server/include/util.h

/*
 * This file is part of Deliantra, the Roguelike Realtime MMORPG.
 * 
 * Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team
 * 
 * Deliantra is free software: you can redistribute it and/or modify it under
 * the terms of the Affero GNU General Public License as published by the
 * Free Software Foundation, either version 3 of the License, 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.
 * 
 * You should have received a copy of the Affero GNU General Public License
 * and the GNU General Public License along with this program. If not, see
 * <http://www.gnu.org/licenses/>.
 * 
 * The authors can be reached via e-mail to <support@deliantra.net>
 */

#ifndef UTIL_H__
#define UTIL_H__

#include <compiler.h>

#define DEBUG_POISON 0x00 // poison memory before freeing it if != 0
#define DEBUG_SALLOC  0   // add a debug wrapper around all sallocs
#define PREFER_MALLOC 0   // use malloc and not the slice allocator

#include <pthread.h>

#include <cstddef>
#include <cmath>
#include <new>
#include <vector>

#include <glib.h>

#include <shstr.h>
#include <traits.h>

#if DEBUG_SALLOC
# define g_slice_alloc0(s) debug_slice_alloc0(s)
# define g_slice_alloc(s) debug_slice_alloc(s)
# define g_slice_free1(s,p) debug_slice_free1(s,p)
void *g_slice_alloc (unsigned long size);
void *g_slice_alloc0 (unsigned long size);
void g_slice_free1 (unsigned long size, void *ptr);
#elif PREFER_MALLOC
# define g_slice_alloc0(s) calloc (1, (s))
# define g_slice_alloc(s) malloc ((s))
# define g_slice_free1(s,p) free ((p))
#endif

// use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever)
#define auto(var,expr) decltype(expr) var = (expr)

// very ugly macro that basically declares and initialises a variable
// that is in scope for the next statement only
// works only for stuff that can be assigned 0 and converts to false
// (note: works great for pointers)
// most ugly macro I ever wrote
#define statementvar(type, name, value) if (type name = 0) { } else if (((name) = (value)), 1)

// in range including end
#define IN_RANGE_INC(val,beg,end) \
  ((unsigned int)(val) - (unsigned int)(beg) <= (unsigned int)(end) - (unsigned int)(beg))

// in range excluding end
#define IN_RANGE_EXC(val,beg,end) \
  ((unsigned int)(val) - (unsigned int)(beg) <  (unsigned int)(end) - (unsigned int)(beg))

void cleanup (const char *cause, bool make_core = false);
void fork_abort (const char *msg);

// rationale for using (U) not (T) is to reduce signed/unsigned issues,
// as a is often a constant while b is the variable. it is still a bug, though.
template<typename T, typename U> static inline T min (T a, U b) { return (U)a < b ? (U)a : b; }
template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; }
template<typename T, typename U, typename V> static inline T clamp (T v, U a, V b) { return v < (T)a ? (T)a : v >(T)b ? (T)b : v; }

template<typename T, typename U> static inline void min_it (T &v, U m) { v = min (v, (T)m); }
template<typename T, typename U> static inline void max_it (T &v, U m) { v = max (v, (T)m); }
template<typename T, typename U, typename V> static inline void clamp_it (T &v, U a, V b) { v = clamp (v, (T)a, (T)b); }

template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; }

template<typename T, typename U, typename V> static inline T min (T a, U b, V c) { return min (a, min (b, c)); }
template<typename T, typename U, typename V> static inline T max (T a, U b, V c) { return max (a, max (b, c)); }

// sign returns -1 or +1
template<typename T>
static inline T sign (T v) { return v < 0 ? -1 : +1; }
// relies on 2c representation
template<>
inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); }

// sign0 returns -1, 0 or +1
template<typename T>
static inline T sign0 (T v) { return v ? sign (v) : 0; }

// div* only work correctly for div > 0
// div, with correct rounding (< 0.5 downwards, >=0.5 upwards)
template<typename T> static inline T div    (T val, T div)
{
  return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div;
}
// div, round-up
template<typename T> static inline T div_ru (T val, T div)
{
  return expect_false (val < 0) ? - ((-val                ) / div) : (val + div - 1) / div;
}
// div, round-down
template<typename T> static inline T div_rd (T val, T div)
{
  return expect_false (val < 0) ? - ((-val + (div - 1)    ) / div) : (val          ) / div;
}

// lerp* only work correctly for min_in < max_in
// Linear intERPolate, scales val from min_in..max_in to min_out..max_out
template<typename T>
static inline T
lerp (T val, T min_in, T max_in, T min_out, T max_out)
{
  return min_out + div   <T> ((val - min_in) * (max_out - min_out), max_in  - min_in);
}

// lerp, round-down
template<typename T>
static inline T
lerp_rd (T val, T min_in, T max_in, T min_out, T max_out)
{
  return min_out + div_rd<T> ((val - min_in) * (max_out - min_out), max_in  - min_in);
}

// lerp, round-up
template<typename T>
static inline T
lerp_ru (T val, T min_in, T max_in, T min_out, T max_out)
{
  return min_out + div_ru<T> ((val - min_in) * (max_out - min_out), max_in  - min_in);
}

// lots of stuff taken from FXT

/* Rotate right. This is used in various places for checksumming */
//TODO: that sucks, use a better checksum algo
static inline uint32_t
rotate_right (uint32_t c, uint32_t count = 1)
{
  return (c << (32 - count)) | (c >> count);
}

static inline uint32_t
rotate_left (uint32_t c, uint32_t count = 1)
{
  return (c >> (32 - count)) | (c << count);
}

// Return abs(a-b)
// Both a and b must not have the most significant bit set
static inline uint32_t
upos_abs_diff (uint32_t a, uint32_t b)
{
  long d1 = b - a;
  long d2 = (d1 & (d1 >> 31)) << 1;

  return d1 - d2;               // == (b - d) - (a + d);
}

// Both a and b must not have the most significant bit set
static inline uint32_t
upos_min (uint32_t a, uint32_t b)
{
  int32_t d = b - a;
  d &= d >> 31;
  return a + d;
}

// Both a and b must not have the most significant bit set
static inline uint32_t
upos_max (uint32_t a, uint32_t b)
{
  int32_t d = b - a;
  d &= d >> 31;
  return b - d;
}

// this is much faster than crossfire's original algorithm
// on modern cpus
inline int
isqrt (int n)
{
  return (int)sqrtf ((float)n);
}

// this is kind of like the ^^ operator, if it would exist, without sequence point.
// more handy than it looks like, due to the implicit !! done on its arguments
inline bool
logical_xor (bool a, bool b)
{
  return a != b;
}

inline bool
logical_implies (bool a, bool b)
{
  return a <= b;
}

// this is only twice as fast as naive sqrtf (dx*dy+dy*dy)
#if 0
// and has a max. error of 6 in the range -100..+100.
#else
// and has a max. error of 9 in the range -100..+100.
#endif
inline int 
idistance (int dx, int dy)
{ 
  unsigned int dx_ = abs (dx);
  unsigned int dy_ = abs (dy);

#if 0
  return dx_ > dy_
      ? (dx_ * 61685 + dy_ * 26870) >> 16
      : (dy_ * 61685 + dx_ * 26870) >> 16;
#else
  return dx_ + dy_ - min (dx_, dy_) * 5 / 8;
#endif
}

/*
 * absdir(int): Returns a number between 1 and 8, which represent
 * the "absolute" direction of a number (it actually takes care of
 * "overflow" in previous calculations of a direction).
 */
inline int
absdir (int d)
{
  return ((d - 1) & 7) + 1;
}

extern ssize_t slice_alloc; // statistics

void *salloc_ (int n)            throw (std::bad_alloc);
void *salloc_ (int n, void *src) throw (std::bad_alloc);

// strictly the same as g_slice_alloc, but never returns 0
template<typename T>
inline T *salloc (int n = 1)     throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T));              }

// also copies src into the new area, like "memdup"
// if src is 0, clears the memory
template<typename T>
inline T *salloc (int n, T *src) throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T), (void *)src); }

// clears the memory
template<typename T>
inline T *salloc0(int n = 1)     throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T), 0);           }

// for symmetry
template<typename T>
inline void sfree (T *ptr, int n = 1) throw ()
{
  if (expect_true (ptr))
    {
      slice_alloc -= n * sizeof (T);
      if (DEBUG_POISON) memset (ptr, DEBUG_POISON, n * sizeof (T));
      g_slice_free1 (n * sizeof (T), (void *)ptr);
      assert (slice_alloc >= 0);//D
    }
}

// nulls the pointer
template<typename T>
inline void sfree0 (T *&ptr, int n = 1) throw ()
{
  sfree<T> (ptr, n);
  ptr = 0;
}

// makes dynamically allocated objects zero-initialised
struct zero_initialised
{
  void *operator new (size_t s, void *p)
  {
    memset (p, 0, s);
    return p;
  }

  void *operator new (size_t s)
  {
    return salloc0<char> (s);
  }

  void *operator new[] (size_t s)
  {
    return salloc0<char> (s);
  }

  void operator delete (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }

  void operator delete[] (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }
};

// makes dynamically allocated objects zero-initialised
struct slice_allocated
{
  void *operator new (size_t s, void *p)
  {
    return p;
  }

  void *operator new (size_t s)
  {
    return salloc<char> (s);
  }

  void *operator new[] (size_t s)
  {
    return salloc<char> (s);
  }

  void operator delete (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }

  void operator delete[] (void *p, size_t s)
  {
    sfree ((char *)p, s);
  }
};

// a STL-compatible allocator that uses g_slice
// boy, this is verbose
template<typename Tp>
struct slice_allocator
{
  typedef size_t size_type;
  typedef ptrdiff_t difference_type;
  typedef Tp *pointer;
  typedef const Tp *const_pointer;
  typedef Tp &reference;
  typedef const Tp &const_reference;
  typedef Tp value_type;

  template <class U> 
  struct rebind
  {
    typedef slice_allocator<U> other;
  };

  slice_allocator () throw () { }
  slice_allocator (const slice_allocator &) throw () { }
  template<typename Tp2>
  slice_allocator (const slice_allocator<Tp2> &) throw () { }

  ~slice_allocator () { }

  pointer address (reference x) const { return &x; }
  const_pointer address (const_reference x) const { return &x; }

  pointer allocate (size_type n, const_pointer = 0)
  {
    return salloc<Tp> (n);
  }

  void deallocate (pointer p, size_type n)
  {
    sfree<Tp> (p, n);
  }

  size_type max_size () const throw ()
  {
    return size_t (-1) / sizeof (Tp);
  }

  void construct (pointer p, const Tp &val)
  {
    ::new (p) Tp (val);
  }

  void destroy (pointer p)
  {
    p->~Tp ();
  }
};

// P. L'Ecuyer, “Maximally Equidistributed Combined Tausworthe Generators”, Mathematics of Computation, 65, 213 (1996), 203–213.
// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps
// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
struct tausworthe_random_generator
{
  uint32_t state [4];

  void operator =(const tausworthe_random_generator &src)
  {
    state [0] = src.state [0];
    state [1] = src.state [1];
    state [2] = src.state [2];
    state [3] = src.state [3];
  }

  void seed (uint32_t seed);
  uint32_t next ();
};

// Xorshift RNGs, George Marsaglia
// http://www.jstatsoft.org/v08/i14/paper
// this one is about 40% faster than the tausworthe one above (i.e. not much),
// despite the inlining, and has the issue of only creating 2**32-1 numbers.
// see also http://www.iro.umontreal.ca/~lecuyer/myftp/papers/xorshift.pdf
struct xorshift_random_generator
{
  uint32_t x, y;

  void operator =(const xorshift_random_generator &src)
  {
    x = src.x;
    y = src.y;
  }

  void seed (uint32_t seed)
  {
    x = seed;
    y = seed * 69069U;
  }

  uint32_t next ()
  {
    uint32_t t = x ^ (x << 10);
    x = y;
    y = y ^ (y >> 13) ^ t ^ (t >> 10);
    return y;
  }
};

template<class generator>
struct random_number_generator : generator
{
  // uniform distribution, 0 .. max (0, num - 1)
  uint32_t operator ()(uint32_t num)
  {
    return !is_constant (num) ? get_range (num)                        // non-constant
         : num & (num - 1)    ? (this->next () * (uint64_t)num) >> 32U // constant, non-power-of-two
         :                      this->next () & (num - 1);             // constant, power-of-two
  }

  // return a number within (min .. max)
  int operator () (int r_min, int r_max)
  {
    return is_constant (r_min) && is_constant (r_max) && r_min <= r_max
              ? r_min + operator ()(r_max - r_min + 1)
              : get_range (r_min, r_max);
  }

  double operator ()()
  {
    return this->next () / (double)0xFFFFFFFFU;
  }

protected:
  uint32_t get_range (uint32_t r_max);
  int get_range (int r_min, int r_max);
};

typedef random_number_generator<tausworthe_random_generator> rand_gen;

extern rand_gen rndm, rmg_rndm;

INTERFACE_CLASS (attachable)
struct refcnt_base
{
  typedef int refcnt_t;
  mutable refcnt_t ACC (RW, refcnt);

  MTH void refcnt_inc () const { ++refcnt; }
  MTH void refcnt_dec () const { --refcnt; }

  refcnt_base () : refcnt (0) { }
};

// to avoid branches with more advanced compilers
extern refcnt_base::refcnt_t refcnt_dummy;

template<class T>
struct refptr
{
  // p if not null
  refcnt_base::refcnt_t *refcnt_ref () { return p ? &p->refcnt : &refcnt_dummy; }

  void refcnt_dec ()
  {
    if (!is_constant (p))
      --*refcnt_ref ();
    else if (p)
      --p->refcnt;
  }

  void refcnt_inc ()
  {
    if (!is_constant (p))
      ++*refcnt_ref ();
    else if (p)
      ++p->refcnt;
  }

  T *p;

  refptr () : p(0) { }
  refptr (const refptr<T> &p) : p(p.p) { refcnt_inc (); }
  refptr (T *p) : p(p) { refcnt_inc (); }
  ~refptr () { refcnt_dec (); }

  const refptr<T> &operator =(T *o)
  {
    // if decrementing ever destroys we need to reverse the order here
    refcnt_dec ();
    p = o;
    refcnt_inc ();
    return *this;
  }

  const refptr<T> &operator =(const refptr<T> &o)
  {
    *this = o.p;
    return *this;
  }

  T &operator * () const { return *p; }
  T *operator ->() const { return  p; }

  operator T *() const { return p; }
};

typedef refptr<maptile>   maptile_ptr;
typedef refptr<object>    object_ptr;
typedef refptr<archetype> arch_ptr;
typedef refptr<client>    client_ptr;
typedef refptr<player>    player_ptr;

struct str_hash
{
  std::size_t operator ()(const char *s) const
  {
#if 0
    uint32_t hash = 0;

    /* use the one-at-a-time hash function, which supposedly is
     * better than the djb2-like one used by perl5.005, but
     * certainly is better then the bug used here before.
     * see http://burtleburtle.net/bob/hash/doobs.html
     */
    while (*s)
      {
        hash += *s++;
        hash += hash << 10;
        hash ^= hash >>  6;
      }

    hash += hash <<  3;
    hash ^= hash >> 11;
    hash += hash << 15;
#else
    // use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/)
    // it is about twice as fast as the one-at-a-time one,
    // with good distribution.
    // FNV-1a is faster on many cpus because the multiplication
    // runs concurrent with the looping logic.
    uint32_t hash = 2166136261;
    
    while (*s)
      hash = (hash ^ *s++) * 16777619;
#endif

    return hash;
  }
};

struct str_equal
{
  bool operator ()(const char *a, const char *b) const
  {
    return !strcmp (a, b);
  }
};

// Mostly the same as std::vector, but insert/erase can reorder
// the elements, making append(=insert)/remove O(1) instead of O(n).
//
// NOTE: only some forms of erase are available
template<class T>
struct unordered_vector : std::vector<T, slice_allocator<T> >
{
  typedef typename unordered_vector::iterator iterator;

  void erase (unsigned int pos)
  {
    if (pos < this->size () - 1)
      (*this)[pos] = (*this)[this->size () - 1];

    this->pop_back ();
  }

  void erase (iterator i)
  {
    erase ((unsigned int )(i - this->begin ()));
  }
};

// This container blends advantages of linked lists
// (efficiency) with vectors (random access) by
// by using an unordered vector and storing the vector
// index inside the object.
//
// + memory-efficient on most 64 bit archs
// + O(1) insert/remove
// + free unique (but varying) id for inserted objects
// + cache-friendly iteration
// - only works for pointers to structs
//
// NOTE: only some forms of erase/insert are available
typedef int object_vector_index;

template<class T, object_vector_index T::*indexmember>
struct object_vector : std::vector<T *, slice_allocator<T *> >
{
  typedef typename object_vector::iterator iterator;

  bool contains (const T *obj) const
  {
    return obj->*indexmember;
  }

  iterator find (const T *obj)
  {
    return obj->*indexmember
      ? this->begin () + obj->*indexmember - 1
      : this->end ();
  }

  void push_back (T *obj)
  {
    std::vector<T *, slice_allocator<T *> >::push_back (obj);
    obj->*indexmember = this->size ();
  }

  void insert (T *obj)
  {
    push_back (obj);
  }

  void insert (T &obj)
  {
    insert (&obj);
  }

  void erase (T *obj)
  {
    unsigned int pos = obj->*indexmember;
    obj->*indexmember = 0;

    if (pos < this->size ())
      {
        (*this)[pos - 1] = (*this)[this->size () - 1];
        (*this)[pos - 1]->*indexmember = pos;
      }

    this->pop_back ();
  }

  void erase (T &obj)
  {
    erase (&obj);
  }
};

// basically does what strncpy should do, but appends "..." to strings exceeding length
// returns the number of bytes actually used (including \0)
int assign (char *dst, const char *src, int maxsize);

// type-safe version of assign
template<int N>
inline int assign (char (&dst)[N], const char *src)
{
  return assign ((char *)&dst, src, N);
}

typedef double tstamp;

// return current time as timestamp
tstamp now ();

int similar_direction (int a, int b);

// like v?sprintf, but returns a "static" buffer
char *vformat (const char *format, va_list ap);
char *format (const char *format, ...) attribute ((format (printf, 1, 2)));

// safety-check player input which will become object->msg
bool msg_is_safe (const char *msg);

/////////////////////////////////////////////////////////////////////////////
// threads, very very thin wrappers around pthreads

struct thread
{
  pthread_t id;

  void start (void *(*start_routine)(void *), void *arg = 0);

  void cancel ()
  {
    pthread_cancel (id);
  }

  void *join ()
  {
    void *ret;

    if (pthread_join (id, &ret))
      cleanup ("pthread_join failed", 1);

    return ret;
  }
};

// note that mutexes are not classes
typedef pthread_mutex_t smutex;

#if __linux && defined (PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP)
 #define SMUTEX_INITIALISER PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP
#else
 #define SMUTEX_INITIALISER PTHREAD_MUTEX_INITIALIZER
#endif

#define SMUTEX(name) smutex name = SMUTEX_INITIALISER
#define SMUTEX_LOCK(name)   pthread_mutex_lock   (&(name))
#define SMUTEX_UNLOCK(name) pthread_mutex_unlock (&(name))

typedef pthread_cond_t scond;

#define SCOND(name) scond name = PTHREAD_COND_INITIALIZER
#define SCOND_SIGNAL(name)     pthread_cond_signal    (&(name))
#define SCOND_BROADCAST(name)  pthread_cond_broadcast (&(name))
#define SCOND_WAIT(name,mutex) pthread_cond_wait      (&(name), &(mutex))

#endif

Revision:	1.94
Committed:	Sun Nov 8 16:13:45 2009 UTC (14 years, 6 months ago) by root
Content type:	text/plain
Branch:	MAIN
Changes since 1.93:	+1 -1 lines
Log Message:	optimisations, cache volume/nrof in mapspace
#	Content
1	/*
2	* This file is part of Deliantra, the Roguelike Realtime MMORPG.
3	*
4	* Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team
5	*
6	* Deliantra is free software: you can redistribute it and/or modify it under
7	* the terms of the Affero GNU General Public License as published by the
8	* Free Software Foundation, either version 3 of the License, or (at your
9	* option) any later version.
10	*
11	* This program is distributed in the hope that it will be useful,
12	* but WITHOUT ANY WARRANTY; without even the implied warranty of
13	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14	* GNU General Public License for more details.
15	*
16	* You should have received a copy of the Affero GNU General Public License
17	* and the GNU General Public License along with this program. If not, see
18	* <http://www.gnu.org/licenses/>.
19	*
20	* The authors can be reached via e-mail to <support@deliantra.net>
21	*/
22
23	#ifndef UTIL_H__
24	#define UTIL_H__
25
26	#include <compiler.h>
27
28	#define DEBUG_POISON 0x00 // poison memory before freeing it if != 0
29	#define DEBUG_SALLOC 0 // add a debug wrapper around all sallocs
30	#define PREFER_MALLOC 0 // use malloc and not the slice allocator
31
32	#include <pthread.h>
33
34	#include <cstddef>
35	#include <cmath>
36	#include <new>
37	#include <vector>
38
39	#include <glib.h>
40
41	#include <shstr.h>
42	#include <traits.h>
43
44	#if DEBUG_SALLOC
45	# define g_slice_alloc0(s) debug_slice_alloc0(s)
46	# define g_slice_alloc(s) debug_slice_alloc(s)
47	# define g_slice_free1(s,p) debug_slice_free1(s,p)
48	void *g_slice_alloc (unsigned long size);
49	void *g_slice_alloc0 (unsigned long size);
50	void g_slice_free1 (unsigned long size, void *ptr);
51	#elif PREFER_MALLOC
52	# define g_slice_alloc0(s) calloc (1, (s))
53	# define g_slice_alloc(s) malloc ((s))
54	# define g_slice_free1(s,p) free ((p))
55	#endif
56
57	// use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever)
58	#define auto(var,expr) decltype(expr) var = (expr)
59
60	// very ugly macro that basically declares and initialises a variable
61	// that is in scope for the next statement only
62	// works only for stuff that can be assigned 0 and converts to false
63	// (note: works great for pointers)
64	// most ugly macro I ever wrote
65	#define statementvar(type, name, value) if (type name = 0) { } else if (((name) = (value)), 1)
66
67	// in range including end
68	#define IN_RANGE_INC(val,beg,end) \
69	((unsigned int)(val) - (unsigned int)(beg) <= (unsigned int)(end) - (unsigned int)(beg))
70
71	// in range excluding end
72	#define IN_RANGE_EXC(val,beg,end) \
73	((unsigned int)(val) - (unsigned int)(beg) < (unsigned int)(end) - (unsigned int)(beg))
74
75	void cleanup (const char *cause, bool make_core = false);
76	void fork_abort (const char *msg);
77
78	// rationale for using (U) not (T) is to reduce signed/unsigned issues,
79	// as a is often a constant while b is the variable. it is still a bug, though.
80	template<typename T, typename U> static inline T min (T a, U b) { return (U)a < b ? (U)a : b; }
81	template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; }
82	template<typename T, typename U, typename V> static inline T clamp (T v, U a, V b) { return v < (T)a ? (T)a : v >(T)b ? (T)b : v; }
83
84	template<typename T, typename U> static inline void min_it (T &v, U m) { v = min (v, (T)m); }
85	template<typename T, typename U> static inline void max_it (T &v, U m) { v = max (v, (T)m); }
86	template<typename T, typename U, typename V> static inline void clamp_it (T &v, U a, V b) { v = clamp (v, (T)a, (T)b); }
87
88	template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; }
89
90	template<typename T, typename U, typename V> static inline T min (T a, U b, V c) { return min (a, min (b, c)); }
91	template<typename T, typename U, typename V> static inline T max (T a, U b, V c) { return max (a, max (b, c)); }
92
93	// sign returns -1 or +1
94	template<typename T>
95	static inline T sign (T v) { return v < 0 ? -1 : +1; }
96	// relies on 2c representation
97	template<>
98	inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); }
99
100	// sign0 returns -1, 0 or +1
101	template<typename T>
102	static inline T sign0 (T v) { return v ? sign (v) : 0; }
103
104	// div* only work correctly for div > 0
105	// div, with correct rounding (< 0.5 downwards, >=0.5 upwards)
106	template<typename T> static inline T div (T val, T div)
107	{
108	return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div;
109	}
110	// div, round-up
111	template<typename T> static inline T div_ru (T val, T div)
112	{
113	return expect_false (val < 0) ? - ((-val ) / div) : (val + div - 1) / div;
114	}
115	// div, round-down
116	template<typename T> static inline T div_rd (T val, T div)
117	{
118	return expect_false (val < 0) ? - ((-val + (div - 1) ) / div) : (val ) / div;
119	}
120
121	// lerp* only work correctly for min_in < max_in
122	// Linear intERPolate, scales val from min_in..max_in to min_out..max_out
123	template<typename T>
124	static inline T
125	lerp (T val, T min_in, T max_in, T min_out, T max_out)
126	{
127	return min_out + div <T> ((val - min_in) * (max_out - min_out), max_in - min_in);
128	}
129
130	// lerp, round-down
131	template<typename T>
132	static inline T
133	lerp_rd (T val, T min_in, T max_in, T min_out, T max_out)
134	{
135	return min_out + div_rd<T> ((val - min_in) * (max_out - min_out), max_in - min_in);
136	}
137
138	// lerp, round-up
139	template<typename T>
140	static inline T
141	lerp_ru (T val, T min_in, T max_in, T min_out, T max_out)
142	{
143	return min_out + div_ru<T> ((val - min_in) * (max_out - min_out), max_in - min_in);
144	}
145
146	// lots of stuff taken from FXT
147
148	/* Rotate right. This is used in various places for checksumming */
149	//TODO: that sucks, use a better checksum algo
150	static inline uint32_t
151	rotate_right (uint32_t c, uint32_t count = 1)
152	{
153	return (c << (32 - count)) \| (c >> count);
154	}
155
156	static inline uint32_t
157	rotate_left (uint32_t c, uint32_t count = 1)
158	{
159	return (c >> (32 - count)) \| (c << count);
160	}
161
162	// Return abs(a-b)
163	// Both a and b must not have the most significant bit set
164	static inline uint32_t
165	upos_abs_diff (uint32_t a, uint32_t b)
166	{
167	long d1 = b - a;
168	long d2 = (d1 & (d1 >> 31)) << 1;
169
170	return d1 - d2; // == (b - d) - (a + d);
171	}
172
173	// Both a and b must not have the most significant bit set
174	static inline uint32_t
175	upos_min (uint32_t a, uint32_t b)
176	{
177	int32_t d = b - a;
178	d &= d >> 31;
179	return a + d;
180	}
181
182	// Both a and b must not have the most significant bit set
183	static inline uint32_t
184	upos_max (uint32_t a, uint32_t b)
185	{
186	int32_t d = b - a;
187	d &= d >> 31;
188	return b - d;
189	}
190
191	// this is much faster than crossfire's original algorithm
192	// on modern cpus
193	inline int
194	isqrt (int n)
195	{
196	return (int)sqrtf ((float)n);
197	}
198
199	// this is kind of like the ^^ operator, if it would exist, without sequence point.
200	// more handy than it looks like, due to the implicit !! done on its arguments
201	inline bool
202	logical_xor (bool a, bool b)
203	{
204	return a != b;
205	}
206
207	inline bool
208	logical_implies (bool a, bool b)
209	{
210	return a <= b;
211	}
212
213	// this is only twice as fast as naive sqrtf (dxdy+dydy)
214	#if 0
215	// and has a max. error of 6 in the range -100..+100.
216	#else
217	// and has a max. error of 9 in the range -100..+100.
218	#endif
219	inline int
220	idistance (int dx, int dy)
221	{
222	unsigned int dx_ = abs (dx);
223	unsigned int dy_ = abs (dy);
224
225	#if 0
226	return dx_ > dy_
227	? (dx_ * 61685 + dy_ * 26870) >> 16
228	: (dy_ * 61685 + dx_ * 26870) >> 16;
229	#else
230	return dx_ + dy_ - min (dx_, dy_) * 5 / 8;
231	#endif
232	}
233
234	/*
235	* absdir(int): Returns a number between 1 and 8, which represent
236	* the "absolute" direction of a number (it actually takes care of
237	* "overflow" in previous calculations of a direction).
238	*/
239	inline int
240	absdir (int d)
241	{
242	return ((d - 1) & 7) + 1;
243	}
244
245	extern ssize_t slice_alloc; // statistics
246
247	void *salloc_ (int n) throw (std::bad_alloc);
248	void salloc_ (int n, void src) throw (std::bad_alloc);
249
250	// strictly the same as g_slice_alloc, but never returns 0
251	template<typename T>
252	inline T salloc (int n = 1) throw (std::bad_alloc) { return (T )salloc_ (n * sizeof (T)); }
253
254	// also copies src into the new area, like "memdup"
255	// if src is 0, clears the memory
256	template<typename T>
257	inline T salloc (int n, T src) throw (std::bad_alloc) { return (T )salloc_ (n sizeof (T), (void *)src); }
258
259	// clears the memory
260	template<typename T>
261	inline T salloc0(int n = 1) throw (std::bad_alloc) { return (T )salloc_ (n * sizeof (T), 0); }
262
263	// for symmetry
264	template<typename T>
265	inline void sfree (T *ptr, int n = 1) throw ()
266	{
267	if (expect_true (ptr))
268	{
269	slice_alloc -= n * sizeof (T);
270	if (DEBUG_POISON) memset (ptr, DEBUG_POISON, n * sizeof (T));
271	g_slice_free1 (n * sizeof (T), (void *)ptr);
272	assert (slice_alloc >= 0);//D
273	}
274	}
275
276	// nulls the pointer
277	template<typename T>
278	inline void sfree0 (T *&ptr, int n = 1) throw ()
279	{
280	sfree<T> (ptr, n);
281	ptr = 0;
282	}
283
284	// makes dynamically allocated objects zero-initialised
285	struct zero_initialised
286	{
287	void operator new (size_t s, void p)
288	{
289	memset (p, 0, s);
290	return p;
291	}
292
293	void *operator new (size_t s)
294	{
295	return salloc0<char> (s);
296	}
297
298	void *operator new[] (size_t s)
299	{
300	return salloc0<char> (s);
301	}
302
303	void operator delete (void *p, size_t s)
304	{
305	sfree ((char *)p, s);
306	}
307
308	void operator delete[] (void *p, size_t s)
309	{
310	sfree ((char *)p, s);
311	}
312	};
313
314	// makes dynamically allocated objects zero-initialised
315	struct slice_allocated
316	{
317	void operator new (size_t s, void p)
318	{
319	return p;
320	}
321
322	void *operator new (size_t s)
323	{
324	return salloc<char> (s);
325	}
326
327	void *operator new[] (size_t s)
328	{
329	return salloc<char> (s);
330	}
331
332	void operator delete (void *p, size_t s)
333	{
334	sfree ((char *)p, s);
335	}
336
337	void operator delete[] (void *p, size_t s)
338	{
339	sfree ((char *)p, s);
340	}
341	};
342
343	// a STL-compatible allocator that uses g_slice
344	// boy, this is verbose
345	template<typename Tp>
346	struct slice_allocator
347	{
348	typedef size_t size_type;
349	typedef ptrdiff_t difference_type;
350	typedef Tp *pointer;
351	typedef const Tp *const_pointer;
352	typedef Tp &reference;
353	typedef const Tp &const_reference;
354	typedef Tp value_type;
355
356	template <class U>
357	struct rebind
358	{
359	typedef slice_allocator<U> other;
360	};
361
362	slice_allocator () throw () { }
363	slice_allocator (const slice_allocator &) throw () { }
364	template<typename Tp2>
365	slice_allocator (const slice_allocator<Tp2> &) throw () { }
366
367	~slice_allocator () { }
368
369	pointer address (reference x) const { return &x; }
370	const_pointer address (const_reference x) const { return &x; }
371
372	pointer allocate (size_type n, const_pointer = 0)
373	{
374	return salloc<Tp> (n);
375	}
376
377	void deallocate (pointer p, size_type n)
378	{
379	sfree<Tp> (p, n);
380	}
381
382	size_type max_size () const throw ()
383	{
384	return size_t (-1) / sizeof (Tp);
385	}
386
387	void construct (pointer p, const Tp &val)
388	{
389	::new (p) Tp (val);
390	}
391
392	void destroy (pointer p)
393	{
394	p->~Tp ();
395	}
396	};
397
398	// P. L'Ecuyer, “Maximally Equidistributed Combined Tausworthe Generators”, Mathematics of Computation, 65, 213 (1996), 203–213.
399	// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps
400	// http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
401	struct tausworthe_random_generator
402	{
403	uint32_t state [4];
404
405	void operator =(const tausworthe_random_generator &src)
406	{
407	state [0] = src.state [0];
408	state [1] = src.state [1];
409	state [2] = src.state [2];
410	state [3] = src.state [3];
411	}
412
413	void seed (uint32_t seed);
414	uint32_t next ();
415	};
416
417	// Xorshift RNGs, George Marsaglia
418	// http://www.jstatsoft.org/v08/i14/paper
419	// this one is about 40% faster than the tausworthe one above (i.e. not much),
420	// despite the inlining, and has the issue of only creating 2**32-1 numbers.
421	// see also http://www.iro.umontreal.ca/~lecuyer/myftp/papers/xorshift.pdf
422	struct xorshift_random_generator
423	{
424	uint32_t x, y;
425
426	void operator =(const xorshift_random_generator &src)
427	{
428	x = src.x;
429	y = src.y;
430	}
431
432	void seed (uint32_t seed)
433	{
434	x = seed;
435	y = seed * 69069U;
436	}
437
438	uint32_t next ()
439	{
440	uint32_t t = x ^ (x << 10);
441	x = y;
442	y = y ^ (y >> 13) ^ t ^ (t >> 10);
443	return y;
444	}
445	};
446
447	template<class generator>
448	struct random_number_generator : generator
449	{
450	// uniform distribution, 0 .. max (0, num - 1)
451	uint32_t operator ()(uint32_t num)
452	{
453	return !is_constant (num) ? get_range (num) // non-constant
454	: num & (num - 1) ? (this->next () * (uint64_t)num) >> 32U // constant, non-power-of-two
455	: this->next () & (num - 1); // constant, power-of-two
456	}
457
458	// return a number within (min .. max)
459	int operator () (int r_min, int r_max)
460	{
461	return is_constant (r_min) && is_constant (r_max) && r_min <= r_max
462	? r_min + operator ()(r_max - r_min + 1)
463	: get_range (r_min, r_max);
464	}
465
466	double operator ()()
467	{
468	return this->next () / (double)0xFFFFFFFFU;
469	}
470
471	protected:
472	uint32_t get_range (uint32_t r_max);
473	int get_range (int r_min, int r_max);
474	};
475
476	typedef random_number_generator<tausworthe_random_generator> rand_gen;
477
478	extern rand_gen rndm, rmg_rndm;
479
480	INTERFACE_CLASS (attachable)
481	struct refcnt_base
482	{
483	typedef int refcnt_t;
484	mutable refcnt_t ACC (RW, refcnt);
485
486	MTH void refcnt_inc () const { ++refcnt; }
487	MTH void refcnt_dec () const { --refcnt; }
488
489	refcnt_base () : refcnt (0) { }
490	};
491
492	// to avoid branches with more advanced compilers
493	extern refcnt_base::refcnt_t refcnt_dummy;
494
495	template<class T>
496	struct refptr
497	{
498	// p if not null
499	refcnt_base::refcnt_t *refcnt_ref () { return p ? &p->refcnt : &refcnt_dummy; }
500
501	void refcnt_dec ()
502	{
503	if (!is_constant (p))
504	--*refcnt_ref ();
505	else if (p)
506	--p->refcnt;
507	}
508
509	void refcnt_inc ()
510	{
511	if (!is_constant (p))
512	++*refcnt_ref ();
513	else if (p)
514	++p->refcnt;
515	}
516
517	T *p;
518
519	refptr () : p(0) { }
520	refptr (const refptr<T> &p) : p(p.p) { refcnt_inc (); }
521	refptr (T *p) : p(p) { refcnt_inc (); }
522	~refptr () { refcnt_dec (); }
523
524	const refptr<T> &operator =(T *o)
525	{
526	// if decrementing ever destroys we need to reverse the order here
527	refcnt_dec ();
528	p = o;
529	refcnt_inc ();
530	return *this;
531	}
532
533	const refptr<T> &operator =(const refptr<T> &o)
534	{
535	*this = o.p;
536	return *this;
537	}
538
539	T &operator * () const { return *p; }
540	T *operator ->() const { return p; }
541
542	operator T *() const { return p; }
543	};
544
545	typedef refptr<maptile> maptile_ptr;
546	typedef refptr<object> object_ptr;
547	typedef refptr<archetype> arch_ptr;
548	typedef refptr<client> client_ptr;
549	typedef refptr<player> player_ptr;
550
551	struct str_hash
552	{
553	std::size_t operator ()(const char *s) const
554	{
555	#if 0
556	uint32_t hash = 0;
557
558	/* use the one-at-a-time hash function, which supposedly is
559	* better than the djb2-like one used by perl5.005, but
560	* certainly is better then the bug used here before.
561	* see http://burtleburtle.net/bob/hash/doobs.html
562	*/
563	while (*s)
564	{
565	hash += *s++;
566	hash += hash << 10;
567	hash ^= hash >> 6;
568	}
569
570	hash += hash << 3;
571	hash ^= hash >> 11;
572	hash += hash << 15;
573	#else
574	// use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/)
575	// it is about twice as fast as the one-at-a-time one,
576	// with good distribution.
577	// FNV-1a is faster on many cpus because the multiplication
578	// runs concurrent with the looping logic.
579	uint32_t hash = 2166136261;
580
581	while (*s)
582	hash = (hash ^ s++) 16777619;
583	#endif
584
585	return hash;
586	}
587	};
588
589	struct str_equal
590	{
591	bool operator ()(const char a, const char b) const
592	{
593	return !strcmp (a, b);
594	}
595	};
596
597	// Mostly the same as std::vector, but insert/erase can reorder
598	// the elements, making append(=insert)/remove O(1) instead of O(n).
599	//
600	// NOTE: only some forms of erase are available
601	template<class T>
602	struct unordered_vector : std::vector<T, slice_allocator<T> >
603	{
604	typedef typename unordered_vector::iterator iterator;
605
606	void erase (unsigned int pos)
607	{
608	if (pos < this->size () - 1)
609	(this)[pos] = (this)[this->size () - 1];
610
611	this->pop_back ();
612	}
613
614	void erase (iterator i)
615	{
616	erase ((unsigned int )(i - this->begin ()));
617	}
618	};
619
620	// This container blends advantages of linked lists
621	// (efficiency) with vectors (random access) by
622	// by using an unordered vector and storing the vector
623	// index inside the object.
624	//
625	// + memory-efficient on most 64 bit archs
626	// + O(1) insert/remove
627	// + free unique (but varying) id for inserted objects
628	// + cache-friendly iteration
629	// - only works for pointers to structs
630	//
631	// NOTE: only some forms of erase/insert are available
632	typedef int object_vector_index;
633
634	template<class T, object_vector_index T::*indexmember>
635	struct object_vector : std::vector<T , slice_allocator<T > >
636	{
637	typedef typename object_vector::iterator iterator;
638
639	bool contains (const T *obj) const
640	{
641	return obj->*indexmember;
642	}
643
644	iterator find (const T *obj)
645	{
646	return obj->*indexmember
647	? this->begin () + obj->*indexmember - 1
648	: this->end ();
649	}
650
651	void push_back (T *obj)
652	{
653	std::vector<T , slice_allocator<T > >::push_back (obj);
654	obj->*indexmember = this->size ();
655	}
656
657	void insert (T *obj)
658	{
659	push_back (obj);
660	}
661
662	void insert (T &obj)
663	{
664	insert (&obj);
665	}
666
667	void erase (T *obj)
668	{
669	unsigned int pos = obj->*indexmember;
670	obj->*indexmember = 0;
671
672	if (pos < this->size ())
673	{
674	(this)[pos - 1] = (this)[this->size () - 1];
675	(this)[pos - 1]->indexmember = pos;
676	}
677
678	this->pop_back ();
679	}
680
681	void erase (T &obj)
682	{
683	erase (&obj);
684	}
685	};
686
687	// basically does what strncpy should do, but appends "..." to strings exceeding length
688	// returns the number of bytes actually used (including \0)
689	int assign (char dst, const char src, int maxsize);
690
691	// type-safe version of assign
692	template<int N>
693	inline int assign (char (&dst)[N], const char *src)
694	{
695	return assign ((char *)&dst, src, N);
696	}
697
698	typedef double tstamp;
699
700	// return current time as timestamp
701	tstamp now ();
702
703	int similar_direction (int a, int b);
704
705	// like v?sprintf, but returns a "static" buffer
706	char vformat (const char format, va_list ap);
707	char format (const char format, ...) attribute ((format (printf, 1, 2)));
708
709	// safety-check player input which will become object->msg
710	bool msg_is_safe (const char *msg);
711
712	/////////////////////////////////////////////////////////////////////////////
713	// threads, very very thin wrappers around pthreads
714
715	struct thread
716	{
717	pthread_t id;
718
719	void start (void (start_routine)(void ), void arg = 0);
720
721	void cancel ()
722	{
723	pthread_cancel (id);
724	}
725
726	void *join ()
727	{
728	void *ret;
729
730	if (pthread_join (id, &ret))
731	cleanup ("pthread_join failed", 1);
732
733	return ret;
734	}
735	};
736
737	// note that mutexes are not classes
738	typedef pthread_mutex_t smutex;
739
740	#if __linux && defined (PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP)
741	#define SMUTEX_INITIALISER PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP
742	#else
743	#define SMUTEX_INITIALISER PTHREAD_MUTEX_INITIALIZER
744	#endif
745
746	#define SMUTEX(name) smutex name = SMUTEX_INITIALISER
747	#define SMUTEX_LOCK(name) pthread_mutex_lock (&(name))
748	#define SMUTEX_UNLOCK(name) pthread_mutex_unlock (&(name))
749
750	typedef pthread_cond_t scond;
751
752	#define SCOND(name) scond name = PTHREAD_COND_INITIALIZER
753	#define SCOND_SIGNAL(name) pthread_cond_signal (&(name))
754	#define SCOND_BROADCAST(name) pthread_cond_broadcast (&(name))
755	#define SCOND_WAIT(name,mutex) pthread_cond_wait (&(name), &(mutex))
756
757	#endif
758