… | |
… | |
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)); } |
… | |
… | |
389 | // 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. |
390 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps |
392 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps |
391 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps |
393 | // http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps |
392 | struct tausworthe_random_generator |
394 | struct tausworthe_random_generator |
393 | { |
395 | { |
394 | // generator |
|
|
395 | uint32_t state [4]; |
396 | uint32_t state [4]; |
396 | |
397 | |
397 | void operator =(const tausworthe_random_generator &src) |
398 | void operator =(const tausworthe_random_generator &src) |
398 | { |
399 | { |
399 | state [0] = src.state [0]; |
400 | state [0] = src.state [0]; |
… | |
… | |
402 | state [3] = src.state [3]; |
403 | state [3] = src.state [3]; |
403 | } |
404 | } |
404 | |
405 | |
405 | void seed (uint32_t seed); |
406 | void seed (uint32_t seed); |
406 | uint32_t next (); |
407 | uint32_t next (); |
|
|
408 | }; |
407 | |
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 | // see also http://www.iro.umontreal.ca/~lecuyer/myftp/papers/xorshift.pdf |
|
|
415 | struct xorshift_random_generator |
|
|
416 | { |
|
|
417 | uint32_t x, y; |
|
|
418 | |
|
|
419 | void operator =(const xorshift_random_generator &src) |
|
|
420 | { |
|
|
421 | x = src.x; |
|
|
422 | y = src.y; |
|
|
423 | } |
|
|
424 | |
|
|
425 | void seed (uint32_t seed) |
|
|
426 | { |
|
|
427 | x = seed; |
|
|
428 | y = seed * 69069U; |
|
|
429 | } |
|
|
430 | |
|
|
431 | uint32_t next () |
|
|
432 | { |
|
|
433 | uint32_t t = x ^ (x << 10); |
|
|
434 | x = y; |
|
|
435 | y = y ^ (y >> 13) ^ t ^ (t >> 10); |
|
|
436 | return y; |
|
|
437 | } |
|
|
438 | }; |
|
|
439 | |
|
|
440 | template<class generator> |
|
|
441 | struct random_number_generator : generator |
|
|
442 | { |
408 | // uniform distribution, 0 .. max (0, num - 1) |
443 | // uniform distribution, 0 .. max (0, num - 1) |
409 | uint32_t operator ()(uint32_t num) |
444 | uint32_t operator ()(uint32_t num) |
410 | { |
445 | { |
411 | return is_constant (num) |
446 | return !is_constant (num) ? get_range (num) // non-constant |
412 | ? (next () * (uint64_t)num) >> 32U |
447 | : num & (num - 1) ? (this->next () * (uint64_t)num) >> 32U // constant, non-power-of-two |
413 | : get_range (num); |
448 | : this->next () & (num - 1); // constant, power-of-two |
414 | } |
449 | } |
415 | |
450 | |
416 | // return a number within (min .. max) |
451 | // return a number within (min .. max) |
417 | int operator () (int r_min, int r_max) |
452 | int operator () (int r_min, int r_max) |
418 | { |
453 | { |
… | |
… | |
429 | protected: |
464 | protected: |
430 | uint32_t get_range (uint32_t r_max); |
465 | uint32_t get_range (uint32_t r_max); |
431 | int get_range (int r_min, int r_max); |
466 | int get_range (int r_min, int r_max); |
432 | }; |
467 | }; |
433 | |
468 | |
434 | typedef tausworthe_random_generator rand_gen; |
469 | typedef random_number_generator<tausworthe_random_generator> rand_gen; |
435 | |
470 | |
436 | extern rand_gen rndm, rmg_rndm; |
471 | extern rand_gen rndm, rmg_rndm; |
437 | |
472 | |
438 | INTERFACE_CLASS (attachable) |
473 | INTERFACE_CLASS (attachable) |
439 | struct refcnt_base |
474 | struct refcnt_base |
… | |
… | |
508 | |
543 | |
509 | struct str_hash |
544 | struct str_hash |
510 | { |
545 | { |
511 | std::size_t operator ()(const char *s) const |
546 | std::size_t operator ()(const char *s) const |
512 | { |
547 | { |
513 | unsigned long hash = 0; |
548 | #if 0 |
|
|
549 | uint32_t hash = 0; |
514 | |
550 | |
515 | /* use the one-at-a-time hash function, which supposedly is |
551 | /* use the one-at-a-time hash function, which supposedly is |
516 | * better than the djb2-like one used by perl5.005, but |
552 | * better than the djb2-like one used by perl5.005, but |
517 | * certainly is better then the bug used here before. |
553 | * certainly is better then the bug used here before. |
518 | * see http://burtleburtle.net/bob/hash/doobs.html |
554 | * see http://burtleburtle.net/bob/hash/doobs.html |
… | |
… | |
525 | } |
561 | } |
526 | |
562 | |
527 | hash += hash << 3; |
563 | hash += hash << 3; |
528 | hash ^= hash >> 11; |
564 | hash ^= hash >> 11; |
529 | hash += hash << 15; |
565 | hash += hash << 15; |
|
|
566 | #else |
|
|
567 | // use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/) |
|
|
568 | // it is about twice as fast as the one-at-a-time one, |
|
|
569 | // with good distribution. |
|
|
570 | // FNV-1a is faster on many cpus because the multiplication |
|
|
571 | // runs concurrent with the looping logic. |
|
|
572 | uint32_t hash = 2166136261; |
|
|
573 | |
|
|
574 | while (*s) |
|
|
575 | hash = (hash ^ *s++) * 16777619; |
|
|
576 | #endif |
530 | |
577 | |
531 | return hash; |
578 | return hash; |
532 | } |
579 | } |
533 | }; |
580 | }; |
534 | |
581 | |