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

                        Multiplication

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


void mul(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b);
// x = a * b

void sqr(ZZ_pX& x, const ZZ_pX& a);
inline ZZ_pX sqr(const ZZ_pX& a)
   { ZZ_pX x; sqr(x, a); NTL_OPT_RETURN(ZZ_pX, x); }
// x = a^2

void mul(ZZ_pX & x, const ZZ_pX& a, const ZZ_p& b); 
void mul(ZZ_pX& x, const ZZ_pX& a, long b);


inline void mul(ZZ_pX& x, const ZZ_p& a, const ZZ_pX& b)
   { mul(x, b, a); }

inline void mul(ZZ_pX& x, long a, const ZZ_pX& b)
   { mul(x, b, a); }

inline ZZ_pX operator*(const ZZ_pX& a, const ZZ_pX& b)
   { ZZ_pX x; mul(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX operator*(const ZZ_pX& a, const ZZ_p& b)
   { ZZ_pX x; mul(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX operator*(const ZZ_pX& a, long b)
   { ZZ_pX x; mul(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX operator*(const ZZ_p& a, const ZZ_pX& b)
   { ZZ_pX x; mul(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX operator*(long a, const ZZ_pX& b)
   { ZZ_pX x; mul(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX& operator*=(ZZ_pX& x, const ZZ_pX& b)
   { mul(x, x, b); return x; }

inline ZZ_pX& operator*=(ZZ_pX& x, const ZZ_p& b)
   { mul(x, x, b); return x; }

inline ZZ_pX& operator*=(ZZ_pX& x, long b)
   { mul(x, x, b); return x; }


void PlainMul(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b);
// always uses the "classical" algorithm

void PlainSqr(ZZ_pX& x, const ZZ_pX& a);
// always uses the "classical" algorithm


void FFTMul(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b);
// always uses the FFT

void FFTSqr(ZZ_pX& x, const ZZ_pX& a);
// always uses the FFT

void MulTrunc(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b, long n);
// x = a * b % X^n

inline ZZ_pX MulTrunc(const ZZ_pX& a, const ZZ_pX& b, long n)
   { ZZ_pX x; MulTrunc(x, a, b, n); NTL_OPT_RETURN(ZZ_pX, x); }

void PlainMulTrunc(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b, long n);
void FFTMulTrunc(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b, long n);

void SqrTrunc(ZZ_pX& x, const ZZ_pX& a, long n);
// x = a^2 % X^n

inline ZZ_pX SqrTrunc(const ZZ_pX& a, long n)
   { ZZ_pX x; SqrTrunc(x, a, n); NTL_OPT_RETURN(ZZ_pX, x); }

void PlainSqrTrunc(ZZ_pX& x, const ZZ_pX& a, long n);
void FFTSqrTrunc(ZZ_pX& x, const ZZ_pX& a, long n);


void power(ZZ_pX& x, const ZZ_pX& a, long e);
inline ZZ_pX power(const ZZ_pX& a, long e)
   { ZZ_pX x; power(x, a, e); NTL_OPT_RETURN(ZZ_pX, x); }


// The following data structures and routines allow one
// to hand-craft various algorithms, using the FFT convolution
// algorithms directly.
// Look in the file ZZ_pX.c for examples.




// FFT representation of polynomials

class FFTRep {
public:
   long k;                // a 2^k point representation
   long MaxK;             // maximum space allocated
   long **tbl;
   long NumPrimes; 

   void SetSize(long NewK);

   FFTRep(const FFTRep& R);
   FFTRep& operator=(const FFTRep& R);

   FFTRep() { k = MaxK = -1; tbl = 0; NumPrimes = 0; }
   FFTRep(INIT_SIZE_TYPE, long InitK) 
   { k = MaxK = -1; tbl = 0; NumPrimes = 0; SetSize(InitK); }
   ~FFTRep();
};


void ToFFTRep(FFTRep& y, const ZZ_pX& x, long k, long lo, long hi);
// computes an n = 2^k point convolution of x[lo..hi].

inline void ToFFTRep(FFTRep& y, const ZZ_pX& x, long k)

   { ToFFTRep(y, x, k, 0, deg(x)); }

void RevToFFTRep(FFTRep& y, const vec_ZZ_p& x,
                 long k, long lo, long hi, long offset);
// computes an n = 2^k point convolution of X^offset*x[lo..hi] mod X^n-1
// using "inverted" evaluation points.


void FromFFTRep(ZZ_pX& x, FFTRep& y, long lo, long hi);
// converts from FFT-representation to coefficient representation
// only the coefficients lo..hi are computed
// NOTE: this version destroys the data in y

// non-destructive versions of the above

void NDFromFFTRep(ZZ_pX& x, const FFTRep& y, long lo, long hi, FFTRep& temp);
void NDFromFFTRep(ZZ_pX& x, const FFTRep& y, long lo, long hi);

void RevFromFFTRep(vec_ZZ_p& x, FFTRep& y, long lo, long hi);

   // converts from FFT-representation to coefficient representation
   // using "inverted" evaluation points.
   // only the coefficients lo..hi are computed




void FromFFTRep(ZZ_p* x, FFTRep& y, long lo, long hi);
// convert out coefficients lo..hi of y, store result in x.
// no normalization is done.


// direct manipulation of FFT reps

void mul(FFTRep& z, const FFTRep& x, const FFTRep& y);
void sub(FFTRep& z, const FFTRep& x, const FFTRep& y);
void add(FFTRep& z, const FFTRep& x, const FFTRep& y);

void reduce(FFTRep& x, const FFTRep& a, long k);
// reduces a 2^l point FFT-rep to a 2^k point FFT-rep

void AddExpand(FFTRep& x, const FFTRep& a);
//  x = x + (an "expanded" version of a)





// This data structure holds unconvoluted modular representations
// of polynomials

class ZZ_pXModRep {
private:
   ZZ_pXModRep(const ZZ_pXModRep&); // disabled
   void operator=(const ZZ_pXModRep&); // disabled

public:
   long n;
   long MaxN;
   long **tbl;
   long NumPrimes; 

   void SetSize(long NewN);

   ZZ_pXModRep() { n = MaxN = 0; tbl = 0; NumPrimes = 0; }
   ZZ_pXModRep(INIT_SIZE_TYPE, long k) 
   { n = MaxN = 0; tbl = 0; NumPrimes = 0; SetSize(k); }
   ~ZZ_pXModRep();
};


void ToZZ_pXModRep(ZZ_pXModRep& x, const ZZ_pX& a, long lo, long hi);

void ToFFTRep(FFTRep& x, const ZZ_pXModRep& a, long k, long lo, long hi);
// converts coefficients lo..hi to a 2^k-point FFTRep.
// must have hi-lo+1 < 2^k





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

                      Division

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

void DivRem(ZZ_pX& q, ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b);
// q = a/b, r = a%b

void div(ZZ_pX& q, const ZZ_pX& a, const ZZ_pX& b);
// q = a/b

void div(ZZ_pX& q, const ZZ_pX& a, const ZZ_p& b);
void div(ZZ_pX& q, const ZZ_pX& a, long b);


void rem(ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b);
// r = a%b

long divide(ZZ_pX& q, const ZZ_pX& a, const ZZ_pX& b);
// if b | a, sets q = a/b and returns 1; otherwise returns 0

long divide(const ZZ_pX& a, const ZZ_pX& b);
// if b | a, sets q = a/b and returns 1; otherwise returns 0

void InvTrunc(ZZ_pX& x, const ZZ_pX& a, long m);
// computes x = a^{-1} % X^m 
// constant term must be non-zero

inline ZZ_pX InvTrunc(const ZZ_pX& a, long m)
   { ZZ_pX x; InvTrunc(x, a, m); NTL_OPT_RETURN(ZZ_pX, x); }



// These always use "classical" arithmetic
void PlainDivRem(ZZ_pX& q, ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b);
void PlainDiv(ZZ_pX& q, const ZZ_pX& a, const ZZ_pX& b);
void PlainRem(ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b);

void PlainRem(ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b, ZZVec& tmp);
void PlainDivRem(ZZ_pX& q, ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b,
                 ZZVec& tmp);


// These always use FFT arithmetic
void FFTDivRem(ZZ_pX& q, ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b);
void FFTDiv(ZZ_pX& q, const ZZ_pX& a, const ZZ_pX& b);
void FFTRem(ZZ_pX& r, const ZZ_pX& a, const ZZ_pX& b);

void PlainInvTrunc(ZZ_pX& x, const ZZ_pX& a, long m);
// always uses "classical" algorithm
// ALIAS RESTRICTION: input may not alias output

void NewtonInvTrunc(ZZ_pX& x, const ZZ_pX& a, long m);
// uses a Newton Iteration with the FFT.
// ALIAS RESTRICTION: input may not alias output


inline ZZ_pX operator/(const ZZ_pX& a, const ZZ_pX& b)
   { ZZ_pX x; div(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX operator/(const ZZ_pX& a, const ZZ_p& b)
   { ZZ_pX x; div(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX operator/(const ZZ_pX& a, long b)
   { ZZ_pX x; div(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX& operator/=(ZZ_pX& x, const ZZ_p& b)
   { div(x, x, b); return x; }

inline ZZ_pX& operator/=(ZZ_pX& x, long b)
   { div(x, x, b); return x; }

inline ZZ_pX& operator/=(ZZ_pX& x, const ZZ_pX& b)
   { div(x, x, b); return x; }


inline ZZ_pX operator%(const ZZ_pX& a, const ZZ_pX& b)
   { ZZ_pX x; rem(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

inline ZZ_pX& operator%=(ZZ_pX& x, const ZZ_pX& b)
   { rem(x, x, b); return x; }


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

                         GCD's

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


void GCD(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b);
// x = GCD(a, b),  x is always monic (or zero if a==b==0).

inline ZZ_pX GCD(const ZZ_pX& a, const ZZ_pX& b)
   { ZZ_pX x; GCD(x, a, b); NTL_OPT_RETURN(ZZ_pX, x); }

void XGCD(ZZ_pX& d, ZZ_pX& s, ZZ_pX& t, const ZZ_pX& a, const ZZ_pX& b);
// d = gcd(a,b), a s + b t = d 

void PlainXGCD(ZZ_pX& d, ZZ_pX& s, ZZ_pX& t, const ZZ_pX& a, const ZZ_pX& b);
// same as above, but uses classical algorithm


void PlainGCD(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b);
// always uses "cdlassical" arithmetic


class ZZ_pXMatrix {
private:

   ZZ_pXMatrix(const ZZ_pXMatrix&);  // disable
   ZZ_pX elts[2][2];

public:

   ZZ_pXMatrix() { }
   ~ZZ_pXMatrix() { }

   void operator=(const ZZ_pXMatrix&);
   ZZ_pX& operator() (long i, long j) { return elts[i][j]; }
   const ZZ_pX& operator() (long i, long j) const { return elts[i][j]; }
};


void HalfGCD(ZZ_pXMatrix& M_out, const ZZ_pX& U, const ZZ_pX& V, long d_red);
// deg(U) > deg(V),   1 <= d_red <= deg(U)+1.
//
// This computes a 2 x 2 polynomial matrix M_out such that
//    M_out * (U, V)^T = (U', V')^T,
// where U', V' are consecutive polynomials in the Euclidean remainder
// sequence of U, V, and V' is the polynomial of highest degree
// satisfying deg(V') <= deg(U) - d_red.

void XHalfGCD(ZZ_pXMatrix& M_out, ZZ_pX& U, ZZ_pX& V, long d_red);

// same as above, except that U is replaced by U', and V by V'


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

             Modular Arithmetic without pre-conditioning

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

// arithmetic mod f.
// all inputs and outputs are polynomials of degree less than deg(f).
// ASSUMPTION: f is assumed monic, and deg(f) > 0.
// NOTE: if you want to do many computations with a fixed f,
//       use the ZZ_pXModulus data structure and associated routines below.



void MulMod(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& b, const ZZ_pX& f);
// x = (a * b) % f

inline ZZ_pX MulMod(const ZZ_pX& a, const ZZ_pX& b, const ZZ_pX& f)
   { ZZ_pX x; MulMod(x, a, b, f); NTL_OPT_RETURN(ZZ_pX, x); }

void SqrMod(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& f);
// x = a^2 % f

inline ZZ_pX SqrMod(const ZZ_pX& a,  const ZZ_pX& f)
   { ZZ_pX x; SqrMod(x, a, f); NTL_OPT_RETURN(ZZ_pX, x); }

void MulByXMod(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& f);
// x = (a * X) mod f

inline ZZ_pX MulByXMod(const ZZ_pX& a,  const ZZ_pX& f)
   { ZZ_pX x; MulByXMod(x, a, f); NTL_OPT_RETURN(ZZ_pX, x); }



void InvMod(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& f);
// x = a^{-1} % f, error is a is not invertible

inline ZZ_pX InvMod(const ZZ_pX& a,  const ZZ_pX& f)
   { ZZ_pX x; InvMod(x, a, f); NTL_OPT_RETURN(ZZ_pX, x); }

long InvModStatus(ZZ_pX& x, const ZZ_pX& a, const ZZ_pX& f);
// if (a, f) = 1, returns 0 and sets x = a^{-1} % f
// otherwise, returns 1 and sets x = (a, f)




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

        Modular Arithmetic with Pre-conditioning

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


// If you need to do a lot of arithmetic modulo a fixed f,
// build ZZ_pXModulus F for f.  This pre-computes information about f
// that speeds up the computation a great deal.


class ZZ_pXModulus {
public:
   ZZ_pXModulus() : UseFFT(0), n(-1)  { }
   ~ZZ_pXModulus() { }
   

   // the following members may become private in future
   ZZ_pX f;   // the modulus
   long UseFFT;// flag indicating whether FFT should be used.
   long n;     // n = deg(f)
   long k;     // least k s/t 2^k >= n
   long l;     // least l s/t 2^l >= 2n-3
   FFTRep FRep; // 2^k point rep of f
                // H = rev((rev(f))^{-1} rem X^{n-1})
   FFTRep HRep; // 2^l point rep of H