--- deliantra/server/include/util.h 2009/10/20 05:57:08 1.92 +++ deliantra/server/include/util.h 2012/10/29 23:55:54 1.120 @@ -1,54 +1,34 @@ /* * This file is part of Deliantra, the Roguelike Realtime MMORPG. - * - * Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team - * + * + * Copyright (©) 2005,2006,2007,2008,2009,2010,2011,2012 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 @@ -77,6 +57,17 @@ // 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 +// 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 @@ -97,8 +88,8 @@ // 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 +106,30 @@ 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 +211,7 @@ 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) @@ -251,6 +254,26 @@ #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 @@ -262,6 +285,20 @@ return ((d - 1) & 7) + 1; } +// avoid ctz name because netbsd or freebsd spams it's namespace with it +#if GCC_VERSION(3,4) +static inline int least_significant_bit (uint32_t x) +{ + return __builtin_ctz (x); +} +#else +int least_significant_bit (uint32_t x); +#endif + +#define for_all_bits_sparse_32(mask, idxvar) \ + for (uint32_t idxvar, mask_ = mask; \ + mask_ && ((idxvar = least_significant_bit (mask_)), mask_ &= ~(1 << idxvar), 1);) + extern ssize_t slice_alloc; // statistics void *salloc_ (int n) throw (std::bad_alloc); @@ -289,7 +326,6 @@ 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 } } @@ -415,88 +451,63 @@ } }; -// 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; + + refcnt_buf (size_t size = 0); + refcnt_buf (void *data, size_t size); - void operator =(const tausworthe_random_generator &src) + 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; + ++_refcnt (); } - 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 (unsigned int) * 2 + }; + + unsigned int &_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) + unsigned int &_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 (unsigned int 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 dec () { - return this->next () / (double)0xFFFFFFFFU; + if (!--_refcnt ()) + sfree (data - overhead, size () + overhead); } - -protected: - uint32_t get_range (uint32_t r_max); - int get_range (int r_min, int r_max); }; -typedef random_number_generator rand_gen; - -extern rand_gen rndm, rmg_rndm; - INTERFACE_CLASS (attachable) struct refcnt_base { @@ -567,42 +578,48 @@ 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); } }; @@ -639,7 +656,7 @@ // 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 @@ -686,7 +703,7 @@ void erase (T *obj) { - unsigned int pos = obj->*indexmember; + object_vector_index pos = obj->*indexmember; obj->*indexmember = 0; if (pos < this->size ()) @@ -704,6 +721,74 @@ } }; +///////////////////////////////////////////////////////////////////////////// + +// 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); @@ -724,7 +809,7 @@ // 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, ...) attribute ((format (printf, 1, 2))); // safety-check player input which will become object->msg bool msg_is_safe (const char *msg);