inline void extreme_value_distribution_impl(
RNGType &rng, std::size_t n, RealType *r, RealType a, RealType b)
{
u01_oo_distribution(rng, n, r);
log(n, r, r);
mul(n, static_cast<RealType>(-1), r, r);
log(n, r, r);
muladd(n, r, -b, a, r);
}
/* SU= 24 */
void BIG_pxmul(DBIG c,BIG a,int b)
{
int j;
chunk carry;
BIG_dzero(c);
carry=0;
for (j=0;j<NLEN;j++)
carry=muladd(a[j],(chunk)b,carry,&c[j]);
c[NLEN]=carry;
#ifdef DEBUG_NORM
c[DNLEN]=0;
#endif
}
/* SU= 72 */
void BIG_mul(DBIG c,BIG a,BIG b)
{
int i,j;
chunk carry;
#ifdef dchunk
dchunk t,co;
#endif
BIG_norm(a); /* needed here to prevent overflow from addition of partial products */
BIG_norm(b);
/* Faster to Combafy it.. Let the compiler unroll the loops! */
#ifdef COMBA
t=(dchunk)a[0]*b[0];
c[0]=(chunk)t&MASK; co=t>>BASEBITS;
t=(dchunk)a[1]*b[0]+(dchunk)a[0]*b[1]+co;
c[1]=(chunk)t&MASK; co=t>>BASEBITS;
for (j=2;j<NLEN;j++)
{
t=co; for (i=0;i<=j;i++) t+=(dchunk)a[j-i]*b[i];
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
}
for (j=NLEN;j<DNLEN-2;j++)
{
t=co; for (i=j-NLEN+1;i<NLEN;i++) t+=(dchunk)a[j-i]*b[i];
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
}
t=(dchunk)a[NLEN-1]*b[NLEN-1]+co;
c[DNLEN-2]=(chunk)t&MASK; co=t>>BASEBITS;
c[DNLEN-1]=(chunk)co;
#else
BIG_dzero(c);
for (i=0;i<NLEN;i++)
{
carry=0;
for (j=0;j<NLEN;j++)
carry=muladd(a[i],b[j],carry,&c[i+j]);
c[NLEN+i]=carry;
}
#endif
#ifdef DEBUG_NORM
c[DNLEN]=0;
#endif
}
/* SU= 24 */
chunk BIG_pmul(BIG r,BIG a,int c)
{
int i;
chunk ak,carry=0;
BIG_norm(a);
for (i=0;i<NLEN;i++)
{
ak=a[i];
r[i]=0;
carry=muladd(ak,(chunk)c,carry,&r[i]);
}
#ifdef DEBUG_NORM
r[NLEN]=0;
#endif
return carry;
}
/* SU= 40 */
void BIG_smul(BIG c,BIG a,BIG b)
{
int i,j;
chunk carry;
BIG_norm(a);
BIG_norm(b);
BIG_zero(c);
for (i=0;i<NLEN;i++)
{
carry=0;
for (j=0;j<NLEN;j++)
if (i+j<NLEN) carry=muladd(a[i],b[j],carry,&c[i+j]);
}
#ifdef DEBUG_NORM
c[NLEN]=0;
#endif
}
double dot(const fvec4d &ov) const
{
vector_type muladd(v[0] * ov.v[0] + v[1] * ov.v[1]);
return muladd[0] + muladd[1];
}
/* SU= 80 */
void BIG_sqr(DBIG c,BIG a)
{
int i,j;
chunk carry;
#ifdef dchunk
dchunk t,co;
#endif
BIG_norm(a);
/* Note 2*a[i] in loop below and extra addition */
#ifdef COMBA
t=(dchunk)a[0]*a[0];
c[0]=(chunk)t&MASK; co=t>>BASEBITS;
t=(dchunk)a[1]*a[0]; t+=t; t+=co;
c[1]=(chunk)t&MASK; co=t>>BASEBITS;
#if NLEN%2==1
for (j=2;j<NLEN-1;j+=2)
{
t=(dchunk)a[j]*a[0]; for (i=1;i<(j+1)/2;i++) t+=(dchunk)a[j-i]*a[i]; t+=t; t+=co; t+=(dchunk)a[j/2]*a[j/2];
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
t=(dchunk)a[j+1]*a[0]; for (i=1;i<(j+2)/2;i++) t+=(dchunk)a[j+1-i]*a[i]; t+=t; t+=co;
c[j+1]=(chunk)t&MASK; co=t>>BASEBITS;
}
j=NLEN-1;
t=(dchunk)a[j]*a[0]; for (i=1;i<(j+1)/2;i++) t+=(dchunk)a[j-i]*a[i]; t+=t; t+=co; t+=(dchunk)a[j/2]*a[j/2];
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
#else
for (j=2;j<NLEN;j+=2)
{
t=(dchunk)a[j]*a[0]; for (i=1;i<(j+1)/2;i++) t+=(dchunk)a[j-i]*a[i]; t+=t; t+=co; t+=(dchunk)a[j/2]*a[j/2];
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
t=(dchunk)a[j+1]*a[0]; for (i=1;i<(j+2)/2;i++) t+=(dchunk)a[j+1-i]*a[i]; t+=t; t+=co;
c[j+1]=(chunk)t&MASK; co=t>>BASEBITS;
}
#endif
#if NLEN%2==1
j=NLEN;
t=(dchunk)a[NLEN-1]*a[j-NLEN+1]; for (i=j-NLEN+2;i<(j+1)/2;i++) t+=(dchunk)a[j-i]*a[i]; t+=t; t+=co;
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
for (j=NLEN+1;j<DNLEN-2;j+=2)
{
t=(dchunk)a[NLEN-1]*a[j-NLEN+1]; for (i=j-NLEN+2;i<(j+1)/2;i++) t+=(dchunk)a[j-i]*a[i]; t+=t; t+=co; t+=(dchunk)a[j/2]*a[j/2];
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
t=(dchunk)a[NLEN-1]*a[j-NLEN+2]; for (i=j-NLEN+3;i<(j+2)/2;i++) t+=(dchunk)a[j+1-i]*a[i]; t+=t; t+=co;
c[j+1]=(chunk)t&MASK; co=t>>BASEBITS;
}
#else
for (j=NLEN;j<DNLEN-2;j+=2)
{
t=(dchunk)a[NLEN-1]*a[j-NLEN+1]; for (i=j-NLEN+2;i<(j+1)/2;i++) t+=(dchunk)a[j-i]*a[i]; t+=t; t+=co; t+=(dchunk)a[j/2]*a[j/2];
c[j]=(chunk)t&MASK; co=t>>BASEBITS;
t=(dchunk)a[NLEN-1]*a[j-NLEN+2]; for (i=j-NLEN+3;i<(j+2)/2;i++) t+=(dchunk)a[j+1-i]*a[i]; t+=t; t+=co;
c[j+1]=(chunk)t&MASK; co=t>>BASEBITS;
}
#endif
t=(dchunk)a[NLEN-1]*a[NLEN-1]+co;
c[DNLEN-2]=(chunk)t&MASK; co=t>>BASEBITS;
c[DNLEN-1]=(chunk)co;
#else
BIG_dzero(c);
for (i=0;i<NLEN;i++)
{
carry=0;
for (j=i+1;j<NLEN;j++)
carry=muladd(a[i],a[j],carry,&c[i+j]);
c[NLEN+i]=carry;
}
for (i=0;i<DNLEN;i++) c[i]*=2;
for (i=0;i<NLEN;i++)
c[2*i+1]+=muladd(a[i],a[i],0,&c[2*i]);
BIG_dnorm(c);
#endif
#ifdef DEBUG_NORM
c[DNLEN]=0;
#endif
}
