How Can I Optimize My Number Theoretic Transform (NTT) and Modular Arithmetic for Faster Big Number Squaring?

Modular arithmetics and NTT (finite field DFT) optimizations

Original question:

I wanted to use NTT for fast squaring (see Fast bignum square computation), but the result is slow even for really big numbers .. more than 12000 bits.

My question is:

1. Is there a way to optimize my NTT transform?
   I did not mean to speed it by parallelism (threads); this is low-level layer only.
2. Is there a way to speed up my modular arithmetics?

Code:

//---------------------------------------------------------------------------
class fourier_NTT                                    // Number theoretic transform
    {

public:
    DWORD r,L,p,N;
    DWORD W,iW,rN;
    fourier_NTT(){ r=0; L=0; p=0; W=0; iW=0; rN=0; }

    // main interface
    void  NTT(DWORD *dst,DWORD *src,DWORD n=0);               // DWORD dst[n] = fast  NTT(DWORD src[n])
    void iNTT(DWORD *dst,DWORD *src,DWORD n=0);               // DWORD dst[n] = fast INTT(DWORD src[n])

    // Helper functions
    bool init(DWORD n);                                       // init r,L,p,W,iW,rN
    void  NTT_fast(DWORD *dst,DWORD *src,DWORD n,DWORD w);    // DWORD dst[n] = fast  NTT(DWORD src[n])

    // Only for testing
    void  NTT_slow(DWORD *dst,DWORD *src,DWORD n,DWORD w);    // DWORD dst[n] = slow  NTT(DWORD src[n])
    void iNTT_slow(DWORD *dst,DWORD *src,DWORD n,DWORD w);    // DWORD dst[n] = slow INTT(DWORD src[n])

    // DWORD arithmetics
    DWORD shl(DWORD a);
    DWORD shr(DWORD a);

    // Modular arithmetics
    DWORD mod(DWORD a);
    DWORD modadd(DWORD a,DWORD b);
    DWORD modsub(DWORD a,DWORD b);
    DWORD modmul(DWORD a,DWORD b);
    DWORD modpow(DWORD a,DWORD b);
    };

//---------------------------------------------------------------------------
void fourier_NTT:: NTT(DWORD *dst,DWORD *src,DWORD n)
    {
    if (n>0) init(n);
    NTT_fast(dst,src,N,W);
//    NTT_slow(dst,src,N,W);
    }

//---------------------------------------------------------------------------
void fourier_NTT::INTT(DWORD *dst,DWORD *src,DWORD n)
    {
    if (n>0) init(n);
    NTT_fast(dst,src,N,iW);
    for (DWORD i=0;i<n dst intt_slow bool fourier_ntt::init n p else ntt overflow can ocur r="2;" if>0x10000000)) { r=0; L=0; p=0; W=0; iW=0; rN=0; N=0; return false; } L=0x30000000/n; // 32:30 bit best for unsigned 32 bit
//    r=2; p=0x78000001; if ((n0x04000000)) { r=0; L=0; p=0; W=0; iW=0; rN=0; N=0; return false; } L=0x3c000000/n; // 31:27 bit best for signed 32 bit
//    r=2; p=0x00010001; if ((n0x00000020)) { r=0; L=0; p=0; W=0; iW=0; rN=0; N=0; return false; } L=0x00000020/n; // 17:16 bit best for 16 bit
//    r=2; p=0x0a000001; if ((n0x01000000)) { r=0; L=0; p=0; W=0; iW=0; rN=0; N=0; return false; } L=0x01000000/n; // 28:25 bit
     N=n;                // size of vectors [DWORDs]
     W=modpow(r,    L);    // Wn for NTT
    iW=modpow(r,p-1-L);    // Wn for INTT
    rN=modpow(n,p-2  );    // scale for INTT
    return true;
    }

//---------------------------------------------------------------------------
void fourier_NTT:: NTT_fast(DWORD *dst,DWORD *src,DWORD n,DWORD w)
    {
    if (n>1,w2=modmul(w,w);
    // reorder even,odd
    for (i=0,j=0;i<n2 dst for j="1;i<n" recursion ntt_fast even odd restore results a0="src[i];" a1="modmul(src[j],w2);" void fourier_ntt:: ntt_slow n w dword i a="0;" wi="modmul(wi,wj);" wj="modmul(wj,w);" fourier_ntt::intt_slow fourier_ntt::shl return fourier_ntt::shr>>1)&0x7FFFFFFF; }

//---------------------------------------------------------------------------
DWORD fourier_NTT::mod(DWORD a)
    {
    DWORD bb;</n2></n>

The above is the detailed content of How Can I Optimize My Number Theoretic Transform (NTT) and Modular Arithmetic for Faster Big Number Squaring?. For more information, please follow other related articles on the PHP Chinese website!