1 | /* |
1 | /* |
2 | * This file is part of Deliantra, the Roguelike Realtime MMORPG. |
2 | * This file is part of Deliantra, the Roguelike Realtime MMORPG. |
3 | * |
3 | * |
4 | * Copyright (©) 2005,2006,2007,2008 Marc Alexander Lehmann / Robin Redeker / the Deliantra team |
4 | * Copyright (©) 2005,2006,2007,2008,2009,2010,2011 Marc Alexander Lehmann / Robin Redeker / the Deliantra team |
5 | * |
5 | * |
6 | * Deliantra is free software: you can redistribute it and/or modify it under |
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 |
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 |
8 | * Free Software Foundation, either version 3 of the License, or (at your |
9 | * option) any later version. |
9 | * option) any later version. |
… | |
… | |
55 | #endif |
55 | #endif |
56 | |
56 | |
57 | // use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever) |
57 | // use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever) |
58 | #define auto(var,expr) decltype(expr) var = (expr) |
58 | #define auto(var,expr) decltype(expr) var = (expr) |
59 | |
59 | |
|
|
60 | #if cplusplus_does_not_suck |
|
|
61 | // does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) |
|
|
62 | template<typename T, int N> |
|
|
63 | static inline int array_length (const T (&arr)[N]) |
|
|
64 | { |
|
|
65 | return N; |
|
|
66 | } |
|
|
67 | #else |
|
|
68 | #define array_length(name) (sizeof (name) / sizeof (name [0])) |
|
|
69 | #endif |
|
|
70 | |
60 | // very ugly macro that basically declares and initialises a variable |
71 | // very ugly macro that basically declares and initialises a variable |
61 | // that is in scope for the next statement only |
72 | // that is in scope for the next statement only |
62 | // works only for stuff that can be assigned 0 and converts to false |
73 | // works only for stuff that can be assigned 0 and converts to false |
63 | // (note: works great for pointers) |
74 | // (note: works great for pointers) |
64 | // most ugly macro I ever wrote |
75 | // most ugly macro I ever wrote |
… | |
… | |
93 | // sign returns -1 or +1 |
104 | // sign returns -1 or +1 |
94 | template<typename T> |
105 | template<typename T> |
95 | static inline T sign (T v) { return v < 0 ? -1 : +1; } |
106 | static inline T sign (T v) { return v < 0 ? -1 : +1; } |
96 | // relies on 2c representation |
107 | // relies on 2c representation |
97 | template<> |
108 | template<> |
98 | inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); } |
109 | inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); } |
|
|
110 | template<> |
|
|
111 | inline sint16 sign (sint16 v) { return 1 - (sint16 (uint16 (v) >> 15) * 2); } |
|
|
112 | template<> |
|
|
113 | inline sint32 sign (sint32 v) { return 1 - (sint32 (uint32 (v) >> 31) * 2); } |
99 | |
114 | |
100 | // sign0 returns -1, 0 or +1 |
115 | // sign0 returns -1, 0 or +1 |
101 | template<typename T> |
116 | template<typename T> |
102 | static inline T sign0 (T v) { return v ? sign (v) : 0; } |
117 | static inline T sign0 (T v) { return v ? sign (v) : 0; } |
|
|
118 | |
|
|
119 | //clashes with C++0x |
|
|
120 | template<typename T, typename U> |
|
|
121 | static inline T copysign (T a, U b) { return a > 0 ? b : -b; } |
103 | |
122 | |
104 | // div* only work correctly for div > 0 |
123 | // div* only work correctly for div > 0 |
105 | // div, with correct rounding (< 0.5 downwards, >=0.5 upwards) |
124 | // div, with correct rounding (< 0.5 downwards, >=0.5 upwards) |
106 | template<typename T> static inline T div (T val, T div) |
125 | template<typename T> static inline T div (T val, T div) |
107 | { |
126 | { |
108 | return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div; |
127 | return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div; |
109 | } |
128 | } |
|
|
129 | |
|
|
130 | template<> inline float div (float val, float div) { return val / div; } |
|
|
131 | template<> inline double div (double val, double div) { return val / div; } |
|
|
132 | |
110 | // div, round-up |
133 | // div, round-up |
111 | template<typename T> static inline T div_ru (T val, T div) |
134 | template<typename T> static inline T div_ru (T val, T div) |
112 | { |
135 | { |
113 | return expect_false (val < 0) ? - ((-val ) / div) : (val + div - 1) / div; |
136 | return expect_false (val < 0) ? - ((-val ) / div) : (val + div - 1) / div; |
114 | } |
137 | } |
… | |
… | |
186 | int32_t d = b - a; |
209 | int32_t d = b - a; |
187 | d &= d >> 31; |
210 | d &= d >> 31; |
188 | return b - d; |
211 | return b - d; |
189 | } |
212 | } |
190 | |
213 | |
191 | // this is much faster than crossfires original algorithm |
214 | // this is much faster than crossfire's original algorithm |
192 | // on modern cpus |
215 | // on modern cpus |
193 | inline int |
216 | inline int |
194 | isqrt (int n) |
217 | isqrt (int n) |
195 | { |
218 | { |
196 | return (int)sqrtf ((float)n); |
219 | return (int)sqrtf ((float)n); |
… | |
… | |
229 | #else |
252 | #else |
230 | return dx_ + dy_ - min (dx_, dy_) * 5 / 8; |
253 | return dx_ + dy_ - min (dx_, dy_) * 5 / 8; |
231 | #endif |
254 | #endif |
232 | } |
255 | } |
233 | |
256 | |
|
|
257 | // can be substantially faster than floor, if your value range allows for it |
|
|
258 | template<typename T> |
|
|
259 | inline T |
|
|
260 | fastfloor (T x) |
|
|
261 | { |
|
|
262 | return std::floor (x); |
|
|
263 | } |
|
|
264 | |
|
|
265 | inline float |
|
|
266 | fastfloor (float x) |
|
|
267 | { |
|
|
268 | return sint32(x) - (x < 0); |
|
|
269 | } |
|
|
270 | |
|
|
271 | inline double |
|
|
272 | fastfloor (double x) |
|
|
273 | { |
|
|
274 | return sint64(x) - (x < 0); |
|
|
275 | } |
|
|
276 | |
234 | /* |
277 | /* |
235 | * absdir(int): Returns a number between 1 and 8, which represent |
278 | * absdir(int): Returns a number between 1 and 8, which represent |
236 | * the "absolute" direction of a number (it actually takes care of |
279 | * the "absolute" direction of a number (it actually takes care of |
237 | * "overflow" in previous calculations of a direction). |
280 | * "overflow" in previous calculations of a direction). |
238 | */ |
281 | */ |
… | |
… | |
240 | absdir (int d) |
283 | absdir (int d) |
241 | { |
284 | { |
242 | return ((d - 1) & 7) + 1; |
285 | return ((d - 1) & 7) + 1; |
243 | } |
286 | } |
244 | |
287 | |
|
|
288 | // avoid ctz name because netbsd or freebsd spams it's namespace with it |
|
|
289 | #if GCC_VERSION(3,4) |
|
|
290 | static inline int least_significant_bit (uint32_t x) |
|
|
291 | { |
|
|
292 | return __builtin_ctz (x); |
|
|
293 | } |
|
|
294 | #else |
|
|
295 | int least_significant_bit (uint32_t x); |
|
|
296 | #endif |
|
|
297 | |
|
|
298 | #define for_all_bits_sparse_32(mask, idxvar) \ |
|
|
299 | for (uint32_t idxvar, mask_ = mask; \ |
|
|
300 | mask_ && ((idxvar = least_significant_bit (mask_)), mask_ &= ~(1 << idxvar), 1);) |
|
|
301 | |
245 | extern ssize_t slice_alloc; // statistics |
302 | extern ssize_t slice_alloc; // statistics |
246 | |
303 | |
247 | void *salloc_ (int n) throw (std::bad_alloc); |
304 | void *salloc_ (int n) throw (std::bad_alloc); |
248 | void *salloc_ (int n, void *src) throw (std::bad_alloc); |
305 | void *salloc_ (int n, void *src) throw (std::bad_alloc); |
249 | |
306 | |
… | |
… | |
393 | { |
450 | { |
394 | p->~Tp (); |
451 | p->~Tp (); |
395 | } |
452 | } |
396 | }; |
453 | }; |
397 | |
454 | |
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) |
455 | INTERFACE_CLASS (attachable) |
481 | struct refcnt_base |
456 | struct refcnt_base |
482 | { |
457 | { |
483 | typedef int refcnt_t; |
458 | typedef int refcnt_t; |
484 | mutable refcnt_t ACC (RW, refcnt); |
459 | mutable refcnt_t ACC (RW, refcnt); |
… | |
… | |
545 | typedef refptr<maptile> maptile_ptr; |
520 | typedef refptr<maptile> maptile_ptr; |
546 | typedef refptr<object> object_ptr; |
521 | typedef refptr<object> object_ptr; |
547 | typedef refptr<archetype> arch_ptr; |
522 | typedef refptr<archetype> arch_ptr; |
548 | typedef refptr<client> client_ptr; |
523 | typedef refptr<client> client_ptr; |
549 | typedef refptr<player> player_ptr; |
524 | typedef refptr<player> player_ptr; |
|
|
525 | typedef refptr<region> region_ptr; |
|
|
526 | |
|
|
527 | #define STRHSH_NULL 2166136261 |
|
|
528 | |
|
|
529 | static inline uint32_t |
|
|
530 | strhsh (const char *s) |
|
|
531 | { |
|
|
532 | // use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/) |
|
|
533 | // it is about twice as fast as the one-at-a-time one, |
|
|
534 | // with good distribution. |
|
|
535 | // FNV-1a is faster on many cpus because the multiplication |
|
|
536 | // runs concurrently with the looping logic. |
|
|
537 | // we modify the hash a bit to improve its distribution |
|
|
538 | uint32_t hash = STRHSH_NULL; |
|
|
539 | |
|
|
540 | while (*s) |
|
|
541 | hash = (hash ^ *s++) * 16777619U; |
|
|
542 | |
|
|
543 | return hash ^ (hash >> 16); |
|
|
544 | } |
|
|
545 | |
|
|
546 | static inline uint32_t |
|
|
547 | memhsh (const char *s, size_t len) |
|
|
548 | { |
|
|
549 | uint32_t hash = STRHSH_NULL; |
|
|
550 | |
|
|
551 | while (len--) |
|
|
552 | hash = (hash ^ *s++) * 16777619U; |
|
|
553 | |
|
|
554 | return hash; |
|
|
555 | } |
550 | |
556 | |
551 | struct str_hash |
557 | struct str_hash |
552 | { |
558 | { |
553 | std::size_t operator ()(const char *s) const |
559 | std::size_t operator ()(const char *s) const |
554 | { |
560 | { |
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; |
561 | return strhsh (s); |
|
|
562 | } |
|
|
563 | |
|
|
564 | std::size_t operator ()(const shstr &s) const |
|
|
565 | { |
|
|
566 | return strhsh (s); |
586 | } |
567 | } |
587 | }; |
568 | }; |
588 | |
569 | |
589 | struct str_equal |
570 | struct str_equal |
590 | { |
571 | { |
… | |
… | |
682 | { |
663 | { |
683 | erase (&obj); |
664 | erase (&obj); |
684 | } |
665 | } |
685 | }; |
666 | }; |
686 | |
667 | |
|
|
668 | ///////////////////////////////////////////////////////////////////////////// |
|
|
669 | |
|
|
670 | // something like a vector or stack, but without |
|
|
671 | // out of bounds checking |
|
|
672 | template<typename T> |
|
|
673 | struct fixed_stack |
|
|
674 | { |
|
|
675 | T *data; |
|
|
676 | int size; |
|
|
677 | int max; |
|
|
678 | |
|
|
679 | fixed_stack () |
|
|
680 | : size (0), data (0) |
|
|
681 | { |
|
|
682 | } |
|
|
683 | |
|
|
684 | fixed_stack (int max) |
|
|
685 | : size (0), max (max) |
|
|
686 | { |
|
|
687 | data = salloc<T> (max); |
|
|
688 | } |
|
|
689 | |
|
|
690 | void reset (int new_max) |
|
|
691 | { |
|
|
692 | sfree (data, max); |
|
|
693 | size = 0; |
|
|
694 | max = new_max; |
|
|
695 | data = salloc<T> (max); |
|
|
696 | } |
|
|
697 | |
|
|
698 | void free () |
|
|
699 | { |
|
|
700 | sfree (data, max); |
|
|
701 | data = 0; |
|
|
702 | } |
|
|
703 | |
|
|
704 | ~fixed_stack () |
|
|
705 | { |
|
|
706 | sfree (data, max); |
|
|
707 | } |
|
|
708 | |
|
|
709 | T &operator[](int idx) |
|
|
710 | { |
|
|
711 | return data [idx]; |
|
|
712 | } |
|
|
713 | |
|
|
714 | void push (T v) |
|
|
715 | { |
|
|
716 | data [size++] = v; |
|
|
717 | } |
|
|
718 | |
|
|
719 | T &pop () |
|
|
720 | { |
|
|
721 | return data [--size]; |
|
|
722 | } |
|
|
723 | |
|
|
724 | T remove (int idx) |
|
|
725 | { |
|
|
726 | T v = data [idx]; |
|
|
727 | |
|
|
728 | data [idx] = data [--size]; |
|
|
729 | |
|
|
730 | return v; |
|
|
731 | } |
|
|
732 | }; |
|
|
733 | |
|
|
734 | ///////////////////////////////////////////////////////////////////////////// |
|
|
735 | |
687 | // basically does what strncpy should do, but appends "..." to strings exceeding length |
736 | // basically does what strncpy should do, but appends "..." to strings exceeding length |
688 | // returns the number of bytes actually used (including \0) |
737 | // returns the number of bytes actually used (including \0) |
689 | int assign (char *dst, const char *src, int maxsize); |
738 | int assign (char *dst, const char *src, int maxsize); |
690 | |
739 | |
691 | // type-safe version of assign |
740 | // type-safe version of assign |