/*Converting from double precision to Multi-precision and calculating e^x */
double
SECTION
__slowexp (double x)
{
#ifndef USE_LONG_DOUBLE_FOR_MP
double w, z, res, eps = 3.0e-26;
int p;
mp_no mpx, mpy, mpz, mpw, mpeps, mpcor;
/* Use the multiple precision __MPEXP function to compute the exponential
First at 144 bits and if it is not accurate enough, at 768 bits. */
p = 6;
__dbl_mp (x, &mpx, p);
__mpexp (&mpx, &mpy, p);
__dbl_mp (eps, &mpeps, p);
__mul (&mpeps, &mpy, &mpcor, p);
__add (&mpy, &mpcor, &mpw, p);
__sub (&mpy, &mpcor, &mpz, p);
__mp_dbl (&mpw, &w, p);
__mp_dbl (&mpz, &z, p);
if (w == z)
return w;
else
{
p = 32;
__dbl_mp (x, &mpx, p);
__mpexp (&mpx, &mpy, p);
__mp_dbl (&mpy, &res, p);
return res;
}
#else
return (double) __ieee754_expl((long double)x);
#endif
}
/* Stage 3: Perform a multi-Precision computation */
static double
SECTION
atan2Mp (double x, double y, const int pr[])
{
double z1, z2;
int i, p;
mp_no mpx, mpy, mpz, mpz1, mpz2, mperr, mpt1;
for (i = 0; i < MM; i++)
{
p = pr[i];
__dbl_mp (x, &mpx, p);
__dbl_mp (y, &mpy, p);
__mpatan2 (&mpy, &mpx, &mpz, p);
__dbl_mp (ud[i].d, &mpt1, p);
__mul (&mpz, &mpt1, &mperr, p);
__add (&mpz, &mperr, &mpz1, p);
__sub (&mpz, &mperr, &mpz2, p);
__mp_dbl (&mpz1, &z1, p);
__mp_dbl (&mpz2, &z2, p);
if (z1 == z2)
{
LIBC_PROBE (slowatan2, 4, &p, &x, &y, &z1);
return z1;
}
}
LIBC_PROBE (slowatan2_inexact, 4, &p, &x, &y, &z1);
return z1; /*if impossible to do exact computing */
}
double __slowpow(double x, double y, double z) {
double res,res1;
mp_no mpx, mpy, mpz,mpw,mpp,mpr,mpr1;
static const mp_no eps = {-3,{1.0,4.0}};
int p;
res = __halfulp(x,y); /* halfulp() returns -10 or x^y */
if (res >= 0) return res; /* if result was really computed by halfulp */
/* else, if result was not really computed by halfulp */
p = 10; /* p=precision */
__dbl_mp(x,&mpx,p);
__dbl_mp(y,&mpy,p);
__dbl_mp(z,&mpz,p);
__mplog(&mpx, &mpz, p); /* log(x) = z */
__mul(&mpy,&mpz,&mpw,p); /* y * z =w */
__mpexp(&mpw, &mpp, p); /* e^w =pp */
__add(&mpp,&eps,&mpr,p); /* pp+eps =r */
__mp_dbl(&mpr, &res, p);
__sub(&mpp,&eps,&mpr1,p); /* pp -eps =r1 */
__mp_dbl(&mpr1, &res1, p); /* converting into double precision */
if (res == res1) return res;
p = 32; /* if we get here result wasn't calculated exactly, continue */
__dbl_mp(x,&mpx,p); /* for more exact calculation */
__dbl_mp(y,&mpy,p);
__dbl_mp(z,&mpz,p);
__mplog(&mpx, &mpz, p); /* log(c)=z */
__mul(&mpy,&mpz,&mpw,p); /* y*z =w */
__mpexp(&mpw, &mpp, p); /* e^w=pp */
__mp_dbl(&mpp, &res, p); /* converting into double precision */
return res;
}
void
SECTION
__mpsqrt (mp_no *x, mp_no *y, int p)
{
int i, m, ey;
double dx, dy;
static const mp_no mphalf = {0, {1.0, HALFRAD}};
static const mp_no mp3halfs = {1, {1.0, 1.0, HALFRAD}};
mp_no mpxn, mpz, mpu, mpt1, mpt2;
ey = EX / 2;
__cpy (x, &mpxn, p);
mpxn.e -= (ey + ey);
__mp_dbl (&mpxn, &dx, p);
dy = fastiroot (dx);
__dbl_mp (dy, &mpu, p);
__mul (&mpxn, &mphalf, &mpz, p);
m = __mpsqrt_mp[p];
for (i = 0; i < m; i++)
{
__sqr (&mpu, &mpt1, p);
__mul (&mpt1, &mpz, &mpt2, p);
__sub (&mp3halfs, &mpt2, &mpt1, p);
__mul (&mpu, &mpt1, &mpt2, p);
__cpy (&mpt2, &mpu, p);
}
__mul (&mpxn, &mpu, y, p);
EY += ey;
}
/* Final stages. Compute atan(x) by multiple precision arithmetic */
static double
atanMp (double x, const int pr[])
{
mp_no mpx, mpy, mpy2, mperr, mpt1, mpy1;
double y1, y2;
int i, p;
for (i = 0; i < M; i++)
{
p = pr[i];
__dbl_mp (x, &mpx, p);
__mpatan (&mpx, &mpy, p);
__dbl_mp (u9[i].d, &mpt1, p);
__mul (&mpy, &mpt1, &mperr, p);
__add (&mpy, &mperr, &mpy1, p);
__sub (&mpy, &mperr, &mpy2, p);
__mp_dbl (&mpy1, &y1, p);
__mp_dbl (&mpy2, &y2, p);
if (y1 == y2)
{
LIBC_PROBE (slowatan, 3, &p, &x, &y1);
return y1;
}
}
LIBC_PROBE (slowatan_inexact, 3, &p, &x, &y1);
return y1; /*if impossible to do exact computing */
}
void __mplog(mp_no *x, mp_no *y, int p) {
#include "mplog.h"
int i,m;
#if 0
int j,k,m1,m2,n;
double a,b;
#endif
static const int mp[33] = {0,0,0,0,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,
4,4,4,4,4,4,4,4,4,4,4,4,4,4};
mp_no mpone = {0,{0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}};
mp_no mpt1,mpt2;
/* Choose m and initiate mpone */
m = mp[p]; mpone.e = 1; mpone.d[0]=mpone.d[1]=ONE;
/* Perform m newton iterations to solve for y: exp(y)-x=0. */
/* The iterations formula is: y(n+1)=y(n)+(x*exp(-y(n))-1). */
__cpy(y,&mpt1,p);
for (i=0; i<m; i++) {
mpt1.d[0]=-mpt1.d[0];
__mpexp(&mpt1,&mpt2,p);
__mul(x,&mpt2,&mpt1,p);
__sub(&mpt1,&mpone,&mpt2,p);
__add(y,&mpt2,&mpt1,p);
__cpy(&mpt1,y,p);
}
return;
}
static
SECTION
void __inv(const mp_no *x, mp_no *y, int p) {
int i;
#if 0
int l;
#endif
double t;
mp_no z,w;
static const int np1[] = {0,0,0,0,1,2,2,2,2,3,3,3,3,3,3,3,3,3,
4,4,4,4,4,4,4,4,4,4,4,4,4,4,4};
const mp_no mptwo = {1,{1.0,2.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}};
__cpy(x,&z,p); z.e=0; __mp_dbl(&z,&t,p);
t=ONE/t; __dbl_mp(t,y,p); EY -= EX;
for (i=0; i<np1[p]; i++) {
__cpy(y,&w,p);
__mul(x,&w,y,p);
__sub(&mptwo,y,&z,p);
__mul(&w,&z,y,p);
}
}
int
SECTION
__mpranred(double x, mp_no *y, int p)
{
number v;
double t,xn;
int i,k,n;
static const mp_no one = {1,{1.0,1.0}};
mp_no a,b,c;
if (ABS(x) < 2.8e14) {
t = (x*hpinv.d + toint.d);
xn = t - toint.d;
v.d = t;
n =v.i[LOW_HALF]&3;
__dbl_mp(xn,&a,p);
__mul(&a,&hp,&b,p);
__dbl_mp(x,&c,p);
__sub(&c,&b,y,p);
return n;
}
else { /* if x is very big more precision required */
__dbl_mp(x,&a,p);
a.d[0]=1.0;
k = a.e-5;
if (k < 0) k=0;
b.e = -k;
b.d[0] = 1.0;
for (i=0;i<p;i++) b.d[i+1] = toverp[i+k];
__mul(&a,&b,&c,p);
t = c.d[c.e];
for (i=1;i<=p-c.e;i++) c.d[i]=c.d[i+c.e];
for (i=p+1-c.e;i<=p;i++) c.d[i]=0;
c.e=0;
if (c.d[1] >= 8388608.0)
{ t +=1.0;
__sub(&c,&one,&b,p);
__mul(&b,&hp,y,p);
}
else __mul(&c,&hp,y,p);
n = (int) t;
if (x < 0) { y->d[0] = - y->d[0]; n = -n; }
return (n&3);
}
}
double __sin32(double x, double res, double res1) {
int p;
mp_no a,b,c;
p=32;
__dbl_mp(res,&a,p);
__dbl_mp(0.5*(res1-res),&b,p);
__add(&a,&b,&c,p);
if (x>0.8)
{ __sub(&hp,&c,&a,p);
__c32(&a,&b,&c,p);
}
else __c32(&c,&a,&b,p); /* b=sin(0.5*(res+res1)) */
__dbl_mp(x,&c,p); /* c = x */
__sub(&b,&c,&a,p);
/* if a>0 return min(res,res1), otherwise return max(res,res1) */
if (a.d[0]>0) return (res<res1)?res:res1;
else return (res>res1)?res:res1;
}
void __c32(mp_no *x, mp_no *y, mp_no *z, int p) {
static const mp_no mpt={1,{1.0,2.0}}, one={1,{1.0,1.0}};
mp_no u,t,t1,t2,c,s;
int i;
__cpy(x,&u,p);
u.e=u.e-1;
cc32(&u,&c,p);
ss32(&u,&s,p);
for (i=0;i<24;i++) {
__mul(&c,&s,&t,p);
__sub(&s,&t,&t1,p);
__add(&t1,&t1,&s,p);
__sub(&mpt,&c,&t1,p);
__mul(&t1,&c,&t2,p);
__add(&t2,&t2,&c,p);
}
__sub(&one,&c,y,p);
__cpy(&s,z,p);
}
/* Compute cos() of double-length number (X + DX) as Multi Precision number and
return result as double. If REDUCE_RANGE is true, X is assumed to be the
original input and DX is ignored. */
double
SECTION
__mpcos (double x, double dx, bool reduce_range)
{
double y;
mp_no a, b, c, s;
int n;
int p = 32;
if (reduce_range)
{
n = __mpranred (x, &a, p); /* n is 0, 1, 2 or 3. */
__c32 (&a, &c, &s, p);
}
else
{
n = -1;
__dbl_mp (x, &b, p);
__dbl_mp (dx, &c, p);
__add (&b, &c, &a, p);
if (x > 0.8)
{
__sub (&hp, &a, &b, p);
__c32 (&b, &s, &c, p);
}
else
__c32 (&a, &c, &s, p); /* a = cos(x+dx) */
}
/* Convert result based on which quarter of unit circle y is in. */
switch (n)
{
case 1:
__mp_dbl (&s, &y, p);
y = -y;
break;
case 3:
__mp_dbl (&s, &y, p);
break;
case 2:
__mp_dbl (&c, &y, p);
y = -y;
break;
/* Quadrant not set, so the result must be cos (X + DX), which is also
stored in C. */
case 0:
default:
__mp_dbl (&c, &y, p);
}
LIBC_PROBE (slowcos, 3, &x, &dx, &y);
return y;
}
void
SECTION
__c32(mp_no *x, mp_no *y, mp_no *z, int p) {
mp_no u,t,t1,t2,c,s;
int i;
__cpy(x,&u,p);
u.e=u.e-1;
cc32(&u,&c,p);
ss32(&u,&s,p);
for (i=0;i<24;i++) {
__mul(&c,&s,&t,p);
__sub(&s,&t,&t1,p);
__add(&t1,&t1,&s,p);
__sub(&mptwo,&c,&t1,p);
__mul(&t1,&c,&t2,p);
__add(&t2,&t2,&c,p);
}
__sub(&mpone,&c,y,p);
__cpy(&s,z,p);
}
double __mpsin(double x, double dx) {
int p;
double y;
mp_no a,b,c;
p=32;
__dbl_mp(x,&a,p);
__dbl_mp(dx,&b,p);
__add(&a,&b,&c,p);
if (x>0.8) { __sub(&hp,&c,&a,p); __c32(&a,&b,&c,p); }
else __c32(&c,&a,&b,p); /* b = sin(x+dx) */
__mp_dbl(&b,&y,p);
return y;
}
/* Treat the Denormalized case */
static double
SECTION
normalized (double ax, double ay, double y, double z)
{
int p;
mp_no mpx, mpy, mpz, mperr, mpz2, mpt1;
p = 6;
__dbl_mp (ax, &mpx, p);
__dbl_mp (ay, &mpy, p);
__dvd (&mpy, &mpx, &mpz, p);
__dbl_mp (ue.d, &mpt1, p);
__mul (&mpz, &mpt1, &mperr, p);
__sub (&mpz, &mperr, &mpz2, p);
__mp_dbl (&mpz2, &z, p);
return signArctan2 (y, z);
}
static void __subTest(int only)
{
if (only && (2 != only)) {
return;
}
std::cout << "\n****************************" << __func__ << "()\n";
org::yuiwong::times::Datetime t1;
org::yuiwong::times::Datetime t2;
for (int i = 0; i < 3; ++i) {
t1.update();
std::cout << "Wait ..." << "\n";
usleep(i * 500 * 1e3);
usleep(100 * 1e3);
t2.update();
__sub(t1, t2);
}
}
/*Converting from double precision to Multi-precision and calculating e^x */
double
SECTION
__slowexp (double x)
{
#ifndef USE_LONG_DOUBLE_FOR_MP
double w, z, res, eps = 3.0e-26;
int p;
mp_no mpx, mpy, mpz, mpw, mpeps, mpcor;
/* Use the multiple precision __MPEXP function to compute the exponential
First at 144 bits and if it is not accurate enough, at 768 bits. */
p = 6;
__dbl_mp (x, &mpx, p);
__mpexp (&mpx, &mpy, p);
__dbl_mp (eps, &mpeps, p);
__mul (&mpeps, &mpy, &mpcor, p);
__add (&mpy, &mpcor, &mpw, p);
__sub (&mpy, &mpcor, &mpz, p);
__mp_dbl (&mpw, &w, p);
__mp_dbl (&mpz, &z, p);
if (w == z)
{
/* Track how often we get to the slow exp code plus
its input/output values. */
LIBC_PROBE (slowexp_p6, 2, &x, &w);
return w;
}
else
{
p = 32;
__dbl_mp (x, &mpx, p);
__mpexp (&mpx, &mpy, p);
__mp_dbl (&mpy, &res, p);
/* Track how often we get to the uber-slow exp code plus
its input/output values. */
LIBC_PROBE (slowexp_p32, 2, &x, &res);
return res;
}
#else
return (double) __ieee754_expl((long double)x);
#endif
}
static void
SECTION
ss32(mp_no *x, mp_no *y, int p) {
int i;
double a;
mp_no mpt1,x2,gor,sum ,mpk={1,{1.0}};
for (i=1;i<=p;i++) mpk.d[i]=0;
__sqr(x,&x2,p);
__cpy(&oofac27,&gor,p);
__cpy(&gor,&sum,p);
for (a=27.0;a>1.0;a-=2.0) {
mpk.d[1]=a*(a-1.0);
__mul(&gor,&mpk,&mpt1,p);
__cpy(&mpt1,&gor,p);
__mul(&x2,&sum,&mpt1,p);
__sub(&gor,&mpt1,&sum,p);
}
__mul(x,&sum,y,p);
}
/*Converting from double precision to Multi-precision and calculating e^x */
double __slowexp(double x) {
double w,z,res,eps=3.0e-26;
#if 0
double y;
#endif
int p;
#if 0
int orig,i;
#endif
mp_no mpx, mpy, mpz,mpw,mpeps,mpcor;
p=6;
__dbl_mp(x,&mpx,p); /* Convert a double precision number x */
/* into a multiple precision number mpx with prec. p. */
__mpexp(&mpx, &mpy, p); /* Multi-Precision exponential function */
__dbl_mp(eps,&mpeps,p);
__mul(&mpeps,&mpy,&mpcor,p);
__add(&mpy,&mpcor,&mpw,p);
__sub(&mpy,&mpcor,&mpz,p);
__mp_dbl(&mpw, &w, p);
__mp_dbl(&mpz, &z, p);
if (w == z) {
/* Track how often we get to the slow exp code plus
its input/output values. */
LIBC_PROBE (slowexp_p6, 2, &x, &w);
return w;
}
else { /* if calculating is not exactly */
p = 32;
__dbl_mp(x,&mpx,p);
__mpexp(&mpx, &mpy, p);
__mp_dbl(&mpy, &res, p);
/* Track how often we get to the uber-slow exp code plus
its input/output values. */
LIBC_PROBE (slowexp_p32, 2, &x, &res);
return res;
}
}
void
SECTION
__mpatan(mp_no *x, mp_no *y, int p) {
int i,m,n;
double dx;
mp_no
mpone = {0,{0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}},
mptwo = {0,{0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}},
mptwoim1 = {0,{0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}};
mp_no mps,mpsm,mpt,mpt1,mpt2,mpt3;
/* Choose m and initiate mpone, mptwo & mptwoim1 */
if (EX>0) m=7;
else if (EX<0) m=0;
else {
__mp_dbl(x,&dx,p); dx=ABS(dx);
for (m=6; m>0; m--)
{if (dx>__atan_xm[m].d) break;}
}
mpone.e = mptwo.e = mptwoim1.e = 1;
mpone.d[0] = mpone.d[1] = mptwo.d[0] = mptwoim1.d[0] = ONE;
mptwo.d[1] = TWO;
/* Reduce x m times */
__mul(x,x,&mpsm,p);
if (m==0) __cpy(x,&mps,p);
else {
for (i=0; i<m; i++) {
__add(&mpone,&mpsm,&mpt1,p);
__mpsqrt(&mpt1,&mpt2,p);
__add(&mpt2,&mpt2,&mpt1,p);
__add(&mptwo,&mpsm,&mpt2,p);
__add(&mpt1,&mpt2,&mpt3,p);
__dvd(&mpsm,&mpt3,&mpt1,p);
__cpy(&mpt1,&mpsm,p);
}
__mpsqrt(&mpsm,&mps,p); mps.d[0] = X[0];
}
/* Evaluate a truncated power series for Atan(s) */
n=__atan_np[p]; mptwoim1.d[1] = __atan_twonm1[p].d;
__dvd(&mpsm,&mptwoim1,&mpt,p);
for (i=n-1; i>1; i--) {
mptwoim1.d[1] -= TWO;
__dvd(&mpsm,&mptwoim1,&mpt1,p);
__mul(&mpsm,&mpt,&mpt2,p);
__sub(&mpt1,&mpt2,&mpt,p);
}
__mul(&mps,&mpt,&mpt1,p);
__sub(&mps,&mpt1,&mpt,p);
/* Compute Atan(x) */
mptwoim1.d[1] = __atan_twom[m].d;
__mul(&mptwoim1,&mpt,y,p);
return;
}
slong _nmod_poly_xgcd_hgcd(mp_ptr G, mp_ptr S, mp_ptr T,
mp_srcptr A, slong lenA, mp_srcptr B, slong lenB,
nmod_t mod)
{
const slong cutoff = FLINT_BIT_COUNT(mod.n) <= 8 ?
NMOD_POLY_SMALL_GCD_CUTOFF : NMOD_POLY_GCD_CUTOFF;
slong lenG, lenS, lenT;
if (lenB == 1)
{
G[0] = B[0];
T[0] = 1;
lenG = 1;
lenS = 0;
lenT = 1;
}
else
{
mp_ptr q = _nmod_vec_init(lenA + lenB);
mp_ptr r = q + lenA;
slong lenq, lenr;
__divrem(q, lenq, r, lenr, A, lenA, B, lenB);
if (lenr == 0)
{
__set(G, lenG, B, lenB);
T[0] = 1;
lenS = 0;
lenT = 1;
}
else
{
mp_ptr h, j, v, w, R[4], X;
slong lenh, lenj, lenv, lenw, lenR[4];
int sgnR;
lenh = lenj = lenB;
lenv = lenw = lenA + lenB - 2;
lenR[0] = lenR[1] = lenR[2] = lenR[3] = (lenB + 1) / 2;
X = _nmod_vec_init(2 * lenh + 2 * lenv + 4 * lenR[0]);
h = X;
j = h + lenh;
v = j + lenj;
w = v + lenv;
R[0] = w + lenw;
R[1] = R[0] + lenR[0];
R[2] = R[1] + lenR[1];
R[3] = R[2] + lenR[2];
sgnR = _nmod_poly_hgcd(R, lenR, h, &lenh, j, &lenj, B, lenB, r, lenr, mod);
if (sgnR > 0)
{
_nmod_vec_neg(S, R[1], lenR[1], mod);
_nmod_vec_set(T, R[0], lenR[0]);
}
else
{
_nmod_vec_set(S, R[1], lenR[1]);
_nmod_vec_neg(T, R[0], lenR[0], mod);
}
lenS = lenR[1];
lenT = lenR[0];
while (lenj != 0)
{
__divrem(q, lenq, r, lenr, h, lenh, j, lenj);
__mul(v, lenv, q, lenq, T, lenT);
{
slong l;
_nmod_vec_swap(S, T, FLINT_MAX(lenS, lenT));
l = lenS; lenS = lenT; lenT = l;
}
__sub(T, lenT, T, lenT, v, lenv);
if (lenr == 0)
{
__set(G, lenG, j, lenj);
goto cofactor;
}
if (lenj < cutoff)
{
mp_ptr u0 = R[0], u1 = R[1];
slong lenu0 = lenr - 1, lenu1 = lenj - 1;
lenG = _nmod_poly_xgcd_euclidean(G, u0, u1, j, lenj, r, lenr, mod);
MPN_NORM(u0, lenu0);
MPN_NORM(u1, lenu1);
__mul(v, lenv, S, lenS, u0, lenu0);
__mul(w, lenw, T, lenT, u1, lenu1);
__add(S, lenS, v, lenv, w, lenw);
goto cofactor;
}
sgnR = _nmod_poly_hgcd(R, lenR, h, &lenh, j, &lenj, j,lenj, r, lenr, mod);
__mul(v, lenv, R[1], lenR[1], T, lenT);
__mul(w, lenw, R[2], lenR[2], S, lenS);
__mul(q, lenq, S, lenS, R[3], lenR[3]);
if (sgnR > 0)
__sub(S, lenS, q, lenq, v, lenv);
else
__sub(S, lenS, v, lenv, q, lenq);
__mul(q, lenq, T, lenT, R[0], lenR[0]);
if (sgnR > WORD(0))
__sub(T, lenT, q, lenq, w, lenw);
else
__sub(T, lenT, w, lenw, q, lenq);
}
__set(G, lenG, h, lenh);
cofactor:
__mul(v, lenv, S, lenS, A, lenA);
__sub(w, lenw, G, lenG, v, lenv);
__div(T, lenT, w, lenw, B, lenB);
_nmod_vec_clear(X);
}
_nmod_vec_clear(q);
}
flint_mpn_zero(S + lenS, lenB - 1 - lenS);
flint_mpn_zero(T + lenT, lenA - 1 - lenT);
return lenG;
}
double
SECTION
__ieee754_log(double x) {
#define M 4
static const int pr[M]={8,10,18,32};
int i,j,n,ux,dx,p;
#if 0
int k;
#endif
double dbl_n,u,p0,q,r0,w,nln2a,luai,lubi,lvaj,lvbj,
sij,ssij,ttij,A,B,B0,y,y1,y2,polI,polII,sa,sb,
t1,t2,t7,t8,t,ra,rb,ww,
a0,aa0,s1,s2,ss2,s3,ss3,a1,aa1,a,aa,b,bb,c;
#ifndef DLA_FMS
double t3,t4,t5,t6;
#endif
number num;
mp_no mpx,mpy,mpy1,mpy2,mperr;
#include "ulog.tbl"
#include "ulog.h"
/* Treating special values of x ( x<=0, x=INF, x=NaN etc.). */
num.d = x; ux = num.i[HIGH_HALF]; dx = num.i[LOW_HALF];
n=0;
if (__builtin_expect(ux < 0x00100000, 0)) {
if (__builtin_expect(((ux & 0x7fffffff) | dx) == 0, 0))
return MHALF/ZERO; /* return -INF */
if (__builtin_expect(ux < 0, 0))
return (x-x)/ZERO; /* return NaN */
n -= 54; x *= two54.d; /* scale x */
num.d = x;
}
if (__builtin_expect(ux >= 0x7ff00000, 0))
return x+x; /* INF or NaN */
/* Regular values of x */
w = x-ONE;
if (__builtin_expect(ABS(w) > U03, 1)) { goto case_03; }
/*--- Stage I, the case abs(x-1) < 0.03 */
t8 = MHALF*w;
EMULV(t8,w,a,aa,t1,t2,t3,t4,t5)
EADD(w,a,b,bb)
/* Evaluate polynomial II */
polII = (b0.d+w*(b1.d+w*(b2.d+w*(b3.d+w*(b4.d+
w*(b5.d+w*(b6.d+w*(b7.d+w*b8.d))))))))*w*w*w;
c = (aa+bb)+polII;
/* End stage I, case abs(x-1) < 0.03 */
if ((y=b+(c+b*E2)) == b+(c-b*E2)) return y;
/*--- Stage II, the case abs(x-1) < 0.03 */
a = d11.d+w*(d12.d+w*(d13.d+w*(d14.d+w*(d15.d+w*(d16.d+
w*(d17.d+w*(d18.d+w*(d19.d+w*d20.d))))))));
EMULV(w,a,s2,ss2,t1,t2,t3,t4,t5)
ADD2(d10.d,dd10.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d9.d,dd9.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d8.d,dd8.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d7.d,dd7.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d6.d,dd6.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d5.d,dd5.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d4.d,dd4.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d3.d,dd3.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(d2.d,dd2.d,s2,ss2,s3,ss3,t1,t2)
MUL2(w,ZERO,s3,ss3,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
MUL2(w,ZERO,s2,ss2,s3,ss3,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(w,ZERO, s3,ss3, b, bb,t1,t2)
/* End stage II, case abs(x-1) < 0.03 */
if ((y=b+(bb+b*E4)) == b+(bb-b*E4)) return y;
goto stage_n;
/*--- Stage I, the case abs(x-1) > 0.03 */
case_03:
/* Find n,u such that x = u*2**n, 1/sqrt(2) < u < sqrt(2) */
n += (num.i[HIGH_HALF] >> 20) - 1023;
num.i[HIGH_HALF] = (num.i[HIGH_HALF] & 0x000fffff) | 0x3ff00000;
if (num.d > SQRT_2) { num.d *= HALF; n++; }
u = num.d; dbl_n = (double) n;
/* Find i such that ui=1+(i-75)/2**8 is closest to u (i= 0,1,2,...,181) */
num.d += h1.d;
i = (num.i[HIGH_HALF] & 0x000fffff) >> 12;
/* Find j such that vj=1+(j-180)/2**16 is closest to v=u/ui (j= 0,...,361) */
num.d = u*Iu[i].d + h2.d;
j = (num.i[HIGH_HALF] & 0x000fffff) >> 4;
/* Compute w=(u-ui*vj)/(ui*vj) */
p0=(ONE+(i-75)*DEL_U)*(ONE+(j-180)*DEL_V);
q=u-p0; r0=Iu[i].d*Iv[j].d; w=q*r0;
/* Evaluate polynomial I */
polI = w+(a2.d+a3.d*w)*w*w;
/* Add up everything */
nln2a = dbl_n*LN2A;
luai = Lu[i][0].d; lubi = Lu[i][1].d;
lvaj = Lv[j][0].d; lvbj = Lv[j][1].d;
EADD(luai,lvaj,sij,ssij)
EADD(nln2a,sij,A ,ttij)
B0 = (((lubi+lvbj)+ssij)+ttij)+dbl_n*LN2B;
B = polI+B0;
/* End stage I, case abs(x-1) >= 0.03 */
if ((y=A+(B+E1)) == A+(B-E1)) return y;
/*--- Stage II, the case abs(x-1) > 0.03 */
/* Improve the accuracy of r0 */
EMULV(p0,r0,sa,sb,t1,t2,t3,t4,t5)
t=r0*((ONE-sa)-sb);
EADD(r0,t,ra,rb)
/* Compute w */
MUL2(q,ZERO,ra,rb,w,ww,t1,t2,t3,t4,t5,t6,t7,t8)
EADD(A,B0,a0,aa0)
/* Evaluate polynomial III */
s1 = (c3.d+(c4.d+c5.d*w)*w)*w;
EADD(c2.d,s1,s2,ss2)
MUL2(s2,ss2,w,ww,s3,ss3,t1,t2,t3,t4,t5,t6,t7,t8)
MUL2(s3,ss3,w,ww,s2,ss2,t1,t2,t3,t4,t5,t6,t7,t8)
ADD2(s2,ss2,w,ww,s3,ss3,t1,t2)
ADD2(s3,ss3,a0,aa0,a1,aa1,t1,t2)
/* End stage II, case abs(x-1) >= 0.03 */
if ((y=a1+(aa1+E3)) == a1+(aa1-E3)) return y;
/* Final stages. Use multi-precision arithmetic. */
stage_n:
for (i=0; i<M; i++) {
p = pr[i];
__dbl_mp(x,&mpx,p); __dbl_mp(y,&mpy,p);
__mplog(&mpx,&mpy,p);
__dbl_mp(e[i].d,&mperr,p);
__add(&mpy,&mperr,&mpy1,p); __sub(&mpy,&mperr,&mpy2,p);
__mp_dbl(&mpy1,&y1,p); __mp_dbl(&mpy2,&y2,p);
if (y1==y2) return y1;
}
return y1;
}
