vector< info_t<tdeg_t> > * f4buchberger_info;
    vector< zinfo_t<tdeg_t> > * zf4buchberger_info;
    bool zdata;
    bool eliminate_flag; // if true, for double revlex order returns only the gbasis part made of polynomials that do not depend on variables to eliminate
    bool interred;
    int parallel; // max number of parallel threads for 1 modular computation
  };

  template<class tdeg_t>
  void * thread_gbasis(void * ptr_){
    thread_gbasis_t<tdeg_t> * ptr=(thread_gbasis_t<tdeg_t> *) ptr_;
    ptr->G.clear();
    if (ptr->zdata){
      if (!zgbasis(ptr->current,ptr->resmod,ptr->G,ptr->p,true,
		   ptr->reduceto0,*ptr->zf4buchberger_info,false,false,ptr->eliminate_flag,true,ptr->parallel,ptr->interred))
	return 0;
    }
    else {
      if (!in_gbasisf4buchbergermod(ptr->current,ptr->resmod,ptr->G,ptr->p,true/*totaldeg*/,
				    ptr->reduceto0,ptr->f4buchberger_info,false))
	return 0;
    }
    return ptr_;
  }
#endif

  template<class tdeg_t>
  bool check_initial_generators(vectpoly8<tdeg_t> & res,const vectpoly8<tdeg_t> & Wi,vector<unsigned> & G,double eps){
    int initial=int(res.size());
    poly8<tdeg_t> tmp0,tmp1,tmp2;
    vectpoly8<tdeg_t> wtmp;
    unsigned j=0,finalchecks=initial;
    if (eps>0)
      finalchecks=giacmin(2*Wi.front().dim,initial);
    if (debug_infolevel)
      CERR << CLOCK()*1e-6 << " begin final check, checking that " << finalchecks << " initial generators belongs to the ideal" << '\n';
    G.resize(Wi.size());
    for (j=0;j<Wi.size();++j)
      G[j]=j;
    vector<bool> Gused(G.size());
    for (j=0;j<finalchecks;++j){
      if (debug_infolevel)
	CERR << "+";
      sort(res[j].coord.begin(),res[j].coord.end(),tdeg_t_sort_t<tdeg_t>(res[j].order));
      reduce(res[j],Wi,G,-1,wtmp,tmp0,tmp1,tmp2,0,&Gused);
      if (!tmp0.coord.empty()){
	break;
      }
      if (debug_infolevel && (j%10==9))
	CERR << j+1 << '\n';
    }
    if (debug_infolevel){
      CERR << '\n' << " Elements used for reduction ";
      for (size_t i=0;i<G.size();++i){
	CERR << (Gused[i]?'+':'-');
      }
      CERR << '\n';
    }
    if (j!=finalchecks){
      if (debug_infolevel){
	CERR << CLOCK()*1e-6 << " final check failure, retrying with another prime " << '\n';
	CERR << "Non-zero remainder " << tmp0 << '\n';
	CERR << "checking res[j], " << j << "<" << initial << '\n';
	CERR << "res[j]=" << res[j] << '\n';
	CERR << "basis candidate " << Wi << '\n';
      }
      return false;
    }
    return true;
  }

  // return 0 (failure), 1 (success), -1: parts of the gbasis reconstructed
  template<class tdeg_t>
  int in_mod_gbasis(vectpoly8<tdeg_t> & res,bool modularcheck,bool zdata,int & rur,GIAC_CONTEXT,gbasis_param_t gbasis_par,int gbasis_logz_age){
    if (debug_infolevel)
      CERR << CLOCK()*1e-6 << " modular gbasis algorithm start, mem " << memory_usage() << '\n';
    bool & eliminate_flag=gbasis_par.eliminate_flag;
    bool interred=gbasis_logz_age==0; // final interreduce
    unsigned initial=unsigned(res.size());
    double eps=proba_epsilon(contextptr); int rechecked=0;
    order_t order={0,0};
    bool multithread_enabled=true;
    // multithread disabled for more than 14 vars because otherwise
    // threads:=2; n:=9;P:=mul(1+x[j]*t,j=0..n-1);
    // X:=[seq(x[j],j=0..n-1)];
    // S:=seq(p[j]-coeff(P,t,j), j=1..n-1);
    //  N:=sum(x[j]^(n-1),j=0..n-1);
    // I:=[N,S]:;eliminate(I,X)
    // segfaults and valgrind does not help...
    for (unsigned i=0;i<res.size();++i){
      const poly8<tdeg_t> & P=res[i];
      if (multithread_enabled && !P.coord.empty())
	multithread_enabled=!P.coord.front().u.vars64();
      order=P.order;
      for (unsigned j=0;j<P.coord.size();++j){
	if (!is_integer(P.coord[j].g)) // improve: accept complex numbers
	  return 0;
      }
    }
    if (order.o!=_REVLEX_ORDER && order.o!=_3VAR_ORDER && order.o!=_7VAR_ORDER && order.o!=_11VAR_ORDER && order.o!=_16VAR_ORDER && order.o!=_32VAR_ORDER && order.o!=_48VAR_ORDER && order.o!=_64VAR_ORDER)
      return 0;
    vectpoly8<tdeg_t> toreinject;
    if (gbasis_par.reinject_begin>=0 && gbasis_par.reinject_end>gbasis_par.reinject_begin){
      toreinject=res; //vectpoly8<tdeg_t>(res.begin()+gbasis_par.reinject_begin,res.begin()+gbasis_par.reinject_end);
      if (gbasis_par.reinject_for_calc>0 && gbasis_par.reinject_for_calc<res.size())
	res.resize(gbasis_par.reinject_for_calc);
      sort(res.begin(),res.end(),tripolymod_tri<poly8<tdeg_t> >(gbasis_logz_age));
    }
    // if (order!=_REVLEX_ORDER) zdata=false;
    vectpoly8<tdeg_t> current,current_orig,current_gbasis,vtmp,afewpolys;
    vectpolymod<tdeg_t> resmod,gbmod;
    poly8<tdeg_t> poly8tmp;
#ifdef EMCC
    // use smaller primes
    gen p=94906249-_floor(giac_rand(contextptr)/32e3,contextptr);
    // gen p=(1<<24)-_floor(giac_rand(contextptr)/32e3,contextptr);
#else
    gen p=(1<<29)-_floor(giac_rand(contextptr)/1e3,contextptr);
#endif
    // unless we are unlucky these lists should contain only 1 element
    vector< vectpoly8<tdeg_t> > V; // list of (chinrem reconstructed) modular groebner basis
    vector< vectpoly8<tdeg_t> > W; // list of rational reconstructed groebner basis
    vector< vectpoly8<tdeg_t> > Wlast;
    vecteur P; // list of associate (product of) modulo
    polymod<tdeg_t> lmmod,lmmodradical,s; vectpolymod<tdeg_t> rurv; // variables for rational univar. reconstr.
    // environment env;
    // env.moduloon=true;
    vector<unsigned> G;
    vector< paire > reduceto0;
    vector< info_t<tdeg_t> > f4buchberger_info;
    f4buchberger_info.reserve(GBASISF4_MAXITER);
    vector<zinfo_t<tdeg_t> > zf4buchberger_info;
    zf4buchberger_info.reserve(GBASISF4_MAXITER);
    mpz_t zu,zd,zu1,zd1,zabsd1,zsqrtm,zq,zur,zr,ztmp;
    mpz_init(zu);
    mpz_init(zd);
    mpz_init(zu1);
    mpz_init(zd1);
    mpz_init(zabsd1);
    mpz_init(zsqrtm);
    mpz_init(zq);
    mpz_init(zur);
    mpz_init(zr);
    mpz_init(ztmp);
    bool ok=true;
#ifdef HAVE_LIBPTHREAD
    int nthreads=(threads_allowed && multithread_enabled)?threads:1,th,parallel=1;
    pthread_t tab[64];
    thread_gbasis_t<tdeg_t> gbasis_param[64];
#else
    int nthreads=1,th,parallel=1;
#endif
    int pend=p.val,p0;
    int recon_n2=-1,recon_n1=-1,recon_n0=-1,recon_added=0,recon_count=0; // reconstr. gbasis element number history
    double augmentgbasis=gbasis_reinject_ratio,prevreconpart=1.0,time1strun=-1.0,time2ndrun=-1.0; current_orig=res; current_gbasis=res;
    // if the ratio of reconstructed is more than augmentgbasis,
    // we clear info and add reconstruction to the gbasis
    for (int count=0;ok;++count,++recon_count){
      p=pend;
      if (count==0 || nthreads==1 || (zdata && augmentgbasis && reduceto0.empty())){
	th=0;
	parallel=nthreads;
      }
      else {
	unsigned sp=simult_primes;
	if (count>=simult_primes_seuil2)
	  sp=simult_primes2;
	if (count>=simult_primes_seuil3)
	  sp=simult_primes3;
	th=giacmin(nthreads-1,sp-1); // no more than simult_primes
	th=giacmin(th,63);
	parallel=nthreads/(th+1);
      }
#ifndef EMCC
      if (count==1 && p.val<(1<<24)){
#ifdef PSEUDO_MOD
	p=(1<<29)-1;
#else
	p=(1<<30)+((1<<30)-1);
#endif
      }
#endif
      // FIXME we should avoid primes that divide one of leading coeff of current_gbasis
      p=prevprime(p-1);
      p0=p.val; // 1st prime used by all threads
      // compute gbasis mod p
      // env.modulo=p;
#ifdef HAVE_LIBPTHREAD
      for (unsigned j=0;j<th;++j){
	gbasis_param[j].current=current_gbasis;
	gbasis_param[j].p=p.val;
	gbasis_param[j].reduceto0=&reduceto0;
	gbasis_param[j].f4buchberger_info=&f4buchberger_info;
	gbasis_param[j].zf4buchberger_info=&zf4buchberger_info;
	gbasis_param[j].zdata=zdata;
	gbasis_param[j].eliminate_flag=eliminate_flag;
	gbasis_param[j].parallel=parallel;
	gbasis_param[j].interred=interred;
	if (count==1)
	  gbasis_param[j].resmod.reserve(resmod.size());
	bool res=true;
	// CERR << "write " << j << " " << p << '\n';
	res=pthread_create(&tab[j],(pthread_attr_t *) NULL,thread_gbasis<tdeg_t>,(void *) &gbasis_param[j]);
	if (res)
	  thread_gbasis<tdeg_t>((void *)&gbasis_param[j]);
#if 1
	p=prevprime(p-1);
#else
	p=nextprime(p+1);
#endif
      }
#endif // thread
      current=current_gbasis;
      G.clear();
      double t_0=CLOCK()*1e-6;
#ifndef KHICAS
      if (debug_infolevel)
	CERR << std::setprecision(15) << t_0 << " begin computing basis modulo " << p << " batch/threads " << th+1 << "/" << parallel << '\n';
#endif
      // CERR << "write " << th << " " << p << '\n';
#ifdef GBASISF4_BUCHBERGER
      if (zdata){
	if (!zgbasis(current,resmod,G,p.val,true,&reduceto0,zf4buchberger_info,false,false,eliminate_flag,true,parallel,interred)){
	  if (augmentgbasis>0)
	    augmentgbasis=2;
	  reduceto0.clear();
	  zf4buchberger_info.clear();
	  zf4buchberger_info.reserve(4*zf4buchberger_info.capacity());
	  G.clear();
	  if (!zgbasis(current,resmod,G,p.val,true/*totaldeg*/,&reduceto0,zf4buchberger_info,false,false,eliminate_flag,true,parallel,interred)){
	    ok=false;
	    break;
	  }
	}
      }
      else {
	resmod.clear();
	if (!in_gbasisf4buchbergermod(current,resmod,G,p.val,true/*totaldeg*/,
				      //		  0,0
				      &reduceto0,&f4buchberger_info,
#if 1
				      false /* not useful */
#else
				      (count==1) /* recompute R and quo at 2nd iteration*/
#endif
				      )){
	  // retry
	  reduceto0.clear();
	  f4buchberger_info.clear(); G.clear();
	  if (!in_gbasisf4buchbergermod(current,resmod,G,p.val,true/*totaldeg*/,&reduceto0,&f4buchberger_info,false)){
	    ok=false;
	    break;
	  }
	  reduceto0.clear();
	  f4buchberger_info.clear();
	}
      }
#else
      if (!in_gbasismod(current,resmod,G,p.val,true,&reduceto0)){
	ok=false;
	break;
      }
      // CERR << "reduceto0 " << reduceto0.size() << '\n';
      //if (!in_gbasis(current,G,&env)) return false;
#endif
#ifdef HAVE_LIBPTHREAD
      // finish threads before chinese remaindering
      void * threadretval[64];
      for (int t=0;t<th;++t){
	threadretval[t]=&threadretval; // non-0 initialization
	pthread_join(tab[t],&threadretval[t]);
      }
#endif
      // cleanup G
      for (unsigned i=0;i<G.size();++i){
	if (resmod[G[i]].coord.empty()){
	  G.erase(G.begin()+i);
	  --i;
	}
      }
      double t_1=CLOCK()*1e-6;
      if (time1strun<0){
	time1strun=t_1-t_0;
	prevreconpart=0.0; // (1.0+res.size())/G.size();
      }
      else {
	if (time2ndrun<0){
	  time2ndrun=(t_1-t_0)/(th+1); // we are computing th+1 primes
	  if (debug_infolevel)
	    CERR << "2nd run " << time2ndrun << " 1st run " << time1strun << '\n';
	  if (time2ndrun<time1strun*gbasis_reinject_speed_ratio
	      || gbasis_par.reinject_for_calc>0
	      //|| time2ndrun<0.5
	      ){
	    // learning is fast enough
	    if (augmentgbasis>0)
	      augmentgbasis=2;
	  }
	}
      }
      pend=p.val; // last prime used
      if (debug_infolevel){
	CERR << t_1 << " end, basis size " << G.size() << " prime number " << count+1 << '\n';
      }
      unsigned i=0;
      for (int t=0;t<=th;++t){
	if (t==th){
	  // extract from current
	  if (rur || gbmod.size()<G.size())
	    gbmod.resize(G.size());
	  for (i=0;i<G.size();++i){
	    gbmod[i]=resmod[G[i]];
	  }
	  p=pend;
	  // CERR << "read " << t << " " << p << '\n';
	}
#ifdef HAVE_LIBPTHREAD
	else {
	  void * ptr_=(void *)threadretval[t]; // saved value from ptr_join
	  if (!ptr_)
	    continue;
	  thread_gbasis_t<tdeg_t> * ptr = (thread_gbasis_t<tdeg_t> *) ptr_;
	  // extract from current
	  if (rur || gbmod.size()<ptr->G.size())
	    gbmod.resize(ptr->G.size());
	  for (i=0;i<ptr->G.size();++i)
	    gbmod[i]=ptr->resmod[ptr->G[i]];
	  p=ptr->p;
	  // CERR << "read " << t << " " << p << '\n';
	  ++count;
	  ++recon_count;
	}
#endif
	if (!ok)
	  continue;
	// remove 0 from gbmod
	for (i=0;i<gbmod.size();){
	  if (gbmod[i].coord.empty())
	    gbmod.erase(gbmod.begin()+i);
	  else
	    ++i;
	}
	// if augmentgbasis>0 (at least) gbmod must be sorted
	//if (augmentgbasis>0)
	sort(gbmod.begin(),gbmod.end(),tripolymod_tri<polymod<tdeg_t> >(gbasis_logz_age));
	if (!rur && gbasis_stop<0)
	  gbmod.resize(-gbasis_stop);
	if (count==0 && V.empty() && gbasis_par.reinject_begin>=0 && gbasis_par.reinject_end>gbasis_par.reinject_begin){
	  // initial reinjection
	  int K=gbasis_par.reinject_end-gbasis_par.reinject_begin;
	  Wlast.push_back(vectpoly8<tdeg_t>());
	  reverse(toreinject.begin(),toreinject.end());
	  toreinject.resize(K);
	  Wlast[0].swap(toreinject);
	  for (int k=0;k<K;++k){
	    if (!chk_equal_mod(Wlast[0][k],gbmod[k],p.val)){
	      *logptr(contextptr) << CLOCK() << " reinjection failure at position " << k << '\n';
	      Wlast.clear();
	      break;
	    }
	  }
	  if (!Wlast.empty()){ // reinjection ok
	    *logptr(contextptr) << CLOCK() << " reinjection success " << K << '\n';
	    V.push_back(vectpoly8<tdeg_t>());
	    W.push_back(vectpoly8<tdeg_t>());
	    convert(gbmod,V.back(),p.val);
	    recon_added=gbasis_par.reinject_end-gbasis_par.reinject_begin;
	    prevreconpart=recon_added/double(gbmod.size());
	    for (int k=0;k<K;++k){
	      V[0][k].coord.clear();
	    }
	    P.push_back(p);
	    continue; // next prime
	  }
	}
	if (debug_infolevel && count==0){
	  CERR << "G= ";
	  for (size_t i=0;i<G.size();++i){
	    CERR << i << ":" << G[i] << "(" << resmod[G[i]].age<<"," << resmod[G[i]].logz << ":" << resmod[G[i]].fromleft << "," << resmod[G[i]].fromright << ")" << '\n';
	  }
	  CERR << "sorted" << '\n';
	  ulonglong nmonoms=0;
	  for (size_t i=0;i<gbmod.size();++i){
	    CERR << i << "(" << gbmod[i].age << "," << gbmod[i].logz << ":" << gbmod[i].fromleft << "," << gbmod[i].fromright << ")" << '\n';
	    nmonoms += gbmod[i].coord.size();
	  }
	  CERR << '\n' << "Partial number of monoms " << nmonoms << '\n';
	}
	// compare gb to existing computed basis
#if 1
	if (rur){
	  gbmod.resize(G.size());
	  int dim=res.front().dim;
	  polymod<tdeg_t> lmtmp(lmmodradical.order,dim);
	  // FIXME rur_quotient_ideal etc. should take care of parameters!
	  if (rur_quotient_ideal_dimension(gbmod,lmtmp)<0){
	    rur=0;
	    continue;
	  }
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " begin modular rur computation" << '\n';
	  if (rur==2){
	    vecteur m,M,res;
	    polymod<tdeg_t> s(order,dim);
	    index_t l(dim);
	    l[dim-1]=1;
	    s.coord.push_back(T_unsigned<modint,tdeg_t>(1,tdeg_t(l,order)));
	    ok=rur_minpoly(gbmod,lmtmp,s,p.val,m,M);
	    rur_convert_univariate(m,dim-1,gbmod[0]);
	    gbmod.resize(1);
	  }
	  else {
	    if (!rur_compute(gbmod,lmtmp,lmmodradical,p.val,s,rurv)){
	      CERR << CLOCK()*1e-6 << "Unable to compute modular rur\n";
	      ok = false; rur = 0;
	      continue;
	    }
	    if (debug_infolevel)
	      CERR << CLOCK()*1e-6 << " end modular rur computation" << '\n';
	    gbmod.swap(rurv); // reconstruct the rur instead of the gbasis
	  } // check for bad primes
	  if (lmmodradical.coord.empty())
	    lmmodradical=lmtmp;
	  else {
	    int i=rur_compare(lmmodradical,lmtmp);
	    if (i!=0){
	      if (i==1) // lmmodradical!=lmtmp and contains lmtmp, bad prime
		continue;
	      // clear existing reconstruction
	      f4buchberger_info.clear();
	      zf4buchberger_info.clear();
	      reduceto0.clear();
	      V.clear(); W.clear(); Wlast.clear(); P.clear();
	      if (i==-1)
		continue;
	      // restart with this prime
	    }
	  }
	} // end if (rur)
	unsigned jpos; gen num,den;
	if (debug_infolevel>2)
	  CERR << "p=" << p << ":" << gbmod << '\n';
	for (i=0;i<V.size();++i){
	  if (W.size()<V.size())
	    W.resize(V.size());
	  if (Wlast.size()<V.size()){
	    Wlast.resize(V.size());
	    Wlast.back().reserve(gbmod.size());
	  }
	  if (V[i].size()!=gbmod.size())
	    continue;
	  for (jpos=recon_added;jpos<gbmod.size();++jpos){
	    if (V[i][jpos].coord.empty() && gbmod[jpos].coord.empty())
	      continue;
	    if (V[i][jpos].coord.empty())
	      break;
	    if (gbmod[jpos].coord.empty())
	      break;
	    if (V[i][jpos].coord.front().u!=gbmod[jpos].coord.front().u)
	      break;
	  }
	  if (jpos!=gbmod.size()){
	    rechecked=0;
	    continue;
	  }
	  jpos=recon_added; // 0 or recon_added (do not check already reconstructed)
	  // check existing Wlast
	  for (;jpos<Wlast[i].size();++jpos){
	    if (!chk_equal_mod(Wlast[i][jpos],gbmod[jpos],p.val)){
	      Wlast[i].resize(jpos);
	      rechecked=0;
	      break;
	    }
	  }
	  if (jpos!=Wlast[i].size() || P[i].type==_INT_){
	    // CERR << jpos << '\n';
	    // IMPROVE: make it work for rur!
	    if (!rur && eps>0 && P[i].type==_INT_ && recon_added==0 && gbasis_stop!=0){
	      // check for non modular gb with early reconstruction */
	      // first build a candidate in early with V[i]
	      vectpoly8<tdeg_t> early(V[i]);
	      int d;
	      for (jpos=0;jpos<early.size();++jpos){
		d=1;
		if (!findmultmod(early[jpos],P[i].val,d)){
		  if (debug_infolevel>1)
		    COUT << "early reconstr. failure pos " << jpos << " P=" << early[jpos] << " d=" << d << " modulo " << P[i].val << '\n';
		  break;
		}
		int s=int(early[jpos].coord.size());
		for (int k=0;k<s;++k){
		  early[jpos].coord[k].g=smod(longlong(early[jpos].coord[k].g.val)*d,P[i].val);
		}
	      }
	      // then check
	      if (jpos==early.size()){
		for (jpos=0;jpos<early.size();++jpos){
		  polymod<tdeg_t> tmp(gbmod[jpos]);
		  smallmultmod(early[jpos].coord.front().g.val,tmp,p.val);
		  if (!chk_equal_mod(early[jpos],tmp,p.val)){
		    if (debug_infolevel>1)
		      COUT << "early recons. failure jpos=" << jpos << " " << early[jpos] << " " << tmp << " modulo " << p.val << '\n';
		    break;
		  }
		}
		rechecked=0;
		if (jpos==early.size() && (eliminate_flag || check_initial_generators(res,early,G,eps))){
		  if (debug_infolevel)
		    CERR << CLOCK()*1e-6 << " end final check " << '\n';
		  swap(res,early);
		  mpz_clear(zd);
		  mpz_clear(zu);
		  mpz_clear(zu1);
		  mpz_clear(zd1);
		  mpz_clear(zabsd1);
		  mpz_clear(zsqrtm);
		  mpz_clear(zq);
		  mpz_clear(zur);
		  mpz_clear(zr);
		  mpz_clear(ztmp);
		  return 1;
		}
	      } // end jpos==early.size()
	    }
	    break; // find another prime
	  }
	  for (;jpos<V[i].size();++jpos){
	    unsigned Vijs=unsigned(V[i][jpos].coord.size());
	    if (Vijs!=gbmod[jpos].coord.size()){
	      rechecked=0;
	      if (debug_infolevel>1)
		CERR << jpos << '\n';
	      break;
	    }
	    //Vijs=1;
	    Vijs/=2;
	    if (Vijs && V[i][jpos].coord[Vijs].g.type==_ZINT){
	      if (!in_fracmod(P[i],V[i][jpos].coord[Vijs].g,
			      zd,zd1,zabsd1,zu,zu1,zur,zq,zr,zsqrtm,ztmp,num,den)){
		rechecked=0;
		if (debug_infolevel>1)
		  CERR << jpos << '\n';
		break;
	      }
	      modint gg=gbmod[jpos].coord[Vijs].g;
	      if (!chk_equal_mod(num/den,gg,p.val)){
		rechecked=0;
		if (debug_infolevel>1)
		  CERR << jpos << '\n';
		break;
	      }
	    }
	    if (!fracmod(V[i][jpos],P[i],
			 zd,zd1,zabsd1,zu,zu1,zur,zq,zr,zsqrtm,ztmp,
			 poly8tmp)){
	      rechecked=0;
	      CERR << CLOCK()*1e-6 << " reconstruction failure at position " << jpos << '\n';
	      break;
	    }
	    if (rur && !poly8tmp.coord.empty() && !chk_equal_mod(poly8tmp.coord.front().g,gbmod[jpos].coord.front().g,p.val)){
	      rechecked=0;
	      break;
	    }
	    if (!chk_equal_mod(poly8tmp,gbmod[jpos],p.val)){
	      rechecked=0;
	      break;
	    }
	    poly8<tdeg_t> tmptmp(poly8tmp.order,poly8tmp.dim);
	    Wlast[i].push_back(tmptmp);
	    Wlast[i].back().coord.swap(poly8tmp.coord);
	  }
	  if (debug_infolevel>0)
	    CERR << CLOCK()*1e-6 << " unstable mod " << p << " from " << V[i].size() << " reconstructed " << Wlast[i].size() << " (#" << i << ")" << '\n';
	  // possible improvement: if t==th and i==0 and Wlast.size()/V[i].size()
	  // has increased significantly
	  // it might be a good idea to add it's component
	  // to current, and clear info (if zdata: reduceto0, zf4buchberger_info)
	  recon_n2=recon_n1;
	  recon_n1=recon_n0;
	  recon_n0=Wlast[i].size();
	  if (eps>1e-20 &&
	      // recon_n2==recon_n1 && recon_n1==recon_n0 &&
	      zdata && augmentgbasis && t==th && i==0){
	    double reconpart=recon_n2/double(V[i].size());
	    if (!rur && recon_n0/double(V[i].size())<0.95 &&
		(reconpart-prevreconpart>augmentgbasis
		 // || (reconpart>prevreconpart && recon_count>=giacmax(128,th*4))
		 )
		){
	      CERR << CLOCK()*1e-6 << " adding reconstructed ideal generators " << recon_n2 << " (reconpart " << reconpart << " prev " << prevreconpart << " augment " << augmentgbasis << " recon_count " << recon_count << " th " << th << " recon_n2 " << recon_n2 << " V[i] " << V[i].size() << ")" << '\n';
	      recon_count=0;
	      prevreconpart=reconpart;
	      //current_gbasis=current_orig;
	      int insertpos=0;
	      for (int k=recon_added;k<recon_n2;++k){
		V[i][k].coord.clear();
		poly8<tdeg_t> tmp=Wlast[i][k];
		cleardeno(tmp);
		for (;insertpos<current_gbasis.size();++insertpos){
		  if (tdeg_t_greater(current_gbasis[insertpos].coord.front().u,tmp.coord.front().u,order)){
		    if (!(current_gbasis[insertpos]==tmp)){
		      current_gbasis.insert(current_gbasis.begin()+insertpos,tmp);
		      ++insertpos;
		    }
		    break;
		  }
		}
		if (insertpos==current_gbasis.size()){
		  current_gbasis.push_back(tmp);
		  ++insertpos;
		}
	      }
	      recon_added=recon_n2; // Wlast[i].size();
	      CERR << CLOCK()*1e-6 << " # new ideal generators " << current_gbasis.size() << '\n';
	      reduceto0.clear();
	      zf4buchberger_info.clear();
	      if (gbasis_logz_age){
		res.swap(current_gbasis);
		mpz_clear(zd);
		mpz_clear(zu);
		mpz_clear(zu1);
		mpz_clear(zd1);
		mpz_clear(zabsd1);
		mpz_clear(zsqrtm);
		mpz_clear(zq);
		mpz_clear(zur);
		mpz_clear(zr);
		mpz_clear(ztmp);
		return -1;
	      }
	    }
	  }
	  break;
	} // end for loop on i
	if (i==V.size()){
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " creating reconstruction #" << i << '\n';
	  // not found
	  V.push_back(vectpoly8<tdeg_t>());
	  convert(gbmod,V.back(),p.val);
	  W.push_back(vectpoly8<tdeg_t>()); // no reconstruction yet, wait at least another prime
	  Wlast.push_back(vectpoly8<tdeg_t>());
	  P.push_back(p);
	  continue; // next prime
	}
	if (!rur && gbasis_stop<0 && recon_n0>=-gbasis_stop){
	  if (recon_n2<-gbasis_stop)
	    continue;
	  // stop here
	  W[i]=Wlast[i];
	  W[i].resize(recon_n2);
	  cleardeno(W[i]); // clear denominators
	  CERR << CLOCK()*1e-6 << " Max number of generators reconstructed " << jpos << ">=" << -gbasis_stop << '\n';
	  swap(res,W[i]);
	  mpz_clear(zd);
	  mpz_clear(zu);
	  mpz_clear(zu1);
	  mpz_clear(zd1);
	  mpz_clear(zabsd1);
	  mpz_clear(zsqrtm);
	  mpz_clear(zq);
	  mpz_clear(zur);
	  mpz_clear(zr);
	  mpz_clear(ztmp);
	  return 1;
	}
	if (jpos<gbmod.size()){
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " i=" << i << " begin chinese remaindering " << p << " (" << count+(t==th) << ")" << '\n';
	  int r=chinrem(V[i],P[i],gbmod,p.val,poly8tmp,recon_added); // IMPROVE: maybe start at jpos in V[i]? at least start at recon_added
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " end chinese remaindering" << '\n';
	  if (r==-1){
	    ok=false;
	    continue;
	  }
	  P[i]=P[i]*p;
	  continue; // next prime
	}
	else { // final check
	  W[i]=Wlast[i];
	  if (!rur){
	    if (debug_infolevel)
	      CERR << CLOCK()*1e-6 << " stable, clearing denominators " << '\n';
	    cleardeno(W[i]); // clear denominators
	  }
	  ++rechecked;
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " end rational reconstruction " << '\n';
	  // now check if W[i] is a Groebner basis over Q, if so it's the answer
	  if (rur){
	    // a final check could be performed by replacing
	    // res[3..end]/res[2] in the initial gbasis element and check if it's 0
	    swap(res,W[i]);
	    mpz_clear(zd);
	    mpz_clear(zu);
	    mpz_clear(zu1);
	    mpz_clear(zd1);
	    mpz_clear(zabsd1);
	    mpz_clear(zsqrtm);
	    mpz_clear(zq);
	    mpz_clear(zur);
	    mpz_clear(zr);
	    mpz_clear(ztmp);
	    return 1;
	  }
	  // first verify that the initial generators reduce to 0
	  if (!eliminate_flag && !check_initial_generators(res,W[i],G,eps))
	    continue;
	  if (int(W[i].size())<=GBASIS_DETERMINISTIC)
	    eps=0;
	  if (eliminate_flag && eps==0)
	    eps=1e-7;
	  double eps2=std::pow(double(p.val),double(rechecked))*eps;
	  // recheck by computing gbasis modulo another prime
	  if (eps2>0 && eps2<1){
	    if (debug_infolevel)
	      CERR << CLOCK()*1e-6 << " Final check successful, running another prime to increase confidence." << '\n';
	    continue;
	  }
	  if (eps>0){
	    double terms=0;
	    int termsmin=RAND_MAX; // estimate of the number of terms of a reduced non-0 spoly
	    for (unsigned k=0;k<W[i].size();++k){
	      terms += W[i][k].coord.size();
	      termsmin = giacmin(termsmin,unsigned(W[i][k].coord.size()));
	    }
	    termsmin = 7*(2*termsmin-1);
	    int epsp=P[i].type==_ZINT?mpz_sizeinbase(*P[i]._ZINTptr,10):8-int(std::ceil(2*std::log10(terms)));
	    if (epsp>termsmin)
	      epsp=termsmin;
	    *logptr(contextptr) << gettext("Running a probabilistic check for the reconstructed Groebner basis. If successfull, error probability is less than ") << eps << gettext(" and is estimated to be less than 10^-") << epsp << gettext(". Use proba_epsilon:=0 to certify (this takes more time).") << '\n';
	  }
	  G.clear();
	  if (eps<1.01e-10){
	    // check modulo another prime that W[i] is a gbasis
	    vector<unsigned> G;
	    vectpoly8<tdeg_t> res_(W[i]);
	    vectpolymod<tdeg_t> resmod;
	    vector< zinfo_t<tdeg_t> > zf4buchberger_info;
	    int p=268435399;
	    if (debug_infolevel)
	      CERR << CLOCK()*1e-6 << " Checking that the basis is a gbasis modulo " << p << '\n';
	    if (!zgbasis<tdeg_t>(res_,resmod,G,p,true,0,zf4buchberger_info,false,false,false,false,threads /* parallel*/,true))
	      return 0;
	    sort(resmod.begin(),resmod.end(),tripolymod_tri<polymod<tdeg_t> >(false));
	    sort(W[i].begin(),W[i].end(),tripolymod_tri<poly8<tdeg_t> >(false));
	    for (size_t jpos=0;jpos<G.size();++jpos){
	      if (!chk_equal_mod(W[i][jpos],resmod[jpos],p))
		return 0;
	    }
	    if (debug_infolevel)
	      CERR << CLOCK()*1e-6 << " Check successful mod " << p << '\n';
	  }
	  if (eps2<1 && !is_gbasis(W[i],eps2,modularcheck)){
	    ok=false;
	    continue; // in_gbasis(W[i],G,0,true);
	  }
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " end final check " << '\n';
	  swap(res,W[i]);
	  mpz_clear(zd);
	  mpz_clear(zu);
	  mpz_clear(zu1);
	  mpz_clear(zd1);
	  mpz_clear(zabsd1);
	  mpz_clear(zsqrtm);
	  mpz_clear(zq);
	  mpz_clear(zur);
	  mpz_clear(zr);
	  mpz_clear(ztmp);
	  return 1;
	}
#else
	for (i=0;i<V.size();++i){
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " i= " << i << " begin chinese remaindering" << '\n';
	  int r=chinrem(V[i],P[i],gbmod,p.val,poly8tmp);
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " end chinese remaindering" << '\n';
	  if (r==-1){
	    ok=false;
	    break;
	  }
	  if (r==0){
	    CERR << CLOCK()*1e-6 << " leading terms do not match with reconstruction " << i << " modulo " << p << '\n';
	    continue;
	  }
	  // found one! V is already updated, update W
	  if (W.size()<V.size())
	    W.resize(V.size());
	  if (Wlast.size()<V.size())
	    Wlast.resize(V.size());
	  P[i]=P[i]*p;
	  unsigned jpos=0;
	  // afewpolys.clear();
	  for (;jpos<V[i].size();++jpos){
	    if (int(Wlast[i].size())>jpos && chk_equal_mod(Wlast[i][jpos],gb[jpos],p.val)){
	      if (afewpolys.size()<=jpos)
		afewpolys.push_back(Wlast[i][jpos]);
	      else {
		if (!(afewpolys[jpos]==Wlast[i][jpos]))
		  afewpolys[jpos]=Wlast[i][jpos];
	      }
	    }
	    else {
	      if (!fracmod(V[i][jpos],P[i],
			   zd,zd1,zabsd1,zu,zu1,zur,zq,zr,zsqrtm,ztmp,
			   poly8tmp)){
		CERR << CLOCK()*1e-6 << " reconstruction failure at position " << jpos << '\n';
		break;
	      }
	      if (afewpolys.size()<=jpos){
		poly8<tdeg_t> tmp(poly8tmp.order,poly8tmp.dim);
		afewpolys.push_back(tmp);
	      }
	      afewpolys[jpos].coord.swap(poly8tmp.coord);
	    }
	    if (Wlast[i].size()>jpos && !(afewpolys[jpos]==Wlast[i][jpos])){
	      if (debug_infolevel){
		unsigned j=0,js=giacmin(afewpolys[jpos].coord.size(),Wlast[i][jpos].coord.size());
		for (;j<js;++j){
		  if (!(afewpolys[jpos].coord[j]==Wlast[i][jpos].coord[j]))
		    break;
		}
		CERR << "Diagnostic: chinrem reconstruction mismatch at positions " << jpos << "," << j << '\n';
		if (j<js)
		  CERR << gb[jpos].coord[j].g << "*" << gb[jpos].coord[j].u << '\n';
		else
		  CERR << afewpolys[jpos].coord.size() << "," << Wlast[i][jpos].coord.size() << '\n';
	      }
	      afewpolys.resize(jpos+1);
	      break;
	    }
	    if (jpos > Wlast[i].size()*1.35+2 )
	      break;
	  }
	  if (afewpolys!=Wlast[i]){
	    swap(afewpolys,Wlast[i]);
	    if (debug_infolevel>0)
	      CERR << CLOCK()*1e-6 << " unstable mod " << p << " from " << V[i].size() << " reconstructed " << Wlast[i].size() << '\n';
	    break;
	  }
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " stable, clearing denominators " << '\n';
	  W[i]=Wlast[i];
	  cleardeno(W[i]); // clear denominators
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " end rational reconstruction " << '\n';
	  // now check if W[i] is a Groebner basis over Q, if so it's the answer
	  // first verify that the initial generators reduce to 0
	  poly8<tdeg_t> tmp0,tmp1,tmp2;
	  vectpoly8<tdeg_t> wtmp;
	  unsigned j=0,finalchecks=initial;
	  if (eps>0)
	    finalchecks=giacmin(2*W[i].front().dim,initial);
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " begin final check, checking that " << finalchecks << " initial generators belongs to the ideal" << '\n';
	  G.resize(W[i].size());
	  for (j=0;j<W[i].size();++j)
	    G[j]=j;
	  for (j=0;j<finalchecks;++j){
	    if (debug_infolevel)
	      CERR << "+";
	    reduce(res[j],W[i],G,-1,wtmp,tmp0,tmp1,tmp2,0);
	    if (!tmp0.coord.empty()){
	      break;
	    }
	    if (debug_infolevel	&& (j%10==9))
	      CERR << j+1 << '\n';
	  }
	  if (j!=finalchecks){
	    if (debug_infolevel){
	      CERR << CLOCK()*1e-6 << " final check failure, retrying with another prime " << '\n';
	      CERR << "Non-zero remainder " << tmp0 << '\n';
	      CERR << "checking res[j], " << j << "<" << initial << '\n';
	      CERR << "res[j]=" << res[j] << '\n';
	      CERR << "basis candidate " << W[i] << '\n';
	    }
	    break;
	}
	  /* (final check requires that we have reconstructed a Groebner basis,
	     Modular algorithms for computing Groebner bases Elizabeth A. Arnold
	     Journal of Symbolic Computation 35 (2003) 403–419)
	  */
#if 1
	  if (int(W[i].size())<=GBASIS_DETERMINISTIC)
	    eps=0;
	  if (eps>0){
	    double terms=0;
	    int termsmin=RAND_MAX; // estimate of the number of terms of a reduced non-0 spoly
	    for (unsigned k=0;k<W[i].size();++k){
	      terms += W[i][k].coord.size();
	      termsmin = giacmin(termsmin,W[i][k].coord.size());
	    }
	    termsmin = 7*(2*termsmin-1);
	    int epsp=mpz_sizeinbase(*P[i]._ZINTptr,10)-int(std::ceil(2*std::log10(terms)));
	    if (epsp>termsmin)
	      epsp=termsmin;
	    *logptr(contextptr) << gettext("Running a probabilistic check for the reconstructed Groebner basis. If successfull, error probability is less than ") << eps << gettext(" and is estimated to be less than 10^-") << epsp << gettext(". Use proba_epsilon:=0 to certify (this takes more time).") << '\n';
	  }
	  G.clear();
	  if (eps<6e-8 && !is_gbasis(W[i],eps*1.677e7,modularcheck)){
	    ok=false;
	    break; // in_gbasis(W[i],G,0,true);
	  }
#endif
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " end final check " << '\n';
	  swap(res,W[i]);
	  mpz_clear(zd);
	  mpz_clear(zu);
	  mpz_clear(zu1);
	  mpz_clear(zd1);
	  mpz_clear(zabsd1);
	  mpz_clear(zsqrtm);
	  mpz_clear(zq);
	  mpz_clear(zur);
	  mpz_clear(zr);
	  mpz_clear(ztmp);
	  return 1;
	} // end for (i<V.size())
	if (i==V.size()){
	  if (debug_infolevel)
	    CERR << CLOCK()*1e-6 << " creating reconstruction #" << i << '\n';
	  // not found
	  V.push_back(gb);
	  W.push_back(vectpoly8<tdeg_t>()); // no reconstruction yet, wait at least another prime
	  Wlast.push_back(vectpoly8<tdeg_t>());
	  P.push_back(p);
	}
#endif
      } // end loop on threads
    } //end for int count
    mpz_clear(zd);
    mpz_clear(zu);
    mpz_clear(zu1);
    mpz_clear(zd1);
    mpz_clear(zabsd1);
    mpz_clear(zsqrtm);
    mpz_clear(zq);
    mpz_clear(zur);
    mpz_clear(zr);
    mpz_clear(ztmp);
    return 0;
  }

  template<class tdeg_t>
  bool mod_gbasis(vectpoly8<tdeg_t> & res,bool modularcheck,bool zdata,int & rur,GIAC_CONTEXT,gbasis_param_t gbasis_param){
    int gbasis_logz_age=gbasis_logz_age_sort;
    for (;;){
      int tmp=in_mod_gbasis(res,modularcheck,zdata,rur,contextptr,gbasis_param,gbasis_logz_age);
      if (tmp!=-1) // -1 means part of the gbasis has been reconstructed
	return tmp;
      if (gbasis_logz_age)
	gbasis_logz_age=0; // special sorting is not meaningfull after 1 reinjection, and we want to have interreduction
    }
  }
#ifndef BIGENDIAN
#define GBASIS_SWAP
#endif

#if !defined NO_STDEXCEPT && !defined BIGENDIAN
  #define GIAC_TDEG_T14
#endif

  // other tdeg_t types
#ifdef GIAC_TDEG_T14
#undef INT128 // it's slower!
  struct tdeg_t14 {
    bool vars64() const { return false;}
    int hash_index(void * ptr_) const {
      // if (!ptr_)
	return -1;
    }
    bool add_to_hash(void *ptr_,int no) const {
      return false;
    }
    void dbgprint() const;
    // data
    union {
      unsigned char tab[16]; // tab[0] and 1 is for total degree
      struct {
	unsigned char tdeg;
	unsigned char tdeg2;
	order_t order_;
	longlong * ui;
      };
    };
    int front(){ return tab[2];}
    // methods
    inline unsigned selection_degree(order_t order) const {
#ifdef GBASIS_SELECT_TOTAL_DEGREE
      return total_degree(order);
#endif
      return tdeg;
    }
    inline unsigned total_degree(order_t order) const {
      return tdeg+tdeg2;
    }
    // void set_total_degree(unsigned d) { tab[0]=d;}
    tdeg_t14() {
      longlong * ptr = (longlong *) tab;
      ptr[1]=ptr[0]=0;
    }
    tdeg_t14(int i){
      longlong * ptr = (longlong *) tab;
      ptr[1]=ptr[0]=0;
    }
    void get_tab(short * ptr,order_t order) const {
#ifdef GBASIS_SWAP
      tdeg_t14 t(*this);
      swap_indices14(t.tab);
#else
      const tdeg_t14 & t=*this;
#endif
      ptr[0]=t.tab[0];
      for (unsigned i=1;i<15;++i)
	ptr[i]=t.tab[i+1];
    }
    tdeg_t14(const index_m & lm,order_t order){
      longlong * ptr_ = (longlong *) tab;
      ptr_[1]=ptr_[0]=0;
      unsigned char * ptr=tab;
      vector<deg_t>::const_iterator it=lm.begin(),itend=lm.end();
      if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
	unsigned td=sum_degree(lm);
	if (td>=128)
	  gensizeerr("Degree too large");
	*ptr=td;
	++ptr;
	*ptr=0;
	++ptr;
      }
      if (order.o==_REVLEX_ORDER){
	for (--itend,--it;it!=itend;++ptr,--itend)
	  *ptr=*itend;
      }
      else {
	for (;it!=itend;++ptr,++it)
	  *ptr=*it;
      }
#ifdef GBASIS_SWAP
      swap_indices14(tab);
#endif
    }
  };


#ifdef NSPIRE
  template<class T>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os,const tdeg_t14 & x){
    os << "[";
    for (unsigned i=0; i<14;++i){
      os << unsigned(x.tab[i+2]) << ",";
    }
    return os << "]";
  }
#else
  ostream & operator << (ostream & os,const tdeg_t14 & x){
    os << "[";
    for (unsigned i=0; i<14;++i){
      os << unsigned(x.tab[i+2]) << ",";
    }
    return os << "]";
  }
#endif
  void tdeg_t14::dbgprint() const { COUT << * this << '\n'; }
  inline tdeg_t14 & operator += (tdeg_t14 & x,const tdeg_t14 & y){
#ifdef INT128
    * (uint128_t *) &x += * (const uint128_t *) &y;
#else
    // ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    *((ulonglong *)&x) += *((ulonglong *)&y);
    ((ulonglong *)&x)[1] += ((ulonglong *)&y)[1];
#endif
    if (x.tab[0]>=128){
      gensizeerr("Degree too large");
    }
    return x;
  }
  inline tdeg_t14 operator + (const tdeg_t14 & x,const tdeg_t14 & y){
    tdeg_t14 res(x);
    return res += y;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]+ytab[0];
    ztab[1]=xtab[1]+ytab[1];
    return res;
  }
  inline void add(const tdeg_t14 & x,const tdeg_t14 & y,tdeg_t14 & res,int dim){
#ifdef INT128
    * (uint128_t *) &res = * (const uint128_t *) &x + * (const uint128_t *) &y;
#else
    // ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    *((ulonglong *)&res)=*((ulonglong *)&x)+*((ulonglong *)&y);
    ((ulonglong *)&res)[1]=((ulonglong *)&x)[1]+((ulonglong *)&y)[1];
#endif
    if (res.tab[0]>=128)
      gensizeerr("Degree too large");
  }
  tdeg_t14 operator - (const tdeg_t14 & x,const tdeg_t14 & y){
    tdeg_t14 res;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]-ytab[0];
    ztab[1]=xtab[1]-ytab[1];
    return res;
  }
  inline bool operator == (const tdeg_t14 & x,const tdeg_t14 & y){
#ifdef INT128
    return * (const uint128_t *) &x == * (uint128_t *) &y;
#else
    return ((longlong *) x.tab)[0] == ((longlong *) y.tab)[0] && ((longlong *) x.tab)[1] == ((longlong *) y.tab)[1];
#endif
  }
  inline bool operator != (const tdeg_t14 & x,const tdeg_t14 & y){
    return !(x==y);
  }

  static inline int tdeg_t14_revlex_greater (const tdeg_t14 & x,const tdeg_t14 & y){
#ifdef GBASIS_SWAP
#if 0
    longlong *xtab=(longlong *)&x,*ytab=(longlong *)&y;
    if (longlong a=*xtab-*ytab) // tdeg test already donne by caller
      return a<=0?1:0;
    if (longlong a=xtab[1]-ytab[1])
      return a<=0?1:0;
#else
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    if (xtab[0]!=ytab[0]) // tdeg test already donne by caller
      return xtab[0]<=ytab[0]?1:0;
    if (xtab[1]!=ytab[1])
      return xtab[1]<=ytab[1]?1:0;
#endif
    return 2;
#else // GBASIS_SWAP
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>=y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      if (x.tab[3]!=y.tab[3])
	return x.tab[3]<=y.tab[3]?1:0;
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]>=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      if (x.tab[11]!=y.tab[11])
	return x.tab[11]<=y.tab[11]?1:0;
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]<=y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
    }
    return 2;
#endif // GBASIS_SWAP
  }

  int tdeg_t14_lex_greater (const tdeg_t14 & x,const tdeg_t14 & y){
#ifdef GBASIS_SWAP
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    ulonglong X=*xtab, Y=*ytab;
    if (X!=Y){
      if ( (X & 0xffff) != (Y &0xffff))
	return (X&0xffff)>=(Y&0xffff)?1:0;
      return X>=Y?1:0;
    }
    if (xtab[1]!=ytab[1])
      return xtab[1]>=ytab[1]?1:0;
    return 2;
#else
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]>y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]>y.tab[2]?1:0;
      if (x.tab[3]!=y.tab[3])
	return x.tab[2]>y.tab[2]?1:0;
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]>y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]>y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]>y.tab[6]?1:0;
      return x.tab[7]>y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]>y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]>y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]>y.tab[10]?1:0;
      if (x.tab[11]!=y.tab[11])
	return x.tab[11]>y.tab[11]?1:0;
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]>y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]>y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]>y.tab[14]?1:0;
      return x.tab[15]>y.tab[15]?1:0;
    }
    return 2;
#endif
  }

  inline int tdeg_t_greater(const tdeg_t14 & x,const tdeg_t14 & y,order_t order){
    short X=x.tab[0];
    if (X!=y.tab[0]) return X>y.tab[0]?1:0; // since tdeg is tab[0] for plex
    if (order.o==_REVLEX_ORDER)
      return tdeg_t14_revlex_greater(x,y);
    return tdeg_t14_lex_greater(x,y);
  }
  inline bool tdeg_t_strictly_greater (const tdeg_t14 & x,const tdeg_t14 & y,order_t order){
    return !tdeg_t_greater(y,x,order); // total order
  }

#ifdef INT128
  uint128_t mask4=(((uint128_t) mask2)<<64)|mask2;
#endif

  inline bool tdeg_t_all_greater(const tdeg_t14 & x,const tdeg_t14 & y,order_t order){
#ifdef INT128
    return !((* (const uint128_t *) &x - * (const uint128_t *) &y) & mask4);
#else
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    if ((xtab[0]-ytab[0]) & mask2)
      return false;
    if ((xtab[1]-ytab[1]) & mask2)
      return false;
    return true;
#endif
  }

  // 1 (all greater), 0 (unknown), -1 (all smaller)
  int tdeg_t_compare_all(const tdeg_t14 & x,const tdeg_t14 & y,order_t order){
    int res=0;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    longlong tmp=xtab[0]-ytab[0];
    if (tmp & mask2){
      if (res==1 || ((-tmp) & mask2)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    tmp=xtab[1]-ytab[1];
    if (tmp & mask2){
      if (res==1 || ((-tmp) & mask2)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    return res;
  }

  void index_lcm(const tdeg_t14 & x,const tdeg_t14 & y,tdeg_t14 & z,order_t order){
    int t=0;
    const unsigned char * xtab=&x.tab[2],*ytab=&y.tab[2];
    unsigned char *ztab=&z.tab[2];
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 2
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 3
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 4
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 5
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 6
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 7
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 8
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 12
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
    if (t>=128){
      gensizeerr("Degree too large");
    }
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
      z.tab[0]=t;
    }
    else {
      z.tab[0]=(x.tab[0]>y.tab[0])?x.tab[0]:y.tab[0];
    }
  }

  void index_lcm_overwrite(const tdeg_t14 & x,const tdeg_t14 & y,tdeg_t14 & z,order_t order){
    index_lcm(x,y,z,order);
  }

  void get_index(const tdeg_t14 & x_,index_t & idx,order_t order,int dim){
    idx.resize(dim);
#ifdef GBASIS_SWAP
    tdeg_t14 x(x_);
    swap_indices14(x.tab);
#else
    const tdeg_t14 & x= x_;
#endif
    const unsigned char * ptr=x.tab+2;
    if (order.o==_REVLEX_ORDER){
      for (int i=1;i<=dim;++ptr,++i)
	idx[dim-i]=*ptr;
    }
    else {
      for (int i=0;i<dim;++ptr,++i)
	idx[i]=*ptr;
    }
  }

  bool disjoint(const tdeg_t14 & a,const tdeg_t14 & b,order_t order,short dim){
    const unsigned char * it=a.tab+2, * jt=b.tab+2;
#ifdef GBASIS_SWAP
    const unsigned char * itend=it+14;
#else
    const unsigned char * itend=it+dim;
#endif
    for (;it<itend;++jt,++it){
      if (*it && *jt)
	return false;
    }
    return true;
  }
#endif

#undef GROEBNER_VARS
#define GROEBNER_VARS 11

  struct tdeg_t11 {
    bool vars64() const { return false;}
    int hash_index(void * ptr_) const {
      // if (!ptr_)
	return -1;
    }
    bool add_to_hash(void *ptr_,int no) const {
      return false;
    }
    void dbgprint() const;
    // data
    union {
      short tab[GROEBNER_VARS+1];
      struct {
	short tdeg; // actually it's twice the total degree+1
	short tdeg2;
	order_t order_;
	longlong * ui;
      };
    };
    int front(){ return tab[1];}
    // methods
    inline unsigned selection_degree(order_t order) const {
      return tab[0];
    }
    inline unsigned total_degree(order_t order) const {
      // works only for revlex and tdeg
      return tab[0];
    }
    // void set_total_degree(unsigned d) { tab[0]=d;}
    tdeg_t11() {
      longlong * ptr = (longlong *) tab;
      ptr[2]=ptr[1]=ptr[0]=0;
    }
    tdeg_t11(int i){
      longlong * ptr = (longlong *) tab;
      ptr[2]=ptr[1]=ptr[0]=0;
    }
    void get_tab(short * ptr,order_t order) const {
      for (unsigned i=0;i<=11;++i)
	ptr[i]=tab[i];
#ifdef GBASIS_SWAP
      swap_indices11(ptr);
#endif
      ptr[15]=ptr[14]=ptr[13]=ptr[12]=0;
    }
    tdeg_t11(const index_m & lm,order_t order){
      longlong * ptr_ = (longlong *) tab;
      ptr_[2]=ptr_[1]=ptr_[0]=0;
      short * ptr=tab;
      // tab[GROEBNER_VARS]=order;
      vector<deg_t>::const_iterator it=lm.begin(),itend=lm.end();
      if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
	*ptr=sum_degree(lm);
	++ptr;
      }
      if (order.o==_REVLEX_ORDER){
	for (--itend,--it;it!=itend;++ptr,--itend)
	  *ptr=*itend;
      }
      else {
	for (;it!=itend;++ptr,++it)
	  *ptr=*it;
      }
#ifdef GBASIS_SWAP
      swap_indices11(tab);
#endif
    }
  };


#ifdef NSPIRE
  template<class T>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os,const tdeg_t11 & x){
    os << "[";
    for (unsigned i=0; i<=GROEBNER_VARS;++i){
      os << x.tab[i] << ",";
    }
    return os << "]";
  }
#else
  ostream & operator << (ostream & os,const tdeg_t11 & x){
    os << "[";
    for (unsigned i=0; i<=GROEBNER_VARS;++i){
      os << x.tab[i] << ",";
    }
    return os << "]";
  }
#endif
  void tdeg_t11::dbgprint() const { COUT << * this << '\n'; }
  tdeg_t11 operator + (const tdeg_t11 & x,const tdeg_t11 & y);
  tdeg_t11 & operator += (tdeg_t11 & x,const tdeg_t11 & y){
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    xtab[0]+=ytab[0];
    xtab[1]+=ytab[1];
    xtab[2]+=ytab[2];
    return x;
  }
  tdeg_t11 operator + (const tdeg_t11 & x,const tdeg_t11 & y){
    tdeg_t11 res(x);
    return res += y;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]+ytab[0];
    ztab[1]=xtab[1]+ytab[1];
    ztab[2]=xtab[2]+ytab[2];
    return res;
  }
  inline void add(const tdeg_t11 & x,const tdeg_t11 & y,tdeg_t11 & res,int dim){
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]+ytab[0];
    ztab[1]=xtab[1]+ytab[1];
    ztab[2]=xtab[2]+ytab[2];
  }
  tdeg_t11 operator - (const tdeg_t11 & x,const tdeg_t11 & y){
    tdeg_t11 res;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]-ytab[0];
    ztab[1]=xtab[1]-ytab[1];
    ztab[2]=xtab[2]-ytab[2];
    return res;
  }
  inline bool operator == (const tdeg_t11 & x,const tdeg_t11 & y){
    return ((longlong *) x.tab)[0] == ((longlong *) y.tab)[0] &&
      ((longlong *) x.tab)[1] == ((longlong *) y.tab)[1] &&
      ((longlong *) x.tab)[2] == ((longlong *) y.tab)[2]
    ;
  }
  inline bool operator != (const tdeg_t11 & x,const tdeg_t11 & y){
    return !(x==y);
  }

  static inline int tdeg_t11_revlex_greater (const tdeg_t11 & x,const tdeg_t11 & y){
#ifdef GBASIS_SWAP
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    if (xtab[0]!=ytab[0]) // tdeg test already donne by caller
      return xtab[0]<=ytab[0]?1:0;
    if (xtab[1]!=ytab[1])
      return xtab[1]<=ytab[1]?1:0;
    if (xtab[2]!=ytab[2])
      return xtab[2]<=ytab[2]?1:0;
#else
    longlong *xtab=(longlong *)&x,*ytab=(longlong *)&y;
    if (longlong a=*xtab-*ytab) // tdeg test already donne by caller
      return a<=0?1:0;
    if (longlong a=xtab[1]-ytab[1])
      return a<=0?1:0;
    if (longlong a=xtab[2]-ytab[2])
      return a<=0?1:0;
#endif
    return 2;
#else // GBASIS_SWAP
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>=y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]<=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
    }
    return 2;
#endif // GBASIS_SWAP
  }

  int tdeg_t11_lex_greater (const tdeg_t11 & x,const tdeg_t11 & y){
#ifdef GBASIS_SWAP
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    ulonglong X=*xtab, Y=*ytab;
    if (X!=Y){
      if ( (X & 0xffff) != (Y &0xffff))
	return (X&0xffff)>=(Y&0xffff)?1:0;
      return X>=Y?1:0;
    }
    if (xtab[1]!=ytab[1])
      return xtab[1]>=ytab[1]?1:0;
    if (xtab[2]!=ytab[2])
      return xtab[2]>=ytab[2]?1:0;
    return 2;
#else
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]>y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]>y.tab[2]?1:0;
      return x.tab[3]>y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]>y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]>y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]>y.tab[6]?1:0;
      return x.tab[7]>y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]>y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]>y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]>y.tab[10]?1:0;
      return x.tab[11]>=y.tab[11]?1:0;
    }
    return 2;
#endif
  }

  inline int tdeg_t_greater(const tdeg_t11 & x,const tdeg_t11 & y,order_t order){
    short X=x.tab[0];
    if (X!=y.tab[0]) return X>y.tab[0]?1:0; // since tdeg is tab[0] for plex
    if (order.o==_REVLEX_ORDER)
      return tdeg_t11_revlex_greater(x,y);
    return tdeg_t11_lex_greater(x,y);
  }
  inline bool tdeg_t_strictly_greater (const tdeg_t11 & x,const tdeg_t11 & y,order_t order){
    return !tdeg_t_greater(y,x,order); // total order
  }

  inline bool tdeg_t_all_greater(const tdeg_t11 & x,const tdeg_t11 & y,order_t order){
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    if ((xtab[0]-ytab[0]) & 0x8000800080008000ULL)
      return false;
    if ((xtab[1]-ytab[1]) & 0x8000800080008000ULL)
      return false;
    if ((xtab[2]-ytab[2]) & 0x8000800080008000ULL)
      return false;
    return true;
  }

  // 1 (all greater), 0 (unknown), -1 (all smaller)
  int tdeg_t_compare_all(const tdeg_t11 & x,const tdeg_t11 & y,order_t order){
    int res=0;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    longlong tmp=xtab[0]-ytab[0];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    tmp=xtab[1]-ytab[1];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    tmp=xtab[2]-ytab[2];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    return res;
  }

  void index_lcm(const tdeg_t11 & x,const tdeg_t11 & y,tdeg_t11 & z,order_t order){
    int t=0;
    const short * xtab=&x.tab[1],*ytab=&y.tab[1];
    short *ztab=&z.tab[1];
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 1
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 2
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 3
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 4
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 5
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 6
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 7
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 8
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
      z.tab[0]=t;
    }
    else {
      z.tab[0]=(x.tab[0]>y.tab[0])?x.tab[0]:y.tab[0];
    }
  }

  inline void index_lcm_overwrite(const tdeg_t11 & x,const tdeg_t11 & y,tdeg_t11 & z,order_t order){
    index_lcm(x,y,z,order);
  }

  void get_index(const tdeg_t11 & x_,index_t & idx,order_t order,int dim){
    idx.resize(dim);
#ifdef GBASIS_SWAP
    tdeg_t11 x(x_);
    swap_indices11(x.tab);
#else
    const tdeg_t11 & x= x_;
#endif
    const short * ptr=x.tab;
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER)
      ++ptr;
    if (order.o==_REVLEX_ORDER){
      for (int i=1;i<=dim;++ptr,++i)
	idx[dim-i]=*ptr;
    }
    else {
      for (int i=0;i<dim;++ptr,++i)
	idx[i]=*ptr;
    }
  }

  bool disjoint(const tdeg_t11 & a,const tdeg_t11 & b,order_t order,short dim){
    const short * it=a.tab, * jt=b.tab;
#ifdef GBASIS_SWAP
    const short * itend=it+GROEBNER_VARS+1;
#endif
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
      ++it; ++jt;
    }
#ifndef GBASIS_SWAP
    const short * itend=it+dim;
#endif
    for (;it<itend;++jt,++it){
      if (*it && *jt)
	return false;
    }
    return true;
  }

#undef GROEBNER_VARS
#define GROEBNER_VARS 15

  struct tdeg_t15 {
    bool vars64() const { return false;}
    int hash_index(void * ptr_) const {
      // if (!ptr_)
	return -1;
    }
    bool add_to_hash(void *ptr_,int no) const {
      return false;
    }
    void dbgprint() const;
    // data
    union {
      short tab[GROEBNER_VARS+1];
      struct {
	short tdeg; // actually it's twice the total degree+1
	short tdeg2;
	order_t order_;
	longlong * ui;
      };
    };
    int front(){ return tab[1];}
    // methods
    inline unsigned selection_degree(order_t order) const {
#ifdef GBASIS_SELECT_TOTAL_DEGREE
      return total_degree(order);
#endif
      return tab[0];
    }
    inline unsigned total_degree(order_t order) const {
      if (order.o==_REVLEX_ORDER)
	return tab[0];      // works only for revlex and tdeg
      if (order.o==_3VAR_ORDER)
	return tab[0]+tab[4];
      if (order.o==_7VAR_ORDER)
	return tab[0]+tab[8];
      if (order.o==_11VAR_ORDER)
	return tab[0]+tab[12];
      return tab[0];
    }
    // void set_total_degree(unsigned d) { tab[0]=d;}
    tdeg_t15() {
      longlong * ptr = (longlong *) tab;
      ptr[2]=ptr[1]=ptr[0]=0;
#if GROEBNER_VARS>11
      ptr[3]=0;
#endif
    }
    tdeg_t15(int i){
      longlong * ptr = (longlong *) tab;
      ptr[2]=ptr[1]=ptr[0]=0;
#if GROEBNER_VARS>11
      ptr[3]=0;
#endif
    }
    void get_tab(short * ptr,order_t order) const {
      for (unsigned i=0;i<=GROEBNER_VARS;++i)
	ptr[i]=tab[i];
#ifdef GBASIS_SWAP
      swap_indices15(ptr,order.o);
#endif
    }
    tdeg_t15(const index_m & lm,order_t order){
      longlong * ptr_ = (longlong *) tab;
      ptr_[2]=ptr_[1]=ptr_[0]=0;
      short * ptr=tab;
#if GROEBNER_VARS>11
      ptr_[3]=0;
#endif
      // tab[GROEBNER_VARS]=order;
#if GROEBNER_VARS==15
      if (order.o==_3VAR_ORDER){
	ptr[0]=lm[0]+lm[1]+lm[2];
	ptr[1]=lm[2];
	ptr[2]=lm[1];
	ptr[3]=lm[0];
	ptr +=5;
	short t=0;
	vector<deg_t>::const_iterator it=lm.begin()+3,itend=lm.end();
	for (--itend,--it;it!=itend;++ptr,--itend){
	  t += *itend;
	  *ptr=*itend;
	}
	tab[4]=t;
#ifdef GBASIS_SWAP
	swap_indices15(tab,order.o);
#endif
	return;
      }
      if (order.o==_7VAR_ORDER){
	ptr[0]=lm[0]+lm[1]+lm[2]+lm[3]+lm[4]+lm[5]+lm[6];
	ptr[1]=lm[6];
	ptr[2]=lm[5];
	ptr[3]=lm[4];
	ptr[4]=lm[3];
	ptr[5]=lm[2];
	ptr[6]=lm[1];
	ptr[7]=lm[0];
	ptr +=9;
	short t=0;
	vector<deg_t>::const_iterator it=lm.begin()+7,itend=lm.end();
	for (--itend,--it;it!=itend;++ptr,--itend){
	  t += *itend;
	  *ptr=*itend;
	}
	tab[8]=t;
#ifdef GBASIS_SWAP
	swap_indices15(tab,order.o);
#endif
	return;
      }
      if (order.o==_11VAR_ORDER){
	ptr[0]=lm[0]+lm[1]+lm[2]+lm[3]+lm[4]+lm[5]+lm[6]+lm[7]+lm[8]+lm[9]+lm[10];
	ptr[1]=lm[10];
	ptr[2]=lm[9];
	ptr[3]=lm[8];
	ptr[4]=lm[7];
	ptr[5]=lm[6];
	ptr[6]=lm[5];
	ptr[7]=lm[4];
	ptr[8]=lm[3];
	ptr[9]=lm[2];
	ptr[10]=lm[1];
	ptr[11]=lm[0];
	ptr += 13;
	short t=0;
	vector<deg_t>::const_iterator it=lm.begin()+11,itend=lm.end();
	for (--itend,--it;it!=itend;++ptr,--itend){
	  t += *itend;
	  *ptr=*itend;
	}
	tab[12]=t;
#ifdef GBASIS_SWAP
	swap_indices15(tab,order.o);
#endif
	return;
      }
#endif
      vector<deg_t>::const_iterator it=lm.begin(),itend=lm.end();
      if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
	*ptr=sum_degree(lm);
	++ptr;
      }
      if (order.o==_REVLEX_ORDER){
	for (--itend,--it;it!=itend;++ptr,--itend)
	  *ptr=*itend;
      }
      else {
	for (;it!=itend;++ptr,++it)
	  *ptr=*it;
      }
#ifdef GBASIS_SWAP
      swap_indices15(tab,order.o);
#endif
    }
  };


#ifdef NSPIRE
  template<class T>
  nio::ios_base<T> & operator << (nio::ios_base<T> & os,const tdeg_t15 & x){
    os << "[";
    for (unsigned i=0; i<=GROEBNER_VARS;++i){
      os << x.tab[i] << ",";
    }
    return os << "]";
  }
#else
  ostream & operator << (ostream & os,const tdeg_t15 & x){
    os << "[";
    for (unsigned i=0; i<=GROEBNER_VARS;++i){
      os << x.tab[i] << ",";
    }
    return os << "]";
  }
#endif
  void tdeg_t15::dbgprint() const { COUT << * this << '\n'; }
  tdeg_t15 operator + (const tdeg_t15 & x,const tdeg_t15 & y);
  tdeg_t15 & operator += (tdeg_t15 & x,const tdeg_t15 & y){
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    xtab[0]+=ytab[0];
    xtab[1]+=ytab[1];
    xtab[2]+=ytab[2];
#if GROEBNER_VARS>11
    xtab[3]+=ytab[3];
#endif
    return x;
  }
  tdeg_t15 operator + (const tdeg_t15 & x,const tdeg_t15 & y){
    tdeg_t15 res(x);
    return res += y;
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]+ytab[0];
    ztab[1]=xtab[1]+ytab[1];
    ztab[2]=xtab[2]+ytab[2];
#if GROEBNER_VARS>11
    ztab[3]=xtab[3]+ytab[3];
#endif
#else
    for (unsigned i=0;i<=GROEBNER_VARS;++i)
      res.tab[i]=x.tab[i]+y.tab[i];
#endif
    return res;
  }
  inline void add(const tdeg_t15 & x,const tdeg_t15 & y,tdeg_t15 & res,int dim){
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]+ytab[0];
    ztab[1]=xtab[1]+ytab[1];
    ztab[2]=xtab[2]+ytab[2];
#if GROEBNER_VARS>11
    ztab[3]=xtab[3]+ytab[3];
#endif
#else
    for (unsigned i=0;i<=dim;++i)
      res.tab[i]=x.tab[i]+y.tab[i];
#endif
  }
  tdeg_t15 operator - (const tdeg_t15 & x,const tdeg_t15 & y){
    tdeg_t15 res;
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y,*ztab=(ulonglong *)&res;
    ztab[0]=xtab[0]-ytab[0];
    ztab[1]=xtab[1]-ytab[1];
    ztab[2]=xtab[2]-ytab[2];
#if GROEBNER_VARS>11
    ztab[3]=xtab[3]-ytab[3];
#endif
#else
    for (unsigned i=0;i<=GROEBNER_VARS;++i)
      res.tab[i]=x.tab[i]-y.tab[i];
#endif
    return res;
  }
  inline bool operator == (const tdeg_t15 & x,const tdeg_t15 & y){
    return ((longlong *) x.tab)[0] == ((longlong *) y.tab)[0] &&
      ((longlong *) x.tab)[1] == ((longlong *) y.tab)[1] &&
      ((longlong *) x.tab)[2] == ((longlong *) y.tab)[2]
#if GROEBNER_VARS>11
      &&  ((longlong *) x.tab)[3] == ((longlong *) y.tab)[3]
#endif
    ;
  }
  inline bool operator != (const tdeg_t15 & x,const tdeg_t15 & y){
    return !(x==y);
  }

  static inline int tdeg_t15_revlex_greater (const tdeg_t15 & x,const tdeg_t15 & y){
#ifdef GBASIS_SWAP
#if 1
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    if (xtab[0]!=ytab[0]) // tdeg test already donne by caller
      return xtab[0]<=ytab[0]?1:0;
    if (xtab[1]!=ytab[1])
      return xtab[1]<=ytab[1]?1:0;
    if (xtab[2]!=ytab[2])
      return xtab[2]<=ytab[2]?1:0;
#else
    longlong *xtab=(longlong *)&x,*ytab=(longlong *)&y;
    if (longlong a=*xtab-*ytab) // tdeg test already donne by caller
      return a<=0?1:0;
    if (longlong a=xtab[1]-ytab[1])
      return a<=0?1:0;
    if (longlong a=xtab[2]-ytab[2])
      return a<=0?1:0;
#endif
#if GROEBNER_VARS>11
    return xtab[3]<=ytab[3]?1:0;
#endif
    return 2;
#else // GBASIS_SWAP
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>=y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]<=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
    }
#if GROEBNER_VARS>11
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]<=y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
    }
#endif
    return 2;
#endif // GBASIS_SWAP
  }


#if GROEBNER_VARS==15

  int tdeg_t15_3var_greater (const tdeg_t15 & x,const tdeg_t15 & y){
    if (x.tab[0]!=y.tab[0])
      return x.tab[0]>=y.tab[0]?1:0;
    if (x.tab[4]!=y.tab[4])
      return x.tab[4]>=y.tab[4]?1:0;
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[0] <= ((longlong *) y.tab)[0];
#else
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
#endif
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[1] <= ((longlong *) y.tab)[1];
#else
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
#endif
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[2] <= ((longlong *) y.tab)[2];
#else
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]<=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
#endif
    }
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[3] <= ((longlong *) y.tab)[3];
#else
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]<=y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
#endif
    }
    return 2;
  }

  int tdeg_t15_7var_greater (const tdeg_t15 & x,const tdeg_t15 & y){
    if (x.tab[0]!=y.tab[0])
      return x.tab[0]>=y.tab[0]?1:0;
    if (x.tab[8]!=y.tab[8])
      return x.tab[8]>=y.tab[8]?1:0;
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[0] <= ((longlong *) y.tab)[0];
#else
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
#endif
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[1] <= ((longlong *) y.tab)[1];
#else
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
#endif
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[2] <= ((longlong *) y.tab)[2];
#else
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
#endif
    }
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[3] <= ((longlong *) y.tab)[3];
#else
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]<=y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
#endif
    }
    return 2;
  }

  int tdeg_t15_11var_greater (const tdeg_t15 & x,const tdeg_t15 & y){
    if (x.tab[0]!=y.tab[0])
      return x.tab[0]>=y.tab[0]?1:0;
    if (x.tab[12]!=y.tab[12])
      return x.tab[12]>=y.tab[12]?1:0;
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[0] <= ((longlong *) y.tab)[0];
#else
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]<=y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]<=y.tab[2]?1:0;
      return x.tab[3]<=y.tab[3]?1:0;
#endif
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[1] <= ((longlong *) y.tab)[1];
#else
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]<=y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]<=y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]<=y.tab[6]?1:0;
      return x.tab[7]<=y.tab[7]?1:0;
#endif
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[2] <= ((longlong *) y.tab)[2];
#else
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]<=y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]<=y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]<=y.tab[10]?1:0;
      return x.tab[11]<=y.tab[11]?1:0;
#endif
    }
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
#ifdef GBASIS_SWAP
      return ((longlong *) x.tab)[3] <= ((longlong *) y.tab)[3];
#else
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]<=y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]<=y.tab[14]?1:0;
      return x.tab[15]<=y.tab[15]?1:0;
#endif
    }
    return 2;
  }
#endif // GROEBNER_VARS==15

  int tdeg_t15_lex_greater (const tdeg_t15 & x,const tdeg_t15 & y){
#if 0 // def GBASIS_SWAP
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    ulonglong X=*xtab, Y=*ytab;
    if (X!=Y){
      if ( (X & 0xffff) != (Y &0xffff))
	return (X&0xffff)>=(Y&0xffff)?1:0;
      return X>=Y?1:0;
    }
    if (xtab[1]!=ytab[1])
      return xtab[1]>=ytab[1]?1:0;
    if (xtab[2]!=ytab[2])
      return xtab[2]>=ytab[2]?1:0;
#if GROEBNER_VARS>11
    return xtab[3]>=ytab[3]?1:0;
#endif
    return 2;
#else
    if (((longlong *) x.tab)[0] != ((longlong *) y.tab)[0]){
      if (x.tab[0]!=y.tab[0])
	return x.tab[0]>y.tab[0]?1:0;
      if (x.tab[1]!=y.tab[1])
	return x.tab[1]>y.tab[1]?1:0;
      if (x.tab[2]!=y.tab[2])
	return x.tab[2]>y.tab[2]?1:0;
      return x.tab[3]>y.tab[3]?1:0;
    }
    if (((longlong *) x.tab)[1] != ((longlong *) y.tab)[1]){
      if (x.tab[4]!=y.tab[4])
	return x.tab[4]>y.tab[4]?1:0;
      if (x.tab[5]!=y.tab[5])
	return x.tab[5]>y.tab[5]?1:0;
      if (x.tab[6]!=y.tab[6])
	return x.tab[6]>y.tab[6]?1:0;
      return x.tab[7]>y.tab[7]?1:0;
    }
    if (((longlong *) x.tab)[2] != ((longlong *) y.tab)[2]){
      if (x.tab[8]!=y.tab[8])
	return x.tab[8]>y.tab[8]?1:0;
      if (x.tab[9]!=y.tab[9])
	return x.tab[9]>y.tab[9]?1:0;
      if (x.tab[10]!=y.tab[10])
	return x.tab[10]>y.tab[10]?1:0;
      return x.tab[11]>=y.tab[11]?1:0;
    }
#if GROEBNER_VARS>11
    if (((longlong *) x.tab)[3] != ((longlong *) y.tab)[3]){
      if (x.tab[12]!=y.tab[12])
	return x.tab[12]>y.tab[12]?1:0;
      if (x.tab[13]!=y.tab[13])
	return x.tab[13]>y.tab[13]?1:0;
      if (x.tab[14]!=y.tab[14])
	return x.tab[14]>y.tab[14]?1:0;
      return x.tab[15]>=y.tab[15]?1:0;
    }
#endif
    return 2;
#endif
  }

  inline int tdeg_t_greater(const tdeg_t15 & x,const tdeg_t15 & y,order_t order){
    short X=x.tab[0];
    if (X!=y.tab[0]) return X>y.tab[0]?1:0; // since tdeg is tab[0] for plex
    if (order.o==_REVLEX_ORDER)
      return tdeg_t15_revlex_greater(x,y);
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER)
      return tdeg_t15_3var_greater(x,y);
    if (order.o==_7VAR_ORDER)
      return tdeg_t15_7var_greater(x,y);
    if (order.o==_11VAR_ORDER)
      return tdeg_t15_11var_greater(x,y);
#endif
    return tdeg_t15_lex_greater(x,y);
  }
  inline bool tdeg_t_strictly_greater (const tdeg_t15 & x,const tdeg_t15 & y,order_t order){
    return !tdeg_t_greater(y,x,order); // total order
  }

  bool tdeg_t_all_greater(const tdeg_t15 & x,const tdeg_t15 & y,order_t order){
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    if ((xtab[0]-ytab[0]) & 0x8000800080008000ULL)
      return false;
    if ((xtab[1]-ytab[1]) & 0x8000800080008000ULL)
      return false;
    if ((xtab[2]-ytab[2]) & 0x8000800080008000ULL)
      return false;
#if GROEBNER_VARS>11
    if ((xtab[3]-ytab[3]) & 0x8000800080008000ULL)
      return false;
#endif
    return true;
  }

  // 1 (all greater), 0 (unknown), -1 (all smaller)
  int tdeg_t_compare_all(const tdeg_t15 & x,const tdeg_t15 & y,order_t order){
    int res=0;
    ulonglong *xtab=(ulonglong *)&x,*ytab=(ulonglong *)&y;
    longlong tmp=xtab[0]-ytab[0];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    tmp=xtab[1]-ytab[1];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
    tmp=xtab[2]-ytab[2];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
#if GROEBNER_VARS>11
    tmp=xtab[3]-ytab[3];
    if (tmp & mask){
      if (res==1 || ((-tmp) & mask)) return 0;
      res=-1;
    }
    else {
      if (res==-1) return 0; else res=1;
    }
#endif
    return res;
  }

  void index_lcm(const tdeg_t15 & x,const tdeg_t15 & y,tdeg_t15 & z,order_t order){
    int t=0;
    const short * xtab=&x.tab[1],*ytab=&y.tab[1];
    short *ztab=&z.tab[1];
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 1
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 2
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 3
    ++xtab; ++ytab; ++ztab;
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER){
      z.tab[0]=t;
      t=0;
      ++xtab;++ytab;++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 5
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 6
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 7
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 8
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 12
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
      z.tab[4]=t; // 4
      return;
    }
#endif
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 4
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 5
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 6
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 7
    ++xtab; ++ytab; ++ztab;
#if GROEBNER_VARS==15
    if (order.o==_7VAR_ORDER){
      z.tab[0]=t;
      t=0;
      ++xtab;++ytab;++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 12
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
      z.tab[8]=t; // 8
      return;
    }
#endif
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 8
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 9
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 10
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 11
#if GROEBNER_VARS>11
    ++xtab; ++ytab; ++ztab;
#if GROEBNER_VARS==15
    if (order.o==_11VAR_ORDER){
      z.tab[0]=t;
      t=0;
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
      ++xtab; ++ytab; ++ztab;
      t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
      z.tab[12]=t; // 12
      return;
    }
#endif
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 12
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 13
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 14
    ++xtab; ++ytab; ++ztab;
    t += (*ztab=(*xtab>*ytab)?*xtab:*ytab); // 15
#endif
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
      z.tab[0]=t;
    }
    else {
      z.tab[0]=(x.tab[0]>y.tab[0])?x.tab[0]:y.tab[0];
    }
  }

  inline void index_lcm_overwrite(const tdeg_t15 & x,const tdeg_t15 & y,tdeg_t15 & z,order_t order){
    index_lcm(x,y,z,order);
  }

  void get_index(const tdeg_t15 & x_,index_t & idx,order_t order,int dim){
    idx.resize(dim);
#ifdef GBASIS_SWAP
    tdeg_t15 x(x_);
    swap_indices15(x.tab,order.o);
#else
    const tdeg_t15 & x= x_;
#endif
    const short * ptr=x.tab;
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER){
      ++ptr;
      for (int i=1;i<=3;++ptr,++i)
	idx[3-i]=*ptr;
      ++ptr;
      for (int i=1;i<=dim-3;++ptr,++i)
	idx[dim-i]=*ptr;
      return;
    }
    if (order.o==_7VAR_ORDER){
      ++ptr;
      for (int i=1;i<=7;++ptr,++i)
	idx[7-i]=*ptr;
      ++ptr;
      for (int i=1;i<=dim-7;++ptr,++i)
	idx[dim-i]=*ptr;
      return;
    }
    if (order.o==_11VAR_ORDER){
      ++ptr;
      for (int i=1;i<=11;++ptr,++i)
	idx[11-i]=*ptr;
      ++ptr;
      for (int i=1;i<=dim-11;++ptr,++i)
	idx[dim-i]=*ptr;
      return;
    }
#endif
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER)
      ++ptr;
    if (order.o==_REVLEX_ORDER){
      for (int i=1;i<=dim;++ptr,++i)
	idx[dim-i]=*ptr;
    }
    else {
      for (int i=0;i<dim;++ptr,++i)
	idx[i]=*ptr;
    }
  }

  bool disjoint(const tdeg_t15 & a,const tdeg_t15 & b,order_t order,short dim){
#if GROEBNER_VARS==15
    if (order.o==_3VAR_ORDER){
      if ( (a.tab[1] && b.tab[1]) ||
	   (a.tab[2] && b.tab[2]) ||
	   (a.tab[3] && b.tab[3]) ||
	   (a.tab[5] && b.tab[5]) ||
	   (a.tab[6] && b.tab[6]) ||
	   (a.tab[7] && b.tab[7]) ||
	   (a.tab[8] && b.tab[8]) ||
	   (a.tab[9] && b.tab[9]) ||
	   (a.tab[10] && b.tab[10]) ||
	   (a.tab[11] && b.tab[11]) ||
	   (a.tab[12] && b.tab[12]) ||
	   (a.tab[13] && b.tab[13]) ||
	   (a.tab[14] && b.tab[14]) ||
	   (a.tab[15] && b.tab[15]) )
	return false;
      return true;
    }
    if (order.o==_7VAR_ORDER){
      if ( (a.tab[1] && b.tab[1]) ||
	   (a.tab[2] && b.tab[2]) ||
	   (a.tab[3] && b.tab[3]) ||
	   (a.tab[4] && b.tab[4]) ||
	   (a.tab[5] && b.tab[5]) ||
	   (a.tab[6] && b.tab[6]) ||
	   (a.tab[7] && b.tab[7]) ||
	   (a.tab[9] && b.tab[9]) ||
	   (a.tab[10] && b.tab[10]) ||
	   (a.tab[11] && b.tab[11]) ||
	   (a.tab[12] && b.tab[12]) ||
	   (a.tab[13] && b.tab[13]) ||
	   (a.tab[14] && b.tab[14]) ||
	   (a.tab[15] && b.tab[15]) )
	return false;
      return true;
    }
    if (order.o==_11VAR_ORDER){
      if ( (a.tab[1] && b.tab[1]) ||
	   (a.tab[2] && b.tab[2]) ||
	   (a.tab[3] && b.tab[3]) ||
	   (a.tab[4] && b.tab[4]) ||
	   (a.tab[5] && b.tab[5]) ||
	   (a.tab[6] && b.tab[6]) ||
	   (a.tab[7] && b.tab[7]) ||
	   (a.tab[8] && b.tab[8]) ||
	   (a.tab[9] && b.tab[9]) ||
	   (a.tab[10] && b.tab[10]) ||
	   (a.tab[11] && b.tab[11]) ||
	   (a.tab[13] && b.tab[13]) ||
	   (a.tab[14] && b.tab[14]) ||
	   (a.tab[15] && b.tab[15]))
	return false;
      return true;
    }
#endif
    const short * it=a.tab, * jt=b.tab;
#ifdef GBASIS_SWAP
    const short * itend=it+GROEBNER_VARS+1;
#endif
    if (order.o==_REVLEX_ORDER || order.o==_TDEG_ORDER){
      ++it; ++jt;
    }
#ifndef GBASIS_SWAP
    const short * itend=it+dim;
#endif
    for (;it<itend;++jt,++it){
      if (*it && *jt)
	return false;
    }
    return true;
  }

  template<class T>
  static void get_newres(const T & resmod,vectpoly & newres,const vectpoly & v,const vector<unsigned> & G){
    newres=vectpoly(G.size(),polynome(v.front().dim,v.front()));
    for (unsigned i=0;i<G.size();++i)
      resmod[G[i]].get_polynome(newres[i]);
  }

  template<class tdeg_t>
  static void get_newres_ckrur(const vectpolymod<tdeg_t> & resmod,vectpoly & newres,const vectpoly & v,const vector<unsigned> & G,modint env,int & rur){
    if (rur && !resmod.empty()){
      vectpolymod<tdeg_t> gbmod; gbmod.reserve(G.size());
      for (int i=0;i<int(G.size());++i)
	gbmod.push_back(resmod[G[i]]);
      order_t order=resmod.front().order; int dim=resmod.front().dim;
      polymod<tdeg_t> lmtmp(order,dim),lmmodradical(order,dim),s(order,dim);
      vectpolymod<tdeg_t> rurv;
      if (rur_quotient_ideal_dimension(gbmod,lmtmp)<0)
	rur=0;
      else {
	if (debug_infolevel)
	  CERR << CLOCK()*1e-6 << " begin modular rur computation" << '\n';
	bool ok;
	if (rur==2){
	  vecteur m,M,res;
	  s.coord.clear();
	  index_t l(dim);
	  l[dim-1]=1;
	  s.coord.push_back(T_unsigned<modint,tdeg_t>(1,tdeg_t(l,order)));
	  ok=rur_minpoly(gbmod,lmtmp,s,env,m,M);
	  rur_convert_univariate(m,dim-1,gbmod[0]);
	  gbmod.resize(1);
	}
	else {
	  ok=rur_compute(gbmod,lmtmp,lmmodradical,env,s,rurv);
	  if (ok)
	    gbmod.swap(rurv);
	}
	if (!ok)
	  rur=0;
      }
      if (debug_infolevel)
	CERR << CLOCK()*1e-6 << " end modular rur computation" << '\n';
      newres=vectpoly(gbmod.size(),polynome(v.front().dim,v.front()));
      for (unsigned i=0;i<int(gbmod.size());++i)
	gbmod[i].get_polynome(newres[i]);
      return;
    }
    newres=vectpoly(G.size(),polynome(v.front().dim,v.front()));
    for (unsigned i=0;i<G.size();++i)
      resmod[G[i]].get_polynome(newres[i]);
  }

  template<class T>
  static void get_newres(const T & resmod,vectpoly & newres,const vectpoly & v){
    newres=vectpoly(resmod.size(),polynome(v.front().dim,v.front()));
    for (unsigned i=0;i<resmod.size();++i)
      resmod[i].get_polynome(newres[i]);
  }

  bool gbasis8(const vectpoly & v,order_t & order,vectpoly & newres,environment * env,bool modularalgo,bool modularcheck,int & rur,GIAC_CONTEXT,gbasis_param_t gbasis_param){
    bool & eliminate_flag=gbasis_param.eliminate_flag;
    int parallel=1;
#ifdef HAVE_LIBPTHREAD
    if (threads_allowed && threads>1)
      parallel=threads;
#endif
    int save_debuginfo=debug_infolevel;
    if (v.empty()){ newres.clear(); return true;}
#ifdef GIAC_TDEG_T14
    // do not use tdeg_t14 for rur because monomials have often large degrees
    if (v.front().dim<=14 && order.o==_REVLEX_ORDER && !rur){
      try {
	vectpoly8<tdeg_t14> res;
	vectpolymod<tdeg_t14> resmod;
	vector<unsigned> G;
	vectpoly_2_vectpoly8<tdeg_t14>(v,order,res);
	// Temporary workaround until rur_compute support parametric rur
	if (rur && absint(order.o)!=-_RUR_REVLEX){
	  rur=0;
	  order.o=absint(order.o);
	}
	CLOCK_T c=CLOCK();
	if (modularalgo && (!env || env->modulo==0 || env->moduloon==false)){
	  if (mod_gbasis(res,modularcheck,
			 //order.o==_REVLEX_ORDER /* zdata*/,
			 1 || !rur /* zdata*/,
			 rur,contextptr,gbasis_param)){
	    *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 << " Memory " << memory_usage()*1e-6 << "M" << '\n';
	    get_newres(res,newres,v);
	    debug_infolevel=save_debuginfo; return true;
	  }
	}
	if (env && env->moduloon && env->modulo.type==_INT_){
	  if (!res.empty() && (res.front().order.o==_REVLEX_ORDER || res.front().order.o==_3VAR_ORDER || res.front().order.o==_7VAR_ORDER || res.front().order.o==_11VAR_ORDER)){
	    vector<zinfo_t<tdeg_t14> > f4buchberger_info;
	    f4buchberger_info.reserve(GBASISF4_MAXITER);
	    if (zgbasis(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,f4buchberger_info,false/* recomputeR*/,false /* don't compute res8*/,eliminate_flag,false /* 1 mod only */,parallel,true)){
	      *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" <<'\n';
	      get_newres_ckrur(resmod,newres,v,G,env->modulo.val,rur);
	      debug_infolevel=save_debuginfo; return true;
	    }
	  }
	  else {
	    if (in_gbasisf4buchbergermod<tdeg_t14>(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,0,false)){
	      *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" <<'\n';
	      get_newres_ckrur(resmod,newres,v,G,env->modulo.val,rur);
	      debug_infolevel=save_debuginfo; return true;
	    }
	  }
	} // end if gbasis computation modulo integer
	else {
#ifdef GIAC_REDUCEMODULO
	  vectpoly w(v);
	  reduce(w,env);
	  sort_vectpoly(w.begin(),w.end());
	  vectpoly_2_vectpoly8(w,order,res);
#endif
	  if (in_gbasis(res,G,env,true)){
	    *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
	    get_newres(res,newres,v,G);
	    debug_infolevel=save_debuginfo; return true;
	  }
	}
      }  catch (std::runtime_error & e){
	last_evaled_argptr(contextptr)=NULL;
	CERR << "Degree too large for compressed monomials. Using uncompressed monomials instead." << '\n';
      }
    }
#endif
    if (v.front().dim<=11 && order.o==_REVLEX_ORDER){
      vectpoly8<tdeg_t11> res;
      vectpolymod<tdeg_t11> resmod;
      vector<unsigned> G;
      vectpoly_2_vectpoly8<tdeg_t11>(v,order,res);
      // Temporary workaround until rur_compute support parametric rur
      if (rur && absint(order.o)!=-_RUR_REVLEX){
	rur=0;
	order.o=absint(order.o);
      }
      CLOCK_T c=CLOCK();
      if (modularalgo && (!env || env->modulo==0 || env->moduloon==false)){
	if (mod_gbasis(res,modularcheck,
		       //order.o==_REVLEX_ORDER /* zdata*/,
		       1 || !rur /* zdata*/,
		       rur,contextptr,gbasis_param)){
	  *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
	  get_newres(res,newres,v);
	  debug_infolevel=save_debuginfo; return true;
	}
      }
      if (env && env->moduloon && env->modulo.type==_INT_){
	if (!res.empty() && (res.front().order.o==_REVLEX_ORDER || res.front().order.o==_3VAR_ORDER || res.front().order.o==_7VAR_ORDER || res.front().order.o==_11VAR_ORDER)){
	  vector<zinfo_t<tdeg_t11> > f4buchberger_info;
	  f4buchberger_info.reserve(GBASISF4_MAXITER);
	  if (zgbasis(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,f4buchberger_info,false/* recomputeR*/,false /* don't compute res8*/,eliminate_flag,false /* 1 mod only */,parallel,true)){
	    *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
	    get_newres_ckrur(resmod,newres,v,G,env->modulo.val,rur);
	    debug_infolevel=save_debuginfo; return true;
	  }
	}
	else {
	  if (in_gbasisf4buchbergermod<tdeg_t11>(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,0,false)){
	    *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
	    get_newres_ckrur(resmod,newres,v,G,env->modulo.val,rur);
	    debug_infolevel=save_debuginfo; return true;
	  }
	}
      } // end if gbasis modulo integer
      else {
#ifdef GIAC_REDUCEMODULO
	vectpoly w(v);
	reduce(w,env);
	sort_vectpoly(w.begin(),w.end());
	vectpoly_2_vectpoly8(w,order,res);
#endif
	if (in_gbasis(res,G,env,true)){
	  *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
	  get_newres(res,newres,v,G);
	  debug_infolevel=save_debuginfo; return true;
	}
      }
    }
    if (v.front().dim<=15
	//&& order.o==_REVLEX_ORDER
	&&order.o<_16VAR_ORDER
	){
      vectpoly8<tdeg_t15> res;
      vectpolymod<tdeg_t15> resmod;
      vector<unsigned> G;
      vectpoly_2_vectpoly8<tdeg_t15>(v,order,res);
      // Temporary workaround until rur_compute support parametric rur
      if (rur && absint(order.o)!=-_RUR_REVLEX){
	rur=0;
	order.o=absint(order.o);
      }
      CLOCK_T c=CLOCK();
      if (modularalgo && (!env || env->modulo==0 || env->moduloon==false)){
	if (mod_gbasis(res,modularcheck,
		       //order.o==_REVLEX_ORDER /* zdata*/,
		       1 || !rur /* zdata*/,
		       rur,contextptr,gbasis_param)){
	  *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
	  newres=vectpoly(res.size(),polynome(v.front().dim,v.front()));
	  for (unsigned i=0;i<res.size();++i)
	    res[i].get_polynome(newres[i]);
	  debug_infolevel=save_debuginfo; return true;
	}
      }
      if (env && env->moduloon && env->modulo.type==_INT_){
#ifdef GBASISF4_BUCHBERGER
	if (!res.empty() && (res.front().order.o==_REVLEX_ORDER || res.front().order.o==_3VAR_ORDER || res.front().order.o==_7VAR_ORDER || res.front().order.o==_11VAR_ORDER)){
	  vector<zinfo_t<tdeg_t15> > f4buchberger_info;
	  f4buchberger_info.reserve(GBASISF4_MAXITER);
	  if (!zgbasis(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,f4buchberger_info,false/* recomputeR*/,false /* don't compute res8*/,eliminate_flag,false/* 1 mod only*/,parallel,true))
	    return false;
	  *logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
#if 1
	  get_newres_ckrur(resmod,newres,v,G,env->modulo.val,rur);
#else
	  newres=vectpoly(G.size(),polynome(v.front().dim,v.front()));
	  for (unsigned i=0;i<G.size();++i)
	    resmod[G[i]].get_polynome(newres[i]);
#endif
	  debug_infolevel=save_debuginfo; return true;
	}
	else
	  in_gbasisf4buchbergermod<tdeg_t15>(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,0,false);
#else
	in_gbasismod(res,resmod,G,env->modulo.val,true,0);
#endif
	if (debug_infolevel)
	  CERR << "G=" << G << '\n';
      }
      else { // env->modoloon etc.
#ifdef GIAC_REDUCEMODULO
	vectpoly w(v);
	reduce(w,env);
	sort_vectpoly(w.begin(),w.end());
	vectpoly_2_vectpoly8(w,order,res);
#endif
	in_gbasis(res,G,env,true);
      }
      newres=vectpoly(G.size(),polynome(v.front().dim,v.front()));
      for (unsigned i=0;i<G.size();++i)
	res[G[i]].get_polynome(newres[i]);
      debug_infolevel=save_debuginfo; return true;
    }
    vectpoly8<tdeg_t64> res;
    vectpolymod<tdeg_t64> resmod;
    vector<unsigned> G;
    vectpoly_2_vectpoly8<tdeg_t64>(v,order,res);
    // Temporary workaround until rur_compute support parametric rur
    if (rur && absint(order.o)!=-_RUR_REVLEX){
      rur=0;
      order.o=absint(order.o);
    }
    CLOCK_T c=CLOCK();
    if (modularalgo && (!env || env->modulo==0 || env->moduloon==false)){
      if (mod_gbasis(res,modularcheck,
		     //order.o==_REVLEX_ORDER /* zdata*/,
		     !rur /* zdata*/,
		     rur,contextptr,gbasis_param)){
	*logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
	newres=vectpoly(res.size(),polynome(v.front().dim,v.front()));
	for (unsigned i=0;i<res.size();++i)
	  res[i].get_polynome(newres[i]);
	debug_infolevel=save_debuginfo; return true;
      }
    }
    if (env && env->moduloon && env->modulo.type==_INT_){
#ifdef GBASISF4_BUCHBERGER
      if (!res.empty() && (res.front().order.o==_REVLEX_ORDER || res.front().order.o==_3VAR_ORDER || res.front().order.o==_7VAR_ORDER || res.front().order.o==_11VAR_ORDER)){
	vector<zinfo_t<tdeg_t64> > f4buchberger_info;
	f4buchberger_info.reserve(GBASISF4_MAXITER);
	zgbasis(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,f4buchberger_info,false/* recomputeR*/,false /* don't compute res8*/,eliminate_flag,false/* 1 mod only*/,parallel,true);
	*logptr(contextptr) << "// Groebner basis computation time " << (CLOCK()-c)*1e-6 <<  " Memory " << memory_usage()*1e-6 << "M" << '\n';
#if 1
	get_newres_ckrur(resmod,newres,v,G,env->modulo.val,rur);
#else
	newres=vectpoly(G.size(),polynome(v.front().dim,v.front()));
	for (unsigned i=0;i<G.size();++i)
	  resmod[G[i]].get_polynome(newres[i]);
#endif
	debug_infolevel=save_debuginfo; return true;
      }
      else
	in_gbasisf4buchbergermod<tdeg_t64>(res,resmod,G,env->modulo.val,true/*totaldeg*/,0,0,false);
#else
      in_gbasismod(res,resmod,G,env->modulo.val,true,0);
#endif
      if (debug_infolevel)
	CERR << "G=" << G << '\n';
    }
    else {
#ifdef GIAC_REDUCEMODULO
      vectpoly w(v);
      reduce(w,env);
      sort_vectpoly(w.begin(),w.end());
      vectpoly_2_vectpoly8(w,order,res);
#endif
      in_gbasis(res,G,env,true);
    }
    newres=vectpoly(G.size(),polynome(v.front().dim,v.front()));
    for (unsigned i=0;i<G.size();++i)
      res[G[i]].get_polynome(newres[i]);
    debug_infolevel=save_debuginfo; return true;
  }

  bool greduce8(const vectpoly & v,const vectpoly & gb_,order_t & order,vectpoly & newres,environment * env,GIAC_CONTEXT){
    if (v.empty()){ newres.clear(); return true;}
    vectpoly8<tdeg_t64> red,gb,quo;
    vectpoly_2_vectpoly8(v,order,red);
    vectpoly_2_vectpoly8(gb_,order,gb);
    poly8<tdeg_t64> rem,TMP1,TMP2;
    vector<unsigned> G(gb_.size());
    for (int i=0;i<int(gb_.size());++i)
      G[i]=i;
    int dim;
    for (int i=0;i<int(v.size());++i){
      quo.clear();
      rem.coord.clear();
      dim=red[i].dim;
      if (debug_infolevel>1)
	COUT << CLOCK()*1e-6 << " begin reduce poly no " << i << " #monomials " << red[i].coord.size() << '\n';
      gen lambda;
      reduce(red[i],gb,G,-1,quo,rem,TMP1,TMP2,lambda,env);
      if (debug_infolevel>1)
	COUT << CLOCK()*1e-6 << " end reduce poly no " << i << " #monomials " << rem.coord.size() << '\n';
      for (int j=0;j<int(rem.coord.size());++j){
	rem.coord[j].g=rem.coord[j].g/lambda;
      }
      red[i]=rem;
    }
    newres=vectpoly(v.size(),polynome(v.front().dim,v.front()));
    for (unsigned i=0;i<int(v.size());++i)
      red[i].get_polynome(newres[i]);
    return true;
  }

#ifdef GIAC_RHASH
  inline size_t tdeg_t64_hash_function_object::operator () (const tdeg_t64 & v) const
  { return v.hash; }
#endif

#endif // CAS38_DISABLED

#ifndef NO_NAMESPACE_GIAC
} // namespace giac
#endif // ndef NO_NAMESPACE_GIAC