1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2011 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
6 | * All rights reserved. |
6 | * All rights reserved. |
7 | * |
7 | * |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
9 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
… | |
… | |
28 | */ |
28 | */ |
29 | |
29 | |
30 | #ifndef ECB_H |
30 | #ifndef ECB_H |
31 | #define ECB_H |
31 | #define ECB_H |
32 | |
32 | |
|
|
33 | /* 16 bits major, 16 bits minor */ |
|
|
34 | #define ECB_VERSION 0x00010001 |
|
|
35 | |
33 | #ifdef _WIN32 |
36 | #ifdef _WIN32 |
34 | typedef signed char int8_t; |
37 | typedef signed char int8_t; |
35 | typedef unsigned char uint8_t; |
38 | typedef unsigned char uint8_t; |
36 | typedef signed short int16_t; |
39 | typedef signed short int16_t; |
37 | typedef unsigned short uint16_t; |
40 | typedef unsigned short uint16_t; |
… | |
… | |
42 | typedef unsigned long long uint64_t; |
45 | typedef unsigned long long uint64_t; |
43 | #else /* _MSC_VER || __BORLANDC__ */ |
46 | #else /* _MSC_VER || __BORLANDC__ */ |
44 | typedef signed __int64 int64_t; |
47 | typedef signed __int64 int64_t; |
45 | typedef unsigned __int64 uint64_t; |
48 | typedef unsigned __int64 uint64_t; |
46 | #endif |
49 | #endif |
|
|
50 | #ifdef _WIN64 |
|
|
51 | #define ECB_PTRSIZE 8 |
|
|
52 | typedef uint64_t uintptr_t; |
|
|
53 | typedef int64_t intptr_t; |
|
|
54 | #else |
|
|
55 | #define ECB_PTRSIZE 4 |
|
|
56 | typedef uint32_t uintptr_t; |
|
|
57 | typedef int32_t intptr_t; |
|
|
58 | #endif |
|
|
59 | typedef intptr_t ptrdiff_t; |
47 | #else |
60 | #else |
48 | #include <inttypes.h> |
61 | #include <inttypes.h> |
|
|
62 | #if UINTMAX_MAX > 0xffffffffU |
|
|
63 | #define ECB_PTRSIZE 8 |
|
|
64 | #else |
|
|
65 | #define ECB_PTRSIZE 4 |
|
|
66 | #endif |
49 | #endif |
67 | #endif |
50 | |
68 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
69 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
70 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enourmous grief in return for some better fake benchmark numbers. |
71 | * causing enormous grief in return for some better fake benchmark numbers. |
54 | * or so. |
72 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
73 | * we try to detect these and simply assume they are not gcc - if they have |
56 | * an issue with that they should have done it right in the first place. |
74 | * an issue with that they should have done it right in the first place. |
57 | */ |
75 | */ |
58 | #ifndef ECB_GCC_VERSION |
76 | #ifndef ECB_GCC_VERSION |
59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
77 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
60 | #define ECB_GCC_VERSION(major,minor) 0 |
78 | #define ECB_GCC_VERSION(major,minor) 0 |
61 | #else |
79 | #else |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
80 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
63 | #endif |
81 | #endif |
64 | #endif |
82 | #endif |
|
|
83 | |
|
|
84 | /*****************************************************************************/ |
|
|
85 | |
|
|
86 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
|
|
87 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
|
|
88 | |
|
|
89 | #if ECB_NO_THREADS |
|
|
90 | # define ECB_NO_SMP 1 |
|
|
91 | #endif |
|
|
92 | |
|
|
93 | #if ECB_NO_THREADS || ECB_NO_SMP |
|
|
94 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
95 | #endif |
|
|
96 | |
|
|
97 | #ifndef ECB_MEMORY_FENCE |
|
|
98 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
99 | #if __i386 || __i386__ |
|
|
100 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
|
|
101 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
|
|
102 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
|
|
103 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
|
|
104 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
|
|
105 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
|
|
106 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
|
|
107 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
|
|
108 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
109 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
|
|
110 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
|
|
111 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
|
|
112 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
|
|
113 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
|
|
114 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
115 | #elif __sparc || __sparc__ |
|
|
116 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
|
|
117 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
118 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
119 | #elif defined __s390__ || defined __s390x__ |
|
|
120 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
121 | #elif defined __mips__ |
|
|
122 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
123 | #elif defined __alpha__ |
|
|
124 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
125 | #endif |
|
|
126 | #endif |
|
|
127 | #endif |
|
|
128 | |
|
|
129 | #ifndef ECB_MEMORY_FENCE |
|
|
130 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
|
|
131 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
132 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
|
|
133 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
|
|
134 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
|
|
135 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
136 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
|
|
137 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
138 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
|
|
139 | #elif defined _WIN32 |
|
|
140 | #include <WinNT.h> |
|
|
141 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
142 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
143 | #include <mbarrier.h> |
|
|
144 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
145 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
146 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
147 | #elif __xlC__ |
|
|
148 | #define ECB_MEMORY_FENCE __sync () |
|
|
149 | #endif |
|
|
150 | #endif |
|
|
151 | |
|
|
152 | #ifndef ECB_MEMORY_FENCE |
|
|
153 | #if !ECB_AVOID_PTHREADS |
|
|
154 | /* |
|
|
155 | * if you get undefined symbol references to pthread_mutex_lock, |
|
|
156 | * or failure to find pthread.h, then you should implement |
|
|
157 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
|
|
158 | * OR provide pthread.h and link against the posix thread library |
|
|
159 | * of your system. |
|
|
160 | */ |
|
|
161 | #include <pthread.h> |
|
|
162 | #define ECB_NEEDS_PTHREADS 1 |
|
|
163 | #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
|
|
164 | |
|
|
165 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
|
|
166 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
|
|
167 | #endif |
|
|
168 | #endif |
|
|
169 | |
|
|
170 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
|
|
171 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
|
|
172 | #endif |
|
|
173 | |
|
|
174 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
|
|
175 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
|
|
176 | #endif |
|
|
177 | |
|
|
178 | /*****************************************************************************/ |
65 | |
179 | |
66 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
180 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
67 | |
181 | |
68 | #if __cplusplus |
182 | #if __cplusplus |
69 | #define ecb_inline static inline |
183 | #define ecb_inline static inline |
… | |
… | |
189 | x *= 0x01010101; |
303 | x *= 0x01010101; |
190 | |
304 | |
191 | return x >> 24; |
305 | return x >> 24; |
192 | } |
306 | } |
193 | |
307 | |
194 | /* you have the choice beetween something with a table lookup, */ |
|
|
195 | /* something using lots of bit arithmetic and a simple loop */ |
|
|
196 | /* we went for the loop */ |
|
|
197 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
308 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
198 | ecb_function_ int ecb_ld32 (uint32_t x) |
309 | ecb_function_ int ecb_ld32 (uint32_t x) |
199 | { |
310 | { |
200 | int r = 0; |
311 | int r = 0; |
201 | |
312 | |
… | |
… | |
216 | if (x >> 32) { x >>= 32; r += 32; } |
327 | if (x >> 32) { x >>= 32; r += 32; } |
217 | |
328 | |
218 | return r + ecb_ld32 (x); |
329 | return r + ecb_ld32 (x); |
219 | } |
330 | } |
220 | #endif |
331 | #endif |
|
|
332 | |
|
|
333 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
334 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
335 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
336 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
337 | |
|
|
338 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
339 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
340 | { |
|
|
341 | return ( (x * 0x0802U & 0x22110U) |
|
|
342 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
343 | } |
|
|
344 | |
|
|
345 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
346 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
347 | { |
|
|
348 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
349 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
350 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
351 | x = ( x >> 8 ) | ( x << 8); |
|
|
352 | |
|
|
353 | return x; |
|
|
354 | } |
|
|
355 | |
|
|
356 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
357 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
358 | { |
|
|
359 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
360 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
361 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
362 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
363 | x = ( x >> 16 ) | ( x << 16); |
|
|
364 | |
|
|
365 | return x; |
|
|
366 | } |
221 | |
367 | |
222 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
368 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
223 | /* so for this version we are lazy */ |
369 | /* so for this version we are lazy */ |
224 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
370 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
225 | ecb_function_ int |
371 | ecb_function_ int |
… | |
… | |
275 | |
421 | |
276 | #if ECB_GCC_VERSION(4,5) |
422 | #if ECB_GCC_VERSION(4,5) |
277 | #define ecb_unreachable() __builtin_unreachable () |
423 | #define ecb_unreachable() __builtin_unreachable () |
278 | #else |
424 | #else |
279 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
425 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
280 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
426 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
281 | ecb_function_ void ecb_unreachable (void) { } |
427 | ecb_inline void ecb_unreachable (void) { } |
282 | #endif |
428 | #endif |
283 | |
429 | |
284 | /* try to tell the compiler that some condition is definitely true */ |
430 | /* try to tell the compiler that some condition is definitely true */ |
285 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
431 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
286 | |
432 | |
287 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
433 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
288 | ecb_function_ unsigned char |
434 | ecb_inline unsigned char |
289 | ecb_byteorder_helper (void) |
435 | ecb_byteorder_helper (void) |
290 | { |
436 | { |
291 | const uint32_t u = 0x11223344; |
437 | const uint32_t u = 0x11223344; |
292 | return *(unsigned char *)&u; |
438 | return *(unsigned char *)&u; |
293 | } |
439 | } |
294 | |
440 | |
295 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
441 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
296 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
442 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
297 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
443 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
298 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
444 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
299 | |
445 | |
300 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
446 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
301 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
447 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
302 | #else |
448 | #else |
303 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
449 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
450 | #endif |
|
|
451 | |
|
|
452 | #if __cplusplus |
|
|
453 | template<typename T> |
|
|
454 | static inline T ecb_div_rd (T val, T div) |
|
|
455 | { |
|
|
456 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
457 | } |
|
|
458 | template<typename T> |
|
|
459 | static inline T ecb_div_ru (T val, T div) |
|
|
460 | { |
|
|
461 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
462 | } |
|
|
463 | #else |
|
|
464 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
465 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
304 | #endif |
466 | #endif |
305 | |
467 | |
306 | #if ecb_cplusplus_does_not_suck |
468 | #if ecb_cplusplus_does_not_suck |
307 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
469 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
308 | template<typename T, int N> |
470 | template<typename T, int N> |