… | |
… | |
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> |
… | |
… | |
116 | |
118 | |
117 | // sign0 returns -1, 0 or +1 |
119 | // sign0 returns -1, 0 or +1 |
118 | template<typename T> |
120 | template<typename T> |
119 | static inline T sign0 (T v) { return v ? sign (v) : 0; } |
121 | static inline T sign0 (T v) { return v ? sign (v) : 0; } |
120 | |
122 | |
|
|
123 | // div* only work correctly for div > 0 |
121 | // div, with correct rounding (< 0.5 downwards, >=0.5 upwards) |
124 | // div, with correct rounding (< 0.5 downwards, >=0.5 upwards) |
122 | template<typename T> static inline T div (T val, T div) { return (val + div / 2) / div; } |
125 | template<typename T> static inline T div (T val, T div) |
|
|
126 | { |
|
|
127 | return expect_false (val < 0) ? - ((-val + (div - 1) / 2) / div) : (val + div / 2) / div; |
|
|
128 | } |
123 | // div, round-up |
129 | // div, round-up |
124 | template<typename T> static inline T div_ru (T val, T div) { return (val + div - 1) / div; } |
130 | template<typename T> static inline T div_ru (T val, T div) |
|
|
131 | { |
|
|
132 | return expect_false (val < 0) ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
133 | } |
125 | // div, round-down |
134 | // div, round-down |
126 | template<typename T> static inline T div_rd (T val, T div) { return (val ) / div; } |
135 | template<typename T> static inline T div_rd (T val, T div) |
|
|
136 | { |
|
|
137 | return expect_false (val < 0) ? - ((-val + (div - 1) ) / div) : (val ) / div; |
|
|
138 | } |
127 | |
139 | |
|
|
140 | // lerp* only work correctly for min_in < max_in |
|
|
141 | // Linear intERPolate, scales val from min_in..max_in to min_out..max_out |
128 | template<typename T> |
142 | template<typename T> |
129 | static inline T |
143 | static inline T |
130 | lerp (T val, T min_in, T max_in, T min_out, T max_out) |
144 | lerp (T val, T min_in, T max_in, T min_out, T max_out) |
131 | { |
145 | { |
132 | return min_out + div <T> ((val - min_in) * (max_out - min_out), max_in - min_in); |
146 | return min_out + div <T> ((val - min_in) * (max_out - min_out), max_in - min_in); |
… | |
… | |
407 | |
421 | |
408 | // Xorshift RNGs, George Marsaglia |
422 | // Xorshift RNGs, George Marsaglia |
409 | // http://www.jstatsoft.org/v08/i14/paper |
423 | // http://www.jstatsoft.org/v08/i14/paper |
410 | // this one is about 40% faster than the tausworthe one above (i.e. not much), |
424 | // this one is about 40% faster than the tausworthe one above (i.e. not much), |
411 | // despite the inlining, and has the issue of only creating 2**32-1 numbers. |
425 | // despite the inlining, and has the issue of only creating 2**32-1 numbers. |
|
|
426 | // see also http://www.iro.umontreal.ca/~lecuyer/myftp/papers/xorshift.pdf |
412 | struct xorshift_random_generator |
427 | struct xorshift_random_generator |
413 | { |
428 | { |
414 | uint32_t x, y; |
429 | uint32_t x, y; |
415 | |
430 | |
416 | void operator =(const xorshift_random_generator &src) |
431 | void operator =(const xorshift_random_generator &src) |
… | |
… | |
540 | |
555 | |
541 | struct str_hash |
556 | struct str_hash |
542 | { |
557 | { |
543 | std::size_t operator ()(const char *s) const |
558 | std::size_t operator ()(const char *s) const |
544 | { |
559 | { |
545 | unsigned long hash = 0; |
560 | #if 0 |
|
|
561 | uint32_t hash = 0; |
546 | |
562 | |
547 | /* use the one-at-a-time hash function, which supposedly is |
563 | /* use the one-at-a-time hash function, which supposedly is |
548 | * better than the djb2-like one used by perl5.005, but |
564 | * better than the djb2-like one used by perl5.005, but |
549 | * certainly is better then the bug used here before. |
565 | * certainly is better then the bug used here before. |
550 | * see http://burtleburtle.net/bob/hash/doobs.html |
566 | * see http://burtleburtle.net/bob/hash/doobs.html |
… | |
… | |
557 | } |
573 | } |
558 | |
574 | |
559 | hash += hash << 3; |
575 | hash += hash << 3; |
560 | hash ^= hash >> 11; |
576 | hash ^= hash >> 11; |
561 | hash += hash << 15; |
577 | hash += hash << 15; |
|
|
578 | #else |
|
|
579 | // use FNV-1a hash (http://isthe.com/chongo/tech/comp/fnv/) |
|
|
580 | // it is about twice as fast as the one-at-a-time one, |
|
|
581 | // with good distribution. |
|
|
582 | // FNV-1a is faster on many cpus because the multiplication |
|
|
583 | // runs concurrent with the looping logic. |
|
|
584 | uint32_t hash = 2166136261; |
|
|
585 | |
|
|
586 | while (*s) |
|
|
587 | hash = (hash ^ *s++) * 16777619; |
|
|
588 | #endif |
562 | |
589 | |
563 | return hash; |
590 | return hash; |
564 | } |
591 | } |
565 | }; |
592 | }; |
566 | |
593 | |
… | |
… | |
661 | erase (&obj); |
688 | erase (&obj); |
662 | } |
689 | } |
663 | }; |
690 | }; |
664 | |
691 | |
665 | // basically does what strncpy should do, but appends "..." to strings exceeding length |
692 | // basically does what strncpy should do, but appends "..." to strings exceeding length |
|
|
693 | // returns the number of bytes actually used (including \0) |
666 | void assign (char *dst, const char *src, int maxlen); |
694 | int assign (char *dst, const char *src, int maxsize); |
667 | |
695 | |
668 | // type-safe version of assign |
696 | // type-safe version of assign |
669 | template<int N> |
697 | template<int N> |
670 | inline void assign (char (&dst)[N], const char *src) |
698 | inline int assign (char (&dst)[N], const char *src) |
671 | { |
699 | { |
672 | assign ((char *)&dst, src, N); |
700 | return assign ((char *)&dst, src, N); |
673 | } |
701 | } |
674 | |
702 | |
675 | typedef double tstamp; |
703 | typedef double tstamp; |
676 | |
704 | |
677 | // return current time as timestamp |
705 | // return current time as timestamp |