/* Dickman's rho function (to compute probability of success of ecm).

Copyright 2004, 2005, 2006, 2008, 2009, 2010, 2011, 2012, 2013
Alexander Kruppa, Paul Zimmermann.

This file is part of the ECM Library.

The ECM Library is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at your
option) any later version.

The ECM Library 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 Lesser General Public
License for more details.

You should have received a copy of the GNU Lesser General Public License
along with the ECM Library; see the file COPYING.LIB.  If not, see
http://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. */

/* define TESTDRIVE to compile rho as a stand-alone program, in which case
   you need to have libgsl installed */

#include "config.h"
#if defined(TESTDRIVE)
#define _ISOC99_SOURCE 1
#endif
#if defined(DEBUG_NUMINTEGRATE) || defined(TESTDRIVE)
# include <stdio.h>
#endif
#include <stdlib.h>
#include <math.h>
#if defined(TESTDRIVE)
#include <string.h>
#include <primesieve.h>
#endif
#if defined(TESTDRIVE)
#include <gsl/gsl_math.h>
#include <gsl/gsl_sf_expint.h>
#include <gsl/gsl_integration.h>
#endif
#include "ecm-impl.h"

/* For Suyama's curves, we have a known torsion factor of 12 = 2^2*3^1, and
   an average extra exponent of 1/2 for 2, and 1/3 for 3 due to the probability
   that the group order divided by 12 is divisible by 2 or 3, thus on average
   we should have 2^2.5*3^1.333 ~ 24.5, however experimentally we have
   2^3.323*3^1.687 ~ 63.9 (see Alexander Kruppa's thesis, Table 5.1 page 96,
   row sigma=2, http://tel.archives-ouvertes.fr/tel-00477005/en/).
   The exp(ECM_EXTRA_SMOOTHNESS) value takes into account the extra
   smoothness with respect to a random number. */
#ifndef ECM_EXTRA_SMOOTHNESS
#define ECM_EXTRA_SMOOTHNESS 3.134
#endif

#define M_PI_SQR   9.869604401089358619 /* Pi^2 */
#define M_PI_SQR_6 1.644934066848226436 /* Pi^2/6 */
/* gsl_math.h defines M_EULER */
#ifndef M_EULER
#define M_EULER    0.577215664901532861
#endif
#define M_EULER_1   0.422784335098467139 /* 1 - Euler */

#ifndef MAX
#define MAX(x,y) ((x) > (y) ? (x) : (y))
#endif
#ifndef MIN
#define MIN(x,y) ((x) < (y) ? (x) : (y))
#endif

void rhoinit (int, int); /* used in stage2.c */

static double *rhotable = NULL;
static int invh = 0;
static double h = 0.;
static int tablemax = 0;
#if defined(TESTDRIVE)
#define PRIME_PI_MAX 10000
#define PRIME_PI_MAP(x) (((x)+1)/2)
/* The number of primes up to i. Use prime_pi[PRIME_PI_MAP(i)].
   Only correct for i >= 2. */
static unsigned int prime_pi[PRIME_PI_MAP(PRIME_PI_MAX)+1];
#endif

/* Fixme: need prime generating funcion without static state variables */
const unsigned char primemap[667] = {
  254, 223, 239, 126, 182, 219, 61, 249, 213, 79, 30, 243, 234, 166, 237, 158, 
  230, 12, 211, 211, 59, 221, 89, 165, 106, 103, 146, 189, 120, 30, 166, 86, 
  86, 227, 173, 45, 222, 42, 76, 85, 217, 163, 240, 159, 3, 84, 161, 248, 46, 
  253, 68, 233, 102, 246, 19, 58, 184, 76, 43, 58, 69, 17, 191, 84, 140, 193, 
  122, 179, 200, 188, 140, 79, 33, 88, 113, 113, 155, 193, 23, 239, 84, 150, 
  26, 8, 229, 131, 140, 70, 114, 251, 174, 101, 146, 143, 88, 135, 210, 146, 
  216, 129, 101, 38, 227, 160, 17, 56, 199, 38, 60, 129, 235, 153, 141, 81, 
  136, 62, 36, 243, 51, 77, 90, 139, 28, 167, 42, 180, 88, 76, 78, 38, 246, 
  25, 130, 220, 131, 195, 44, 241, 56, 2, 181, 205, 205, 2, 178, 74, 148, 12, 
  87, 76, 122, 48, 67, 11, 241, 203, 68, 108, 36, 248, 25, 1, 149, 168, 92, 
  115, 234, 141, 36, 150, 43, 80, 166, 34, 30, 196, 209, 72, 6, 212, 58, 47, 
  116, 156, 7, 106, 5, 136, 191, 104, 21, 46, 96, 85, 227, 183, 81, 152, 8, 
  20, 134, 90, 170, 69, 77, 73, 112, 39, 210, 147, 213, 202, 171, 2, 131, 97, 
  5, 36, 206, 135, 34, 194, 169, 173, 24, 140, 77, 120, 209, 137, 22, 176, 87, 
  199, 98, 162, 192, 52, 36, 82, 174, 90, 64, 50, 141, 33, 8, 67, 52, 182, 
  210, 182, 217, 25, 225, 96, 103, 26, 57, 96, 208, 68, 122, 148, 154, 9, 136, 
  131, 168, 116, 85, 16, 39, 161, 93, 104, 30, 35, 200, 50, 224, 25, 3, 68, 
  115, 72, 177, 56, 195, 230, 42, 87, 97, 152, 181, 28, 10, 104, 197, 129, 
  143, 172, 2, 41, 26, 71, 227, 148, 17, 78, 100, 46, 20, 203, 61, 220, 20, 
  197, 6, 16, 233, 41, 177, 130, 233, 48, 71, 227, 52, 25, 195, 37, 10, 48, 
  48, 180, 108, 193, 229, 70, 68, 216, 142, 76, 93, 34, 36, 112, 120, 146, 
  137, 129, 130, 86, 38, 27, 134, 233, 8, 165, 0, 211, 195, 41, 176, 194, 74, 
  16, 178, 89, 56, 161, 29, 66, 96, 199, 34, 39, 140, 200, 68, 26, 198, 139, 
  130, 129, 26, 70, 16, 166, 49, 9, 240, 84, 47, 24, 210, 216, 169, 21, 6, 46, 
  12, 246, 192, 14, 80, 145, 205, 38, 193, 24, 56, 101, 25, 195, 86, 147, 139, 
  42, 45, 214, 132, 74, 97, 10, 165, 44, 9, 224, 118, 196, 106, 60, 216, 8, 
  232, 20, 102, 27, 176, 164, 2, 99, 54, 16, 49, 7, 213, 146, 72, 66, 18, 195, 
  138, 160, 159, 45, 116, 164, 130, 133, 120, 92, 13, 24, 176, 97, 20, 29, 2, 
  232, 24, 18, 193, 1, 73, 28, 131, 48, 103, 51, 161, 136, 216, 15, 12, 244, 
  152, 136, 88, 215, 102, 66, 71, 177, 22, 168, 150, 8, 24, 65, 89, 21, 181, 
  68, 42, 82, 225, 179, 170, 161, 89, 69, 98, 85, 24, 17, 165, 12, 163, 60, 
  103, 0, 190, 84, 214, 10, 32, 54, 107, 130, 12, 21, 8, 126, 86, 145, 1, 120, 
  208, 97, 10, 132, 168, 44, 1, 87, 14, 86, 160, 80, 11, 152, 140, 71, 108, 
  32, 99, 16, 196, 9, 228, 12, 87, 136, 11, 117, 11, 194, 82, 130, 194, 57, 
  36, 2, 44, 86, 37, 122, 49, 41, 214, 163, 32, 225, 177, 24, 176, 12, 138, 
  50, 193, 17, 50, 9, 197, 173, 48, 55, 8, 188, 145, 130, 207, 32, 37, 107, 
  156, 48, 143, 68, 38, 70, 106, 7, 73, 142, 9, 88, 16, 2, 37, 197, 196, 66, 
  90, 128, 160, 128, 60, 144, 40, 100, 20, 225, 3, 132, 81, 12, 46, 163, 138, 
  164, 8, 192, 71, 126, 211, 43, 3, 205, 84, 42, 0, 4, 179, 146, 108, 66, 41, 
  76, 131, 193, 146, 204, 28};

#ifdef TESTDRIVE
unsigned long
gcd (unsigned long a, unsigned long b)
{
  unsigned long t;

  while (b != 0)
    {
      t = a % b;
      a = b;
      b = t;
    }

  return a;
}

unsigned long
eulerphi (unsigned long n)
{
  unsigned long phi = 1, p;

  for (p = 2; p * p <= n; p += 2)
    {
      if (n % p == 0)
        {
          phi *= p - 1;
          n /= p;
          while (n % p == 0)
            {
              phi *= p;
              n /= p;
            }
        }

      if (p == 2)
        p--;
    }

  /* now n is prime */

  return (n == 1) ? phi : phi * (n - 1);
}


/* The number of positive integers up to x that have no prime factor <= y,
   for x >= y >= 2. Uses Buchstab's identity */
unsigned long
Buchstab_Phi(unsigned long x, unsigned long y) 
{
  unsigned long p, s;
  primesieve_iterator pg[1];

  if (x < 1)
    return 0;
  if (x <= y)
    return 1;
#if 0
  if (x < y^2)
    return(1 + primepi(x) - primepi (y)));
#endif

  s = 1;
  primesieve_init (pg);
  primesieve_skipto (pg, y, x+1);
  for (p = primesieve_next_prime (pg); p <= x; p = primesieve_next_prime (pg))
    s += Buchstab_Phi(x / p, p - 1);
  return (s);
}


/* The number of positive integers up to x that have no prime factor
   greter than y, for x >= y >= 2. Uses Buchstab's identity */
unsigned long 
Buchstab_Psi(const unsigned long x, const unsigned long y) 
{
  unsigned long r, p;
  primesieve_iterator pg[1];

  if (x <= y)
    return (x);

  if (y == 1UL)
    return (1);

  /* If y^2 > x, then
     Psi(x,y) = x - \sum_{y < p < x, p prime} floor(x/p)

     We separate the sum into ranges where floor(x/p) = k,
     which is x/(k+1) < p <= x/k.
     We also need to satisfy y < p, so we need k < x/y - 1,
     or k_max = ceil (x/y) - 2.
     The primes y < p <= x/(k_max + 1) are summed separately. */
  if (x <= PRIME_PI_MAX && x < y * y)
    {
      unsigned long kmax = x / y - 1;
      unsigned long s1, s2, k;
      
        s1 = (kmax + 1) * (prime_pi [PRIME_PI_MAP(x / (kmax + 1))] - 
                           prime_pi [PRIME_PI_MAP(y)]);
        s2 = 0;
        for (k = 1; k <= kmax; k++)
          s2 += prime_pi[PRIME_PI_MAP(x / k)];
        s2 -= kmax * prime_pi [PRIME_PI_MAP(x / (kmax+1))];
        return (x - s1 - s2);
    }

  r = 1;
  primesieve_init (pg);
  for (p = primesieve_next_prime (pg); p <= y; p = primesieve_next_prime (pg))
    r += Buchstab_Psi (x / p, p);
  return (r);
}

#endif /* TESTDRIVE */


#if defined(TESTDRIVE)
static double
Li (const double x)
{
  return (- gsl_sf_expint_E1 (- log(x)));
}
#endif

/*
  Evaluate dilogarithm via the sum 
  \Li_{2}(z)=\sum_{k=1}^{\infty} \frac{z^k}{k^2}, 
  see http://mathworld.wolfram.com/Dilogarithm.html
  Assumes |z| <= 0.5, for which the sum converges quickly.
 */

static double
dilog_series (const double z)
{
  double r = 0.0, zk; /* zk = z^k */
  int k, k2; /* k2 = k^2 */
  /* Doubles have 53 bits in significand, with |z| <= 0.5 the k+1-st term
     is <= 1/(2^k k^2) of the result, so 44 terms should do */
  for (k = 1, k2 = 1, zk = z; k <= 44; k2 += 2 * k + 1, k++, zk *= z)
    r += zk / (double) k2;

  return r;
}

static double
dilog (double x)
{
  ASSERT(x <= -1.0); /* dilog(1-x) is called from rhoexact for 2 < x <= 3 */

  if (x <= -2.0)
    return -dilog_series (1./x) - M_PI_SQR_6 - 0.5 * log(-1./x) * log(-1./x);
  else /* x <= -1.0 */
    {
      /* L2(z) = -L2(1 - z) + 1/6 * Pi^2 - ln(1 - z)*ln(z) 
         L2(z) = -L2(1/z) - 1/6 * Pi^2 - 0.5*ln^2(-1/z)
         ->
         L2(z) = -(-L2(1/(1-z)) - 1/6 * Pi^2 - 0.5*ln^2(-1/(1-z))) + 1/6 * Pi^2 - ln(1 - z)*ln(z)
               = L2(1/(1-z)) - 1/6 * Pi^2 + 0.5*ln(1 - z)^2 - ln(1 - z)*ln(-z)
         z in [-1, -2) -> 1/(1-z) in [1/2, 1/3)
      */
      double log1x = log (1. - x);
      return dilog_series (1. / (1. - x)) 
             - M_PI_SQR_6 + log1x * (0.5 * log1x - log (-x));
    }
}

#if 0
static double 
L2 (double x)
{
  return log (x) * (1 - log (x-1)) + M_PI_SQR_6 - dilog (1 - x);
}
#endif

static double
rhoexact (double x)
{
  ASSERT(x <= 3.);
  if (x <= 0.)
    return 0.;
  else if (x <= 1.)
    return 1.;
  else if (x <= 2.)
    return 1. - log (x);
  else /* 2 < x <= 3 thus -2 <= 1-x < -1 */
    return 1. - log (x) * (1. - log (x - 1.)) + dilog (1. - x) + 0.5 * M_PI_SQR_6;
}


#if defined(TESTDRIVE)

/* The Buchstab omega(x) function, exact for x <= 4 where it can be 
   evaluated without numerical integration, and approximated by 
   exp(gamma) for larger x. */

static double
Buchstab_omega (const double x)
{
  /* magic = dilog(-1) + 1  = Pi^2/12 + 1 */
  const double magic = 1.82246703342411321824; 

  if (x < 1.) return (0.);
  if (x <= 2.) return (1. / x);
  if (x <= 3.) return ((log (x - 1.) + 1.) / x);
  if (x <= 4.)
    return ((dilog(2. - x) + (1. + log(x - 2.)) * log(x - 1.) + magic) / x);

  /* If argument is out of range, return the limiting value for 
     $x->\infty$: e^-gamma. 
     For x only a little larger than 4., this has relative error 2.2e-6,
     for larger x the error rapidly drops further */

  return 0.56145948356688516982;
}

#endif

void 
rhoinit (int parm_invh, int parm_tablemax)
{
  int i;

  if (parm_invh == invh && parm_tablemax == tablemax)
    return;

  if (rhotable != NULL)
    {
      free (rhotable);
      rhotable = NULL;
      invh = 0;
      h = 0.;
      tablemax = 0;
    }
  
  /* The integration below expects 3 * invh > 4 */
  if (parm_tablemax == 0 || parm_invh < 2)
    return;
    
  invh = parm_invh;
  h = 1. / (double) invh;
  tablemax = parm_tablemax;
  
  rhotable = (double *) malloc (parm_invh * parm_tablemax * sizeof (double));
  ASSERT_ALWAYS(rhotable != NULL);
  
  for (i = 0; i < (3 < parm_tablemax ? 3 : parm_tablemax) * invh; i++)
    rhotable[i] = rhoexact (i * h);
  
  for (i = 3 * invh; i < parm_tablemax * invh; i++)
    {
      /* rho(i*h) = 1 - \int_{1}^{i*h} rho(x-1)/x dx
                  = rho((i-4)*h) - \int_{(i-4)*h}^{i*h} rho(x-1)/x dx */
      
      rhotable[i] = rhotable[i - 4] - 2. / 45. * (
          7. * rhotable[i - invh - 4] / (double)(i - 4)
        + 32. * rhotable[i - invh - 3] / (double)(i - 3)
        + 12. * rhotable[i - invh - 2] / (double)(i - 2)
        + 32. * rhotable[i - invh - 1] / (double)(i - 1)
        + 7. * rhotable[i - invh]  / (double)i );
      if (rhotable[i] < 0.)
        {
#ifndef DEBUG_NUMINTEGRATE
          rhotable[i] = 0.;
#else
          printf (stderr, "rhoinit: rhotable[%d] = %.16f\n", i, 
                   rhotable[i]);
          exit (EXIT_FAILURE);
#endif
        }
    }
}

/* assumes alpha < tablemax */
static double
dickmanrho (double alpha)
{
  ASSERT(alpha < tablemax);

  if (alpha <= 3.)
     return rhoexact (alpha);
  {
    int a = floor (alpha * invh);
    double rho1 = rhotable[a];
    double rho2 = (a + 1) < tablemax * invh ? rhotable[a + 1] : 0;
    return rho1 + (rho2 - rho1) * (alpha * invh - (double) a);
  }
}

#if 0
static double 
dickmanrhosigma (double alpha, double x)
{
  if (alpha <= 0.)
    return 0.;
  if (alpha <= 1.)
    return 1.;
  if (alpha < tablemax)
    return dickmanrho (alpha) + M_EULER_1 * dickmanrho (alpha - 1.) / log (x);
  
  return 0.;
}

static double
dickmanrhosigma_i (int ai, double x)
{
  if (ai <= 0)
    return 0.;
  if (ai <= invh)
    return 1.;
  if (ai < tablemax * invh)
    return rhotable[ai] - M_EULER * rhotable[ai - invh] / log(x);
  
  return 0.;
}
#endif

/* return the value of the "local" Dickman rho function, for numbers near x
   (as opposed to numbers <= x for the original Dickman rho function).
   Reference: PhD thesis of Alexander Kruppa,
   http://docnum.univ-lorraine.fr/public/SCD_T_2010_0054_KRUPPA.pdf,
   equation (5.6) page 100 */
static double
dickmanlocal (double alpha, double x)
{
  if (alpha <= 1.)
    return rhoexact (alpha);
  if (alpha < tablemax)
    return dickmanrho (alpha) - M_EULER * dickmanrho (alpha - 1.) / log (x);
  return 0.;
}

static double
dickmanlocal_i (int ai, double x)
{
  if (ai <= 0)
    return 0.;
  if (ai <= invh)
    return 1.;
  if (ai <= 2 * invh && ai < tablemax * invh)
    return rhotable[ai] - M_EULER / log (x);
  if (ai < tablemax * invh)
    {
      double logx = log (x);
      return rhotable[ai] - (M_EULER * rhotable[ai - invh]
             + M_EULER_1 * rhotable[ai - 2 * invh] / logx) / logx;
    }

  return 0.;
}

static int 
isprime(unsigned long n)
{
  unsigned int r;

  if (n % 2 == 0)
    return (n == 2);
  if (n % 3 == 0)
    return (n == 3);
  if (n % 5 == 0)
    return (n == 5);

  if (n / 30 >= sizeof (primemap))
    abort();
  
  r = n % 30; /* 8 possible values: 1,7,11,13,17,19,23,29 */
  r = (r * 16 + r) / 64; /* maps the 8 values onto 0, ..., 7 */

  return ((primemap[n / 30] & (1 << r)) != 0);
}

/* return the sum in Equation (5.10) page 102 of Alexander Kruppa's
   PhD thesis */
static double
dickmanmu_sum (const unsigned long B1, const unsigned long B2, 
	       const double x)
{
  double s = 0.;
  const double inv_logB1 = 1. / log(B1);
  const double logx = log(x); 
  unsigned long p;

  for (p = B1 + 1; p <= B2; p++)
    if (isprime(p))
      s += dickmanlocal ((logx - log(p)) * inv_logB1, x / p) / p;

  return s;
}

/* return the probability that a number < x has its 2nd largest prime factor
   less than x^(1/alpha) and its largest prime factor less than x^(beta/alpha)
*/
static double
dickmanmu (double alpha, double beta, double x)
{
  double a, b, sum;
  int ai, bi, i;
  ai = ceil ((alpha - beta) * invh);
  if (ai > tablemax * invh)
    ai = tablemax * invh;
  a = (double) ai * h;
  bi = floor ((alpha - 1.) * invh);
  if (bi > tablemax * invh)
    bi = tablemax * invh;
  b = (double) bi * h;
  sum = 0.;
  for (i = ai + 1; i < bi; i++)
    sum += dickmanlocal_i (i, x) / (alpha - i * h);
  sum += 0.5 * dickmanlocal_i (ai, x) / (alpha - a);
  sum += 0.5 * dickmanlocal_i (bi, x) / (alpha - b);
  sum *= h;
  sum += (a - alpha + beta) * 0.5 * (dickmanlocal_i (ai, x) / (alpha - a) + dickmanlocal (alpha - beta, x) / beta);
  sum += (alpha - 1. - b) * 0.5 * (dickmanlocal (alpha - 1., x) + dickmanlocal_i (bi, x) / (alpha - b));

  return sum;
}

static double
brentsuyama (double B1, double B2, double N, double nr)
{
  double a, alpha, beta, sum;
  int ai, i;
  alpha = log (N) / log (B1);
  beta = log (B2) / log (B1);
  ai = floor ((alpha - beta) * invh);
  if (ai > tablemax * invh)
    ai = tablemax * invh;
  a = (double) ai * h;
   sum = 0.;
  for (i = 1; i < ai; i++)
    sum += dickmanlocal_i (i, N) / (alpha - i * h) * (1 - exp (-nr * pow (B1, (-alpha + i * h))));
  sum += 0.5 * (1 - exp(-nr / pow (B1, alpha)));
  sum += 0.5 * dickmanlocal_i (ai, N) / (alpha - a) * (1 - exp(-nr * pow (B1, (-alpha + a))));
  sum *= h;
  sum += 0.5 * (alpha - beta - a) * (dickmanlocal_i (ai, N) / (alpha - a) + dickmanlocal (alpha - beta, N) / beta);

  return sum;
}

static double 
brsudickson (double B1, double B2, double N, double nr, int S)
{
  int i, f;
  double sum;
  sum = 0;
  f = eulerphi (S) / 2;
  for (i = 1; i <= S / 2; i++)
      if (gcd (i, S) == 1)
        sum += brentsuyama (B1, B2, N, nr * (gcd (i - 1, S) + gcd (i + 1, S) - 4) / 2);
  
  return sum / (double)f;
}

static double
brsupower (double B1, double B2, double N, double nr, int S)
{
  int i, f;
  double sum;
  sum = 0;
  f = eulerphi (S);
  for (i = 1; i < S; i++)
      if (gcd (i, S) == 1)
        sum += brentsuyama (B1, B2, N, nr * (gcd (i - 1, S) - 2));
  
  return sum / (double)f;
}

/* Assume N is as likely smooth as a number around N/exp(delta) */

static double
prob (double B1, double B2, double N, double nr, int S, double delta)
{
  const double sumthresh = 20000.;
  double alpha, beta, stage1, stage2, brsu;
  const double effN = N / exp (delta);

  ASSERT(rhotable != NULL);
  
  /* What to do if rhotable is not initialised and asserting is not enabled?
     For now, bail out with 0. result. Not really pretty, either */
  if (rhotable == NULL)
    return 0.;

  if (B1 < 2. || N <= 1.)
    return 0.;
  
  if (effN <= B1)
    return 1.;

#ifdef TESTDRIVE
  printf ("B1 = %f, B2 = %f, N = %.0f, nr = %f, S = %d\n", B1, B2, N, nr, S);
#endif
  
  alpha = log (effN) / log (B1);
  stage1 = dickmanlocal (alpha, effN);
  stage2 = 0.;
  if (B2 > B1)
    {
      if (B1 < sumthresh)
	{
	  stage2 += dickmanmu_sum (B1, MIN(B2, sumthresh), effN);
	  beta = log (B2) / log (MIN(B2, sumthresh));
	}
      else
	beta = log (B2) / log (B1);

      if (beta > 1.)
	stage2 += dickmanmu (alpha, beta, effN);
    }
  brsu = 0.;
  if (S < -1)
    brsu = brsudickson (B1, B2, effN, nr, -S * 2);
  if (S > 1)
    brsu = brsupower (B1, B2, effN, nr, S * 2);

#ifdef TESTDRIVE
  printf ("stage 1 : %f, stage 2 : %f, Brent-Suyama : %f\n", stage1, stage2, brsu);
#endif

  return (stage1 + stage2 + brsu) > 0. ? (stage1 + stage2 + brsu) : 0.;
}

double
ecmprob (double B1, double B2, double N, double nr, int S)
{
  return prob (B1, B2, N, nr, S, ECM_EXTRA_SMOOTHNESS);
}

/* see Willemien Ekkelkamp's Phd thesis:
   https://openaccess.leidenuniv.nl/bitstream/handle/1887/14567/proefschrift_041109.pdf?sequence=2:
   Bach-Peralta formula page 12
   Corollary 4 page 18 with a 2nd-order term */
double
pm1prob (double B1, double B2, double N, double nr, int S, const mpz_t go)
{
  mpz_t cof;
  /* A prime power q^k divides p-1, p prime, with probability 1/(q^k-q^(k-1))
     not with probability 1/q^k as for random numbers. This is taken into 
     account by the "smoothness" value here; a prime p-1 is about as likely
     smooth as a random number around (p-1)/exp(smoothness).
     smoothness = \sum_{q in Primes} log(q)/(q-1)^2 */
  /* Note that this routine is also called for P+1, where we assume the same
     behaviour as with P-1. However, if x0=6/5, Kruppa writes in his PhD
     thesis that we get smoothness = 1.92012; with x0=2/7, we get
     smoothness = 2.05093. */
  double smoothness = 1.2269688;
  unsigned long i;
  
  if (go != NULL && mpz_cmp_ui (go, 1UL) > 0)
    {
      double res;
      mpz_init (cof);
      mpz_set (cof, go);
      for (i = 2; i < 100; i++)
        if (mpz_divisible_ui_p (cof, i))
          {
            /* If we know that q divides p-1 with probability 1, we need to
               adjust the smoothness parameter */
            smoothness -= log ((double) i) / (double) ((i-1)*(i-1));
            /* printf ("pm1prob: Dividing out %lu\n", i); */
            while (mpz_divisible_ui_p (cof, i))
              mpz_tdiv_q_ui (cof, cof, i);
          }
      /* printf ("pm1prob: smoothness after dividing out go primes < 100: %f\n", 
               smoothness); */
      res = prob (B1, B2, N, nr, S, smoothness + log(mpz_get_d (cof)));
      mpz_clear (cof);
      return res;
    }

  return prob (B1, B2, N, nr, S, smoothness);
}

#if defined(TESTDRIVE)

/* Compute probability for primes p == r (mod m) */

static double
pm1prob_rm (double B1, double B2, double N, double nr, int S, unsigned long r,
            unsigned long m)
{
  unsigned long cof;
  double smoothness = 1.2269688;
  unsigned long p;
  
  cof = m;
  
  for (p = 2UL; p < 100UL; p++)
    if (cof % p == 0UL) /* For each prime in m */
      {
        unsigned long cof_r, k, i;
        /* Divisibility by i is determined by r and m. We need to
           adjust the smoothness parameter. In P-1, we had estimated the 
           expected value for the exponent of p as p/(p-1)^2. Undo that. */
        smoothness -= (double)p / ((p-1)*(p-1)) * log ((double) p);
        /* The expected value for the exponent of this prime is k s.t.
           p^k || r, plus 1/(p-1) if p^k || m as well */
        cof_r = gcd (r - 1UL, m);
        for (k = 0UL; cof_r % p == 0UL; k++)
          cof_r /= p;
        smoothness += k * log ((double) p);

        cof_r = m;
        for (i = 0UL; cof_r % p == 0UL; i++)
          cof_r /= p;

        if (i == k)
          smoothness += (1./(p - 1.) * log ((double) p));
        
        while (cof % p == 0UL)
          cof /= p;
        printf ("pm1prob_rm: p = %lu, k = %lu, i = %lu, new smoothness = %f\n", 
                p, i, k, smoothness); 
      }

  return prob (B1, B2, N, nr, S, smoothness);
}


/* The \Phi(x,y) function gives the number of natural numbers <= x 
   that have no prime factor <= y, see Tenenbaum, 
   "Introduction the analytical and probabilistic number theory", III.6.
   This function estimates the \Phi(x,y) function via eq. (48) of the 1st
   edition resp. equation (6.49) of the 3rd edition of Tenenbaum's book. */

static double 
integrand1 (double x, double *y)
{
  return pow (*y, x) / x * log(x-1.);
}


static double 
integrand2 (double v, double *y)
{
  return Buchstab_omega (v) * pow (*y, v);
}


/* Return approximate number of integers n with x1 < n <= x2
   that have no prime factor <= y */

double 
no_small_prime (double x1, double x2, double y)
{
  double u1, u2;
  ASSERT (x1 >= 2.);
  ASSERT (x2 >= x1);
  ASSERT (y >= 2.);
  if (x1 == x2 || x2 <= y)
    return 0.;
  if (x1 < y)
    x1 = y;
  
  u1 = log(x1)/log(y);
  u2 = log(x2)/log(y);

   /* If no prime factors <= sqrt(x2), numbers must be a primes > y */
  if (x2 <= y*y)
    return (Li(x2) - Li(x1));
  
  if (u2 <= 3)
    {
      double r, abserr;
      size_t neval;
      gsl_function f;

      f.function = (double (*) (double, void *)) &integrand1;
      f.params = &y;

      /* intnum(v=1,u,buchstab(v)*y^v) */

      /* First part: intnum(v=u1, u, y^v/v*log(v-1.)) */
      gsl_integration_qng (&f, MAX(u1, 2.) , u2, 0., 0.001, &r, &abserr, &neval);

      /* Second part: intnum(v=u1, u2, y^v/v) = Li(x2) - Li(x1) */
      r += Li (x2) - Li (x1);
      
      return r;
    }
    
  {
    double r, abserr;
    size_t neval;
    gsl_function f;
  
    f.function = (double (*) (double, void *)) &integrand2;
    f.params = &y;
    
    gsl_integration_qng (&f, u1, u2, 0., 0.001, &r, &abserr, &neval);
    return r;
  }
}


static double 
integrand3 (double p, double *param)
{
  const double x1 = param[0];
  const double x2 = param[1];
  const double y = param[2];
  
  return no_small_prime (x1 / p, x2 / p, y) / log(p);
}


double 
no_small_prime_factor (const double x1, const double x2, const double y, 
                       const double z1, const double z2)
{
  double r, abserr, param[3];
  size_t neval;
  gsl_function f;

  param[0] = x1;
  param[1] = x2;
  param[2] = y;
  f.function = (double (*) (double, void *)) &integrand3;
  f.params = &param;
  
  gsl_integration_qng (&f, z1, z2, 0., 0.01, &r, &abserr, &neval);
  
  return r;
}

#endif


#ifdef TESTDRIVE
int
main (int argc, char **argv)
{
  double B1, B2, N, nr, r, m;
  int S;
  unsigned long p, i, pi;
  primesieve_iterator pg[1];

  primesieve_init (pg);
  i = pi = 0;
  for (p = primesieve_next_prime (pg); p <= PRIME_PI_MAX; p = primesieve_next_prime (pg))
    {
      for ( ; i < p; i++)
        prime_pi[PRIME_PI_MAP(i)] = pi;
      pi++;
    }
  for ( ; i < p; i++)
    prime_pi[PRIME_PI_MAP(i)] = pi;
  

  if (argc < 2)
    {
      printf ("Usage: rho <B1> <B2> <N> <nr> <S> [<r> <m>]\n\n\n");
      printf (" Calculate the probability of ECM/P-1 finding a factor near N\n"
              " with B1/B2, evaluating nr random distinct points in stage 2,\n"
              " with a degree -S Dickson polynomial (if S < 0) or\n"
              " S'th power as the Brent-Suyama function\n\n");
      printf (" <B1>        B1 limit.\n");
      printf (" <B2>        B2 limit.\n");
      printf (" <N>         N of similiar size, or number of bits in factor (if < 50).\n");
      printf (" <nr>        Number of random points evaluated in stage 2.\n");
      printf (" <S>         Degree of Brent-Suyama polynomial in stage 2.\n");
      printf (" [<r> <m>]   Limit P-1 to primes p == r (mod m).\n");
      return 1;
    }
  
  if (strcmp (argv[1], "-Buchstab_Phi") == 0)
    {
      unsigned long x, y, r;
      if (argc < 4)
        {
          printf ("-Buchstab_Phi needs x and y paramters\n");
          exit (EXIT_FAILURE);
        }
      x = strtoul (argv[2], NULL, 10);
      y = strtoul (argv[3], NULL, 10);
      r = Buchstab_Phi (x, y);
      printf ("Buchstab_Phi (%lu, %lu) = %lu\n", x, y, r);
      exit (EXIT_SUCCESS);
    }
  else if (strcmp (argv[1], "-Buchstab_Psi") == 0)
    {
      unsigned long x, y, r;
      if (argc < 4)
        {
          printf ("-Buchstab_Psi needs x and y paramters\n");
          exit (EXIT_FAILURE);
        }
      x = strtoul (argv[2], NULL, 10);
      y = strtoul (argv[3], NULL, 10);
      r = Buchstab_Psi (x, y);
      printf ("Buchstab_Psi (%lu, %lu) = %lu\n", x, y, r);
      exit (EXIT_SUCCESS);
    }
  else if (strcmp (argv[1], "-nsp") == 0)
    {
      double x1, x2, y, r;
      
      if (argc < 5)
        {
          printf ("-nsp needs x1, x2, and y paramters\n");
          exit (EXIT_FAILURE);
        }
      x1 = atof (argv[2]);
      x2 = atof (argv[3]);
      y = atof (argv[4]);
      r = no_small_prime (x1, x2, y);
      printf ("no_small_prime(%f, %f, %f) = %f\n", x1, x2, y, r);
      exit (EXIT_SUCCESS);
    }
  else if (strcmp (argv[1], "-nspf") == 0)
    {
      double x1, x2, y, z1, z2, r;
      
      if (argc < 7)
        {
          printf ("-nspf needs x1, x2, y, z1, and z2 paramters\n");
          exit (EXIT_FAILURE);
        }
      x1 = atof (argv[2]);
      x2 = atof (argv[3]);
      y = atof (argv[4]);
      z1 = atof (argv[5]);
      z2 = atof (argv[6]);
      r = no_small_prime_factor (x1, x2, y, z1, z2);
      printf ("no_small_prime(%f, %f, %f, %f, %f) = %f\n", x1, x2, y, z1, z2, r);
      exit (EXIT_SUCCESS);
    }


  if (argc < 6)
    {
      printf ("Need 5 or 7 arguments: B1 B2 N nr S [r m]\n");
      exit (EXIT_FAILURE);
    }
  
  B1 = atof (argv[1]);
  B2 = atof (argv[2]);
  N = atof (argv[3]);
  nr = atof (argv[4]);
  S = atoi (argv[5]);
  r = 0; m = 1;
  if (argc > 7)
    {
      r = atoi (argv[6]);
      m = atoi (argv[7]);
    }

  rhoinit (256, 10);
  if (N < 50.)
    {
      double sum;
      sum = ecmprob(B1, B2, exp2 (N), nr, S);
      sum += 4. * ecmprob(B1, B2, 3./2. * exp2 (N), nr, S);
      sum += ecmprob(B1, B2, 2. * exp2 (N), nr, S);
      sum *= 1./6.;
      printf ("ECM: %.16f\n", sum);

      sum = pm1prob_rm (B1, B2, exp2 (N), nr, S, r, m);
      sum += 4. * pm1prob_rm (B1, B2, 3./2. * exp2 (N), nr, S, r, m);
      sum += pm1prob_rm (B1, B2, 2. * exp2 (N), nr, S, r, m);
      sum *= 1./6.;
      printf ("P-1: %.16f\n", sum);
    }
  else
    {
      printf ("ECM: %.16f\n", ecmprob(B1, B2, N, nr, S));
      printf ("P-1: %.16f\n", pm1prob_rm (B1, B2, N, nr, S, r, m));
    }
  rhoinit (0, 0);
  return 0;
}
#endif