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: |
… | |
… | |
48 | #include <inttypes.h> |
48 | #include <inttypes.h> |
49 | #endif |
49 | #endif |
50 | |
50 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
52 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enourmous grief in return for some better fake benchmark numbers. |
53 | * causing enormous grief in return for some better fake benchmark numbers. |
54 | * or so. |
54 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
55 | * 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. |
56 | * an issue with that they should have done it right in the first place. |
57 | */ |
57 | */ |
58 | #ifndef ECB_GCC_VERSION |
58 | #ifndef ECB_GCC_VERSION |
… | |
… | |
63 | #endif |
63 | #endif |
64 | #endif |
64 | #endif |
65 | |
65 | |
66 | /*****************************************************************************/ |
66 | /*****************************************************************************/ |
67 | |
67 | |
|
|
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
|
|
69 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
|
|
70 | |
|
|
71 | #if ECB_NO_THREADS |
|
|
72 | # define ECB_NO_SMP 1 |
|
|
73 | #endif |
|
|
74 | |
|
|
75 | #if ECB_NO_THREADS || ECB_NO_SMP |
|
|
76 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
77 | #endif |
|
|
78 | |
68 | #ifndef ECB_MEMORY_FENCE |
79 | #ifndef ECB_MEMORY_FENCE |
69 | #if ECB_GCC_VERSION(2,5) |
80 | #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
70 | #if __x86 |
81 | #if __i386 || __i386__ |
71 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
72 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
83 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
73 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE /* better be safe than sorry */ |
84 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
74 | #elif __amd64 |
85 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
75 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
76 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
87 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
|
|
88 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
|
|
89 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
77 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") |
90 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
91 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
|
|
92 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) |
|
|
93 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
|
|
94 | #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
|
|
95 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
|
|
96 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
97 | #elif __sparc || __sparc__ |
|
|
98 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
|
|
99 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
100 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
101 | #elif defined(__s390__) || defined(__s390x__) |
|
|
102 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
103 | #elif defined(__mips__) |
|
|
104 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
78 | #endif |
105 | #endif |
79 | #endif |
106 | #endif |
80 | #endif |
107 | #endif |
81 | |
108 | |
82 | #ifndef ECB_MEMORY_FENCE |
109 | #ifndef ECB_MEMORY_FENCE |
83 | #if ECB_GCC_VERSION(4,4) |
110 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__) |
84 | #define ECB_MEMORY_FENCE __sync_synchronize () |
111 | #define ECB_MEMORY_FENCE __sync_synchronize () |
85 | #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) |
112 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
86 | #define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) |
113 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
87 | #elif _MSC_VER >= 1400 |
114 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
88 | #define ECB_MEMORY_FENCE do { } while (0) |
115 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
89 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
|
|
90 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
|
|
91 | #elif defined(_WIN32) && defined(MemoryBarrier) |
|
|
92 | #define ECB_MEMORY_FENCE MemoryBarrier () |
116 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
93 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
117 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
94 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
118 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
|
|
119 | #elif defined(_WIN32) |
|
|
120 | #include <WinNT.h> |
|
|
121 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
122 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
123 | #include <mbarrier.h> |
|
|
124 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
125 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
126 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
127 | #elif __xlC__ |
|
|
128 | #define ECB_MEMORY_FENCE __sync () |
95 | #endif |
129 | #endif |
96 | #endif |
130 | #endif |
97 | |
131 | |
98 | #ifndef ECB_MEMORY_FENCE |
132 | #ifndef ECB_MEMORY_FENCE |
|
|
133 | #if !ECB_AVOID_PTHREADS |
|
|
134 | /* |
|
|
135 | * if you get undefined symbol references to pthread_mutex_lock, |
|
|
136 | * or failure to find pthread.h, then you should implement |
|
|
137 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
|
|
138 | * OR provide pthread.h and link against the posix thread library |
|
|
139 | * of your system. |
|
|
140 | */ |
99 | #include <pthread.h> |
141 | #include <pthread.h> |
|
|
142 | #define ECB_NEEDS_PTHREADS 1 |
|
|
143 | #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
100 | |
144 | |
101 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
145 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
102 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
146 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
|
|
147 | #endif |
|
|
148 | #endif |
|
|
149 | |
|
|
150 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
103 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
151 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
|
|
152 | #endif |
|
|
153 | |
|
|
154 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
104 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
155 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
105 | #endif |
156 | #endif |
106 | |
157 | |
107 | /*****************************************************************************/ |
158 | /*****************************************************************************/ |
108 | |
159 | |
… | |
… | |
232 | x *= 0x01010101; |
283 | x *= 0x01010101; |
233 | |
284 | |
234 | return x >> 24; |
285 | return x >> 24; |
235 | } |
286 | } |
236 | |
287 | |
237 | /* you have the choice beetween something with a table lookup, */ |
|
|
238 | /* something using lots of bit arithmetic and a simple loop */ |
|
|
239 | /* we went for the loop */ |
|
|
240 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
288 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
241 | ecb_function_ int ecb_ld32 (uint32_t x) |
289 | ecb_function_ int ecb_ld32 (uint32_t x) |
242 | { |
290 | { |
243 | int r = 0; |
291 | int r = 0; |
244 | |
292 | |
… | |
… | |
259 | if (x >> 32) { x >>= 32; r += 32; } |
307 | if (x >> 32) { x >>= 32; r += 32; } |
260 | |
308 | |
261 | return r + ecb_ld32 (x); |
309 | return r + ecb_ld32 (x); |
262 | } |
310 | } |
263 | #endif |
311 | #endif |
|
|
312 | |
|
|
313 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
314 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
315 | { |
|
|
316 | return ( (x * 0x0802U & 0x22110U) |
|
|
317 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
318 | } |
|
|
319 | |
|
|
320 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
321 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
322 | { |
|
|
323 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
324 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
325 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
326 | x = ( x >> 8 ) | ( x << 8); |
|
|
327 | |
|
|
328 | return x; |
|
|
329 | } |
|
|
330 | |
|
|
331 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
332 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
333 | { |
|
|
334 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
335 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
336 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
337 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
338 | x = ( x >> 16 ) | ( x << 16); |
|
|
339 | |
|
|
340 | return x; |
|
|
341 | } |
264 | |
342 | |
265 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
343 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
266 | /* so for this version we are lazy */ |
344 | /* so for this version we are lazy */ |
267 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
345 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
268 | ecb_function_ int |
346 | ecb_function_ int |
… | |
… | |
318 | |
396 | |
319 | #if ECB_GCC_VERSION(4,5) |
397 | #if ECB_GCC_VERSION(4,5) |
320 | #define ecb_unreachable() __builtin_unreachable () |
398 | #define ecb_unreachable() __builtin_unreachable () |
321 | #else |
399 | #else |
322 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
400 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
323 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
401 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
324 | ecb_function_ void ecb_unreachable (void) { } |
402 | ecb_inline void ecb_unreachable (void) { } |
325 | #endif |
403 | #endif |
326 | |
404 | |
327 | /* try to tell the compiler that some condition is definitely true */ |
405 | /* try to tell the compiler that some condition is definitely true */ |
328 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
406 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
329 | |
407 | |
330 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
408 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
331 | ecb_function_ unsigned char |
409 | ecb_inline unsigned char |
332 | ecb_byteorder_helper (void) |
410 | ecb_byteorder_helper (void) |
333 | { |
411 | { |
334 | const uint32_t u = 0x11223344; |
412 | const uint32_t u = 0x11223344; |
335 | return *(unsigned char *)&u; |
413 | return *(unsigned char *)&u; |
336 | } |
414 | } |
337 | |
415 | |
338 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
416 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
339 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
417 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
340 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
418 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
341 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
419 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
342 | |
420 | |
343 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
421 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
344 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
422 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
345 | #else |
423 | #else |
346 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
424 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
425 | #endif |
|
|
426 | |
|
|
427 | #if __cplusplus |
|
|
428 | template<typename T> |
|
|
429 | static inline T ecb_div_rd (T val, T div) |
|
|
430 | { |
|
|
431 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
432 | } |
|
|
433 | template<typename T> |
|
|
434 | static inline T ecb_div_ru (T val, T div) |
|
|
435 | { |
|
|
436 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
437 | } |
|
|
438 | #else |
|
|
439 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
440 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
347 | #endif |
441 | #endif |
348 | |
442 | |
349 | #if ecb_cplusplus_does_not_suck |
443 | #if ecb_cplusplus_does_not_suck |
350 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
444 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
351 | template<typename T, int N> |
445 | template<typename T, int N> |