ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
(Generate patch)

Comparing libecb/ecb.h (file contents):
Revision 1.58 by root, Wed Aug 3 15:40:18 2011 UTC vs.
Revision 1.71 by root, Sat Dec 10 11:59:19 2011 UTC

… …
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
… …
67 67
68/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 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 */ 69/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
70 70
71#if ECB_NO_THREADS || ECB_NO_SMP 71#if ECB_NO_THREADS || ECB_NO_SMP
72 #define ECB_MEMORY_FENCE do { } while (0) 72 #define ECB_MEMORY_FENCE do { } while (0)
73 #define ECB_MEMORY_FENCE_ACQUIRE do { } while (0)
74 #define ECB_MEMORY_FENCE_RELEASE do { } while (0)
75#endif 73#endif
76 74
77#ifndef ECB_MEMORY_FENCE 75#ifndef ECB_MEMORY_FENCE
78 #if ECB_GCC_VERSION(2,5) 76 #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__)
79 #if __x86 77 #if __i386__
80 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 78 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
81 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ 79 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */
82 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ 80 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
83 #elif __amd64 81 #elif __amd64
84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
85 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 83 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")
86 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ 84 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
85 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
86 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
87 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \
88 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__)
89 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
90 #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \
91 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ )
92 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
87 #endif 93 #endif
88 #endif 94 #endif
89#endif 95#endif
90 96
91#ifndef ECB_MEMORY_FENCE 97#ifndef ECB_MEMORY_FENCE
92 #if ECB_GCC_VERSION(4,4) 98 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__)
93 #define ECB_MEMORY_FENCE __sync_synchronize () 99 #define ECB_MEMORY_FENCE __sync_synchronize ()
94 #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) 100 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
95 #define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) 101 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
96 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 102 #elif _MSC_VER >= 1400 /* VC++ 2005 */
97 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 103 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
98 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 104 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
99 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 105 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
100 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 106 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
101 #elif defined(_WIN32) 107 #elif defined(_WIN32)
102 #include <WinNT.h> 108 #include <WinNT.h>
103 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 109 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
104 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
105 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
106 #endif 110 #endif
107#endif 111#endif
108 112
109#ifndef ECB_MEMORY_FENCE 113#ifndef ECB_MEMORY_FENCE
114 #if !ECB_AVOID_PTHREADS
110 /* 115 /*
111 * if you get undefined symbol references to pthread_mutex_lock, 116 * if you get undefined symbol references to pthread_mutex_lock,
112 * or failure to find pthread.h, then you should implement 117 * or failure to find pthread.h, then you should implement
113 * the ECB_MEMORY_FENCE operations for your cpu/compiler 118 * the ECB_MEMORY_FENCE operations for your cpu/compiler
114 * OR proide pthread.h and link against the posix thread library 119 * OR provide pthread.h and link against the posix thread library
115 * of your system. 120 * of your system.
116 */ 121 */
117 #include <pthread.h> 122 #include <pthread.h>
123 #define ECB_NEEDS_PTHREADS 1
124 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
118 125
119 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 126 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
120 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) 127 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
128 #endif
129#endif
130
131#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE)
121 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 132 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
133#endif
134
135#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE)
122 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 136 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
123#endif 137#endif
124 138
125/*****************************************************************************/ 139/*****************************************************************************/
126 140
… …
275 289
276 return r + ecb_ld32 (x); 290 return r + ecb_ld32 (x);
277 } 291 }
278#endif 292#endif
279 293
294ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
295ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
296{
297 return ( (x * 0x0802U & 0x22110U)
298 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
299}
300
301ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
302ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
303{
304 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
305 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
306 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
307 x = ( x >> 8 ) | ( x << 8);
308
309 return x;
310}
311
312ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
313ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
314{
315 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
316 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
317 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
318 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
319 x = ( x >> 16 ) | ( x << 16);
320
321 return x;
322}
323
280/* popcount64 is only available on 64 bit cpus as gcc builtin */ 324/* popcount64 is only available on 64 bit cpus as gcc builtin */
281/* so for this version we are lazy */ 325/* so for this version we are lazy */
282ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 326ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
283ecb_function_ int 327ecb_function_ int
284ecb_popcount64 (uint64_t x) 328ecb_popcount64 (uint64_t x)
… …
359 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 403 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
360#else 404#else
361 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 405 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
362#endif 406#endif
363 407
408#if __cplusplus
409 template<typename T>
410 static inline T ecb_div_rd (T val, T div)
411 {
412 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
413 }
414 template<typename T>
415 static inline T ecb_div_ru (T val, T div)
416 {
417 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
418 }
419#else
420 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
421 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
422#endif
423
364#if ecb_cplusplus_does_not_suck 424#if ecb_cplusplus_does_not_suck
365 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 425 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
366 template<typename T, int N> 426 template<typename T, int N>
367 static inline int ecb_array_length (const T (&arr)[N]) 427 static inline int ecb_array_length (const T (&arr)[N])
368 { 428 {

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines