OpenCores
URL https://opencores.org/ocsvn/thor/thor/trunk

Subversion Repositories thor

[/] [thor/] [trunk/] [FT64v5/] [software/] [CC64/] [source/] [rand.cpp] - Rev 48

Compare with Previous | Blame | View Log

/********************************************************************** 
 Freeciv - Copyright (C) 1996 - A Kjeldberg, L Gregersen, P Unold
   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2, or (at your option)
   any later version.
 
   This program 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 General Public License for more details.
***********************************************************************/
 
/*************************************************************************
   The following random number generator can be found in _The Art of 
   Computer Programming Vol 2._ (2nd ed) by Donald E. Knuth. (C)  1998.
   The algorithm is described in section 3.2.2 as Mitchell and Moore's
   variant of a standard additive number generator.  Note that the
   the constants 55 and 24 are not random.  Please become familiar with
   this algorithm before you mess with it.
 
   Since the additive number generator requires a table of numbers from
   which to generate its random sequences, we must invent a way to 
   populate that table from a single seed value.  I have chosen to do
   this with a different PRNG, known as the "linear congruential method" 
   (also found in Knuth, Vol2).  I must admit that my choices of constants
   (3, 257, and MAX_UINT32) are probably not optimal, but they seem to
   work well enough for our purposes.
 
   Original author for this code: Cedric Tefft <cedric@earthling.net>
   Modified to use rand_state struct by David Pfitzner <dwp@mso.anu.edu.au>
*************************************************************************/
 
#include "stdafx.h"
 
RANDOM_STATE RTFClasses::Random::rand_state;
 
/*************************************************************************
  Returns a new random value from the sequence, in the interval 0 to
  (size-1) inclusive, and updates global state for next call.
  This means that if size <= 1 the function will always return 0.
 
  Once we calculate new_rand below uniform (we hope) between 0 and
  MAX_UINT32 inclusive, need to reduce to required range.  Using
  modulus is bad because generators like this are generally less
  random for their low-significance bits, so this can give poor
  results when 'size' is small.  Instead want to divide the range
  0..MAX_UINT32 into (size) blocks, each with (divisor) values, and
  for any remainder, repeat the calculation of new_rand.
  Then:
	 return_val = new_rand / divisor;
  Will repeat for new_rand > max, where:
         max = size * divisor - 1
  Then max <= MAX_UINT32 implies
	 size * divisor <= (MAX_UINT32+1)
  thus   divisor <= (MAX_UINT32+1)/size
 
  Need to calculate this divisor.  Want divisor as large as possible
  given above contraint, but it doesn't hurt us too much if it is a
  bit smaller (just have to repeat more often).  Calculation exactly
  as above is complicated by fact that (MAX_UINT32+1) may not be
  directly representable in type RANDOM_TYPE, so we do instead:
         divisor = MAX_UINT32/size
*************************************************************************/
namespace RTFClasses
{
RANDOM_TYPE Random::rand(RANDOM_TYPE size)
{
  RANDOM_TYPE new_rand, divisor, max;
  int bailout = 0;
 
	if (size > 1) {
		divisor = MAX_UINT32 / size;
		max = size * divisor - 1;
	}
	else {
		/* size == 0 || size == 1 */
 
		/* 
		* These assignments are only here to make the compiler
		* happy. Since each usage is guarded with a if(size>1).
		*/
		max = MAX_UINT32;
		divisor = 1;
	}
 
	do {
		new_rand = (rand_state.v[rand_state.j]
			+ rand_state.v[rand_state.k]) & MAX_UINT32;
 
		rand_state.x = (rand_state.x +1) % 56;
		rand_state.j = (rand_state.j +1) % 56;
		rand_state.k = (rand_state.k +1) % 56;
		rand_state.v[rand_state.x] = new_rand;
 
		if (++bailout > 10000) {
			//      freelog(LOG_ERROR, "Bailout in myrand(%u)", size);
			new_rand = 0;
			break;
		}
 
	}
	while (size > 1 && new_rand > max);
 
	if (size > 1)
		new_rand /= divisor;
	else
		new_rand = 0;
 
	/* freelog(LOG_DEBUG, "rand(%u) = %u", size, new_rand); */
 
	return new_rand;
} 
 
/*************************************************************************
  Initialize the generator; see comment at top of file.
*************************************************************************/
void Random::srand(RANDOM_TYPE seed)
{ 
    int  i; 
 
    rand_state.v[0]=(seed & MAX_UINT32);
 
    for(i=1; i<56; i++)
       rand_state.v[i] = (3 * rand_state.v[i-1] + 257) & MAX_UINT32;
 
    rand_state.j = (55-55);
    rand_state.k = (55-24);
    rand_state.x = (55-0);
 
    rand_state.is_init = true;
 
    /* Heat it up a bit:
     * Using modulus in myrand() this was important to pass
     * test_random1().  Now using divisor in myrand() that particular
     * test no longer indicates problems, but this seems a good idea
     * anyway -- eg, other tests could well reveal other initial
     * problems even using divisor.
     */
	for (i=0; i<10000; i++)
		(void) rand(MAX_UINT32);
} 
 
/*************************************************************************
  Test one aspect of randomness, using n numbers.
  Reports results to LOG_NORMAL; with good randomness, behaviourchange
  and behavioursame should be about the same size.
  Tests current random state; saves and restores state, so can call
  without interrupting current sequence.
*************************************************************************/
void Random::test(int n)
{
  RANDOM_STATE saved_state;
  int i, old_value = 0, new_value;
  bool didchange, olddidchange = false;
  int behaviourchange = 0, behavioursame = 0;
 
  saved_state = getRandState();
  /* mysrand(time(NULL)); */  /* use current state */
 
  for (i = 0; i < n+2; i++) {
	new_value = rand(2);
	if (i > 0) {		/* have old */
		didchange = (new_value != old_value);
		if (i > 1) {		/* have olddidchange */
			if (didchange != olddidchange)
				behaviourchange++;
			else
				behavioursame++;
		}
		olddidchange = didchange;
	}
	old_value = new_value;
  }
 
  /* restore state: */
  setRandState(saved_state);
};
 
}
 

Compare with Previous | Blame | View Log

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.