… | |
… | |
28 | |
28 | |
29 | #if __GNUC__ >= 3 |
29 | #if __GNUC__ >= 3 |
30 | # define is_constant(c) __builtin_constant_p (c) |
30 | # define is_constant(c) __builtin_constant_p (c) |
31 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
31 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
32 | # define prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
32 | # define prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
|
|
33 | # define noinline __attribute__((__noinline__)) |
33 | #else |
34 | #else |
34 | # define is_constant(c) 0 |
35 | # define is_constant(c) 0 |
35 | # define expect(expr,value) (expr) |
36 | # define expect(expr,value) (expr) |
36 | # define prefetch(addr,rw,locality) |
37 | # define prefetch(addr,rw,locality) |
|
|
38 | # define noinline |
37 | #endif |
39 | #endif |
38 | |
40 | |
39 | #if __GNUC__ < 4 || (__GNUC__ == 4 || __GNUC_MINOR__ < 4) |
41 | #if __GNUC__ < 4 || (__GNUC__ == 4 || __GNUC_MINOR__ < 4) |
40 | # define decltype(x) typeof(x) |
42 | # define decltype(x) typeof(x) |
41 | #endif |
43 | #endif |
42 | |
44 | |
43 | // put into ifs if you are very sure that the expression |
45 | // put into ifs if you are very sure that the expression |
44 | // is mostly true or mosty false. note that these return |
46 | // is mostly true or mosty false. note that these return |
45 | // booleans, not the expression. |
47 | // booleans, not the expression. |
46 | #define expect_false(expr) expect ((expr) != 0, 0) |
48 | #define expect_false(expr) expect ((expr) ? 1 : 0, 0) |
47 | #define expect_true(expr) expect ((expr) != 0, 1) |
49 | #define expect_true(expr) expect ((expr) ? 1 : 0, 1) |
48 | |
50 | |
49 | #include <pthread.h> |
51 | #include <pthread.h> |
50 | |
52 | |
51 | #include <cstddef> |
53 | #include <cstddef> |
52 | #include <cmath> |
54 | #include <cmath> |
… | |
… | |
72 | #endif |
74 | #endif |
73 | |
75 | |
74 | // use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever) |
76 | // use C0X decltype for auto declarations until ISO C++ sanctifies them (if ever) |
75 | #define auto(var,expr) decltype(expr) var = (expr) |
77 | #define auto(var,expr) decltype(expr) var = (expr) |
76 | |
78 | |
77 | // very ugly macro that basicaly declares and initialises a variable |
79 | // very ugly macro that basically declares and initialises a variable |
78 | // that is in scope for the next statement only |
80 | // that is in scope for the next statement only |
79 | // works only for stuff that can be assigned 0 and converts to false |
81 | // works only for stuff that can be assigned 0 and converts to false |
80 | // (note: works great for pointers) |
82 | // (note: works great for pointers) |
81 | // most ugly macro I ever wrote |
83 | // most ugly macro I ever wrote |
82 | #define statementvar(type, name, value) if (type name = 0) { } else if (((name) = (value)), 1) |
84 | #define statementvar(type, name, value) if (type name = 0) { } else if (((name) = (value)), 1) |
… | |
… | |
96 | // as a is often a constant while b is the variable. it is still a bug, though. |
98 | // as a is often a constant while b is the variable. it is still a bug, though. |
97 | template<typename T, typename U> static inline T min (T a, U b) { return (U)a < b ? (U)a : b; } |
99 | template<typename T, typename U> static inline T min (T a, U b) { return (U)a < b ? (U)a : b; } |
98 | template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; } |
100 | template<typename T, typename U> static inline T max (T a, U b) { return (U)a > b ? (U)a : b; } |
99 | 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; } |
101 | 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; } |
100 | |
102 | |
101 | template<typename T> static inline void min_it (T &v, T m) { v = min (v, m); } |
103 | template<typename T, typename U> static inline void min_it (T &v, U m) { v = min (v, (T)m); } |
102 | template<typename T> static inline void max_it (T &v, T m) { v = max (v, m); } |
104 | template<typename T, typename U> static inline void max_it (T &v, U m) { v = max (v, (T)m); } |
103 | template<typename T> static inline void clamp_it (T &v, T a, T b) { v = clamp (v, a, b); } |
105 | 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); } |
104 | |
106 | |
105 | template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; } |
107 | template<typename T, typename U> static inline void swap (T& a, U& b) { T t=a; a=(T)b; b=(U)t; } |
106 | |
108 | |
107 | template<typename T, typename U, typename V> static inline T min (T a, U b, V c) { return min (a, min (b, c)); } |
109 | template<typename T, typename U, typename V> static inline T min (T a, U b, V c) { return min (a, min (b, c)); } |
108 | template<typename T, typename U, typename V> static inline T max (T a, U b, V c) { return max (a, max (b, c)); } |
110 | template<typename T, typename U, typename V> static inline T max (T a, U b, V c) { return max (a, max (b, c)); } |
|
|
111 | |
|
|
112 | // sign returns -1 or +1 |
|
|
113 | template<typename T> |
|
|
114 | static inline T sign (T v) { return v < 0 ? -1 : +1; } |
|
|
115 | // relies on 2c representation |
|
|
116 | template<> |
|
|
117 | inline sint8 sign (sint8 v) { return 1 - (sint8 (uint8 (v) >> 7) * 2); } |
|
|
118 | |
|
|
119 | // sign0 returns -1, 0 or +1 |
|
|
120 | template<typename T> |
|
|
121 | static inline T sign0 (T v) { return v ? sign (v) : 0; } |
109 | |
122 | |
110 | // div, with correct rounding (< 0.5 downwards, >=0.5 upwards) |
123 | // div, with correct rounding (< 0.5 downwards, >=0.5 upwards) |
111 | template<typename T> static inline T div (T val, T div) { return (val + div / 2) / div; } |
124 | template<typename T> static inline T div (T val, T div) { return (val + div / 2) / div; } |
112 | // div, round-up |
125 | // div, round-up |
113 | template<typename T> static inline T div_ru (T val, T div) { return (val + div - 1) / div; } |
126 | template<typename T> static inline T div_ru (T val, T div) { return (val + div - 1) / div; } |
… | |
… | |
378 | // P. L'Ecuyer, “Maximally Equidistributed Combined Tausworthe Generators”, Mathematics of Computation, 65, 213 (1996), 203–213. |
391 | // P. L'Ecuyer, “Maximally Equidistributed Combined Tausworthe Generators”, Mathematics of Computation, 65, 213 (1996), 203–213. |
379 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps |
392 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps |
380 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps |
393 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps |
381 | struct tausworthe_random_generator |
394 | struct tausworthe_random_generator |
382 | { |
395 | { |
383 | // generator |
|
|
384 | uint32_t state [4]; |
396 | uint32_t state [4]; |
385 | |
397 | |
386 | void operator =(const tausworthe_random_generator &src) |
398 | void operator =(const tausworthe_random_generator &src) |
387 | { |
399 | { |
388 | state [0] = src.state [0]; |
400 | state [0] = src.state [0]; |
… | |
… | |
391 | state [3] = src.state [3]; |
403 | state [3] = src.state [3]; |
392 | } |
404 | } |
393 | |
405 | |
394 | void seed (uint32_t seed); |
406 | void seed (uint32_t seed); |
395 | uint32_t next (); |
407 | uint32_t next (); |
|
|
408 | }; |
396 | |
409 | |
|
|
410 | // Xorshift RNGs, George Marsaglia |
|
|
411 | // http://www.jstatsoft.org/v08/i14/paper |
|
|
412 | // this one is about 40% faster than the tausworthe one above (i.e. not much), |
|
|
413 | // despite the inlining, and has the issue of only creating 2**32-1 numbers. |
|
|
414 | struct xorshift_random_generator |
|
|
415 | { |
|
|
416 | uint32_t x, y; |
|
|
417 | |
|
|
418 | void operator =(const xorshift_random_generator &src) |
|
|
419 | { |
|
|
420 | x = src.x; |
|
|
421 | y = src.y; |
|
|
422 | } |
|
|
423 | |
|
|
424 | void seed (uint32_t seed) |
|
|
425 | { |
|
|
426 | x = seed; |
|
|
427 | y = seed * 69069U; |
|
|
428 | } |
|
|
429 | |
|
|
430 | uint32_t next () |
|
|
431 | { |
|
|
432 | uint32_t t = x ^ (x << 10); |
|
|
433 | x = y; |
|
|
434 | y = y ^ (y >> 13) ^ t ^ (t >> 10); |
|
|
435 | return y; |
|
|
436 | } |
|
|
437 | }; |
|
|
438 | |
|
|
439 | template<class generator> |
|
|
440 | struct random_number_generator : generator |
|
|
441 | { |
397 | // uniform distribution, 0 .. max (0, num - 1) |
442 | // uniform distribution, 0 .. max (0, num - 1) |
398 | uint32_t operator ()(uint32_t num) |
443 | uint32_t operator ()(uint32_t num) |
399 | { |
444 | { |
400 | return is_constant (num) |
445 | return !is_constant (num) ? get_range (num) // non-constant |
401 | ? (next () * (uint64_t)num) >> 32U |
446 | : num & (num - 1) ? (this->next () * (uint64_t)num) >> 32U // constant, non-power-of-two |
402 | : get_range (num); |
447 | : this->next () & (num - 1); // constant, power-of-two |
403 | } |
448 | } |
404 | |
449 | |
405 | // return a number within (min .. max) |
450 | // return a number within (min .. max) |
406 | int operator () (int r_min, int r_max) |
451 | int operator () (int r_min, int r_max) |
407 | { |
452 | { |
… | |
… | |
418 | protected: |
463 | protected: |
419 | uint32_t get_range (uint32_t r_max); |
464 | uint32_t get_range (uint32_t r_max); |
420 | int get_range (int r_min, int r_max); |
465 | int get_range (int r_min, int r_max); |
421 | }; |
466 | }; |
422 | |
467 | |
423 | typedef tausworthe_random_generator rand_gen; |
468 | typedef random_number_generator<tausworthe_random_generator> rand_gen; |
424 | |
469 | |
425 | extern rand_gen rndm, rmg_rndm; |
470 | extern rand_gen rndm, rmg_rndm; |
426 | |
471 | |
427 | INTERFACE_CLASS (attachable) |
472 | INTERFACE_CLASS (attachable) |
428 | struct refcnt_base |
473 | struct refcnt_base |
… | |
… | |
497 | |
542 | |
498 | struct str_hash |
543 | struct str_hash |
499 | { |
544 | { |
500 | std::size_t operator ()(const char *s) const |
545 | std::size_t operator ()(const char *s) const |
501 | { |
546 | { |
502 | unsigned long hash = 0; |
547 | #if 0 |
|
|
548 | uint32_t hash = 0; |
503 | |
549 | |
504 | /* use the one-at-a-time hash function, which supposedly is |
550 | /* use the one-at-a-time hash function, which supposedly is |
505 | * better than the djb2-like one used by perl5.005, but |
551 | * better than the djb2-like one used by perl5.005, but |
506 | * certainly is better then the bug used here before. |
552 | * certainly is better then the bug used here before. |
507 | * see http://burtleburtle.net/bob/hash/doobs.html |
553 | * see http://burtleburtle.net/bob/hash/doobs.html |
… | |
… | |
514 | } |
560 | } |
515 | |
561 | |
516 | hash += hash << 3; |
562 | hash += hash << 3; |
517 | hash ^= hash >> 11; |
563 | hash ^= hash >> 11; |
518 | hash += hash << 15; |
564 | hash += hash << 15; |
|
|
565 | #else |
|
|
566 | // use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/) |
|
|
567 | // it is about twice as fast as the one-at-a-time one, |
|
|
568 | // with good distribution. |
|
|
569 | // FNV-1a is faster on many cpus because the multiplication |
|
|
570 | // runs concurrent with the looping logic. |
|
|
571 | uint32_t hash = 2166136261; |
|
|
572 | |
|
|
573 | while (*s) |
|
|
574 | hash = (hash ^ *s++) * 16777619; |
|
|
575 | #endif |
519 | |
576 | |
520 | return hash; |
577 | return hash; |
521 | } |
578 | } |
522 | }; |
579 | }; |
523 | |
580 | |