--- deliantra/server/include/util.h 2009/10/15 21:09:32 1.91
+++ deliantra/server/include/util.h 2018/11/27 18:47:35 1.128
@@ -1,54 +1,35 @@
/*
* This file is part of Deliantra, the Roguelike Realtime MMORPG.
- *
- * Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team
- *
+ *
+ * Copyright (©) 2017,2018 Marc Alexander Lehmann / the Deliantra team
+ * Copyright (©) 2005,2006,2007,2008,2009,2010,2011,2012,2013,2014,2015,2016 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
* .
- *
+ *
* The authors can be reached via e-mail to
*/
#ifndef UTIL_H__
#define UTIL_H__
+#include
+
#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
-#if __GNUC__ >= 3
-# define is_constant(c) __builtin_constant_p (c)
-# define expect(expr,value) __builtin_expect ((expr),(value))
-# define prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
-# define noinline __attribute__((__noinline__))
-#else
-# define is_constant(c) 0
-# define expect(expr,value) (expr)
-# define prefetch(addr,rw,locality)
-# define noinline
-#endif
-
-#if __GNUC__ < 4 || (__GNUC__ == 4 || __GNUC_MINOR__ < 4)
-# define decltype(x) typeof(x)
-#endif
-
-// put into ifs if you are very sure that the expression
-// is mostly true or mosty false. note that these return
-// booleans, not the expression.
-#define expect_false(expr) expect ((expr) ? 1 : 0, 0)
-#define expect_true(expr) expect ((expr) ? 1 : 0, 1)
-
#include
#include
@@ -58,6 +39,8 @@
#include
+#include
+
#include
#include
@@ -77,6 +60,17 @@
// use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever)
#define auto(var,expr) decltype(expr) var = (expr)
+#if cplusplus_does_not_suck /* still sucks in codesize with gcc 6, although local types work now */
+// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm)
+template
+static inline int array_length (const T (&arr)[N])
+{
+ return N;
+}
+#else
+#define array_length(name) (sizeof (name) / sizeof (name [0]))
+#endif
+
// 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
@@ -92,13 +86,13 @@
#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);
+ecb_cold void cleanup (const char *cause, bool make_core = false);
+ecb_cold 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 static inline T min (T a, U b) { return (U)a < b ? (U)a : b; }
-template static inline T max (T a, U b) { return (U)a > b ? (U)a : b; }
+template static inline T min (T a, U b) { return a < (T)b ? a : (T)b; }
+template static inline T max (T a, U b) { return a > (T)b ? a : (T)b; }
template static inline T clamp (T v, U a, V b) { return v < (T)a ? (T)a : v >(T)b ? (T)b : v; }
template static inline void min_it (T &v, U m) { v = min (v, (T)m); }
@@ -115,18 +109,30 @@
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); }
+inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); }
+template<>
+inline sint16 sign (sint16 v) { return 1 - (sint16 (uint16 (v) >> 15) * 2); }
+template<>
+inline sint32 sign (sint32 v) { return 1 - (sint32 (uint32 (v) >> 31) * 2); }
// sign0 returns -1, 0 or +1
template
static inline T sign0 (T v) { return v ? sign (v) : 0; }
+//clashes with C++0x
+template
+static inline T copysign (T a, U b) { return a > 0 ? b : -b; }
+
// div* only work correctly for div > 0
// div, with correct rounding (< 0.5 downwards, >=0.5 upwards)
template static inline T div (T val, T div)
{
return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div;
}
+
+template<> inline float div (float val, float div) { return val / div; }
+template<> inline double div (double val, double div) { return val / div; }
+
// div, round-up
template static inline T div_ru (T val, T div)
{
@@ -208,7 +214,7 @@
return b - d;
}
-// this is much faster than crossfires original algorithm
+// this is much faster than crossfire's original algorithm
// on modern cpus
inline int
isqrt (int n)
@@ -216,15 +222,29 @@
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
+inline int
idistance (int dx, int dy)
-{
+{
unsigned int dx_ = abs (dx);
unsigned int dy_ = abs (dy);
@@ -237,6 +257,26 @@
#endif
}
+// can be substantially faster than floor, if your value range allows for it
+template
+inline T
+fastfloor (T x)
+{
+ return std::floor (x);
+}
+
+inline float
+fastfloor (float x)
+{
+ return sint32(x) - (x < 0);
+}
+
+inline double
+fastfloor (double x)
+{
+ return sint64(x) - (x < 0);
+}
+
/*
* absdir(int): Returns a number between 1 and 8, which represent
* the "absolute" direction of a number (it actually takes care of
@@ -248,40 +288,43 @@
return ((d - 1) & 7) + 1;
}
+#define for_all_bits_sparse_32(mask, idxvar) \
+ for (uint32_t idxvar, mask_ = mask; \
+ mask_ && ((idxvar = ecb_ctz32 (mask_)), mask_ &= ~(1 << idxvar), 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);
+void *salloc_ (int n);
+void *salloc_ (int n, void *src);
// strictly the same as g_slice_alloc, but never returns 0
template
-inline T *salloc (int n = 1) throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T)); }
+inline T *salloc (int n = 1) { return (T *)salloc_ (n * sizeof (T)); }
// also copies src into the new area, like "memdup"
// if src is 0, clears the memory
template
-inline T *salloc (int n, T *src) throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T), (void *)src); }
+inline T *salloc (int n, T *src) { return (T *)salloc_ (n * sizeof (T), (void *)src); }
// clears the memory
template
-inline T *salloc0(int n = 1) throw (std::bad_alloc) { return (T *)salloc_ (n * sizeof (T), 0); }
+inline T *salloc0(int n = 1) { return (T *)salloc_ (n * sizeof (T), 0); }
// for symmetry
template
-inline void sfree (T *ptr, int n = 1) throw ()
+inline void sfree (T *ptr, int n = 1) noexcept
{
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
-inline void sfree0 (T *&ptr, int n = 1) throw ()
+inline void sfree0 (T *&ptr, int n = 1) noexcept
{
sfree (ptr, n);
ptr = 0;
@@ -359,16 +402,16 @@
typedef const Tp &const_reference;
typedef Tp value_type;
- template
+ template
struct rebind
{
typedef slice_allocator other;
};
- slice_allocator () throw () { }
- slice_allocator (const slice_allocator &) throw () { }
+ slice_allocator () noexcept { }
+ slice_allocator (const slice_allocator &) noexcept { }
template
- slice_allocator (const slice_allocator &) throw () { }
+ slice_allocator (const slice_allocator &) noexcept { }
~slice_allocator () { }
@@ -385,7 +428,7 @@
sfree (p, n);
}
- size_type max_size () const throw ()
+ size_type max_size () const noexcept
{
return size_t (-1) / sizeof (Tp);
}
@@ -401,88 +444,70 @@
}
};
-// 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
+// basically a memory area, but refcounted
+struct refcnt_buf
{
- uint32_t state [4];
+ char *data;
- void operator =(const tausworthe_random_generator &src)
+ refcnt_buf (size_t size = 0);
+ refcnt_buf (void *data, size_t size);
+
+ refcnt_buf (const refcnt_buf &src)
{
- state [0] = src.state [0];
- state [1] = src.state [1];
- state [2] = src.state [2];
- state [3] = src.state [3];
+ data = src.data;
+ inc ();
}
- void seed (uint32_t seed);
- uint32_t next ();
-};
+ ~refcnt_buf ();
-// 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;
+ refcnt_buf &operator =(const refcnt_buf &src);
- void operator =(const xorshift_random_generator &src)
+ operator char *()
{
- x = src.x;
- y = src.y;
+ return data;
}
- void seed (uint32_t seed)
+ size_t size () const
{
- x = seed;
- y = seed * 69069U;
+ return _size ();
}
- uint32_t next ()
+protected:
+ enum {
+ overhead = sizeof (uint32_t) * 2
+ };
+
+ uint32_t &_size () const
{
- uint32_t t = x ^ (x << 10);
- x = y;
- y = y ^ (y >> 13) ^ t ^ (t >> 10);
- return y;
+ return ((unsigned int *)data)[-2];
}
-};
-template
-struct random_number_generator : generator
-{
- // uniform distribution, 0 .. max (0, num - 1)
- uint32_t operator ()(uint32_t num)
+ uint32_t &_refcnt () const
{
- 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 ((unsigned int *)data)[-1];
}
- // return a number within (min .. max)
- int operator () (int r_min, int r_max)
+ void _alloc (uint32_t size)
{
- 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);
+ data = ((char *)salloc (size + overhead)) + overhead;
+ _size () = size;
+ _refcnt () = 1;
}
- double operator ()()
+ void _dealloc ();
+
+ void inc ()
{
- return this->next () / (double)0xFFFFFFFFU;
+ ++_refcnt ();
}
-protected:
- uint32_t get_range (uint32_t r_max);
- int get_range (int r_min, int r_max);
+ void dec ()
+ {
+ if (!--_refcnt ())
+ _dealloc ();
+ }
};
-typedef random_number_generator rand_gen;
-
-extern rand_gen rndm, rmg_rndm;
-
INTERFACE_CLASS (attachable)
struct refcnt_base
{
@@ -506,7 +531,7 @@
void refcnt_dec ()
{
- if (!is_constant (p))
+ if (!ecb_is_constant (p))
--*refcnt_ref ();
else if (p)
--p->refcnt;
@@ -514,7 +539,7 @@
void refcnt_inc ()
{
- if (!is_constant (p))
+ if (!ecb_is_constant (p))
++*refcnt_ref ();
else if (p)
++p->refcnt;
@@ -553,43 +578,51 @@
typedef refptr arch_ptr;
typedef refptr client_ptr;
typedef refptr player_ptr;
+typedef refptr region_ptr;
+
+#define STRHSH_NULL 2166136261
+
+static inline uint32_t
+strhsh (const char *s)
+{
+ // 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 concurrently with the looping logic.
+ // we modify the hash a bit to improve its distribution
+ uint32_t hash = STRHSH_NULL;
+
+ while (*s)
+ hash = (hash ^ *s++) * 16777619U;
+
+ return hash ^ (hash >> 16);
+}
+
+static inline uint32_t
+memhsh (const char *s, size_t len)
+{
+ uint32_t hash = STRHSH_NULL;
+
+ while (len--)
+ hash = (hash ^ *s++) * 16777619U;
+
+ return hash;
+}
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 strhsh (s);
+ }
- return hash;
+ std::size_t operator ()(const shstr &s) const
+ {
+ return strhsh (s);
}
+
+ typedef ska::power_of_two_hash_policy hash_policy;
};
struct str_equal
@@ -625,7 +658,7 @@
// This container blends advantages of linked lists
// (efficiency) with vectors (random access) by
-// by using an unordered vector and storing the vector
+// using an unordered vector and storing the vector
// index inside the object.
//
// + memory-efficient on most 64 bit archs
@@ -672,7 +705,7 @@
void erase (T *obj)
{
- unsigned int pos = obj->*indexmember;
+ object_vector_index pos = obj->*indexmember;
obj->*indexmember = 0;
if (pos < this->size ())
@@ -690,6 +723,74 @@
}
};
+/////////////////////////////////////////////////////////////////////////////
+
+// something like a vector or stack, but without
+// out of bounds checking
+template
+struct fixed_stack
+{
+ T *data;
+ int size;
+ int max;
+
+ fixed_stack ()
+ : size (0), data (0)
+ {
+ }
+
+ fixed_stack (int max)
+ : size (0), max (max)
+ {
+ data = salloc (max);
+ }
+
+ void reset (int new_max)
+ {
+ sfree (data, max);
+ size = 0;
+ max = new_max;
+ data = salloc (max);
+ }
+
+ void free ()
+ {
+ sfree (data, max);
+ data = 0;
+ }
+
+ ~fixed_stack ()
+ {
+ sfree (data, max);
+ }
+
+ T &operator[](int idx)
+ {
+ return data [idx];
+ }
+
+ void push (T v)
+ {
+ data [size++] = v;
+ }
+
+ T &pop ()
+ {
+ return data [--size];
+ }
+
+ T remove (int idx)
+ {
+ T v = data [idx];
+
+ data [idx] = data [--size];
+
+ return v;
+ }
+};
+
+/////////////////////////////////////////////////////////////////////////////
+
// 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);
@@ -710,7 +811,7 @@
// like v?sprintf, but returns a "static" buffer
char *vformat (const char *format, va_list ap);
-char *format (const char *format, ...);
+char *format (const char *format, ...) ecb_attribute ((format (printf, 1, 2)));
// safety-check player input which will become object->msg
bool msg_is_safe (const char *msg);