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

Comparing libecb/ecb.h (file contents):
Revision 1.52 by root, Mon Jul 18 01:18:17 2011 UTC vs.
Revision 1.70 by root, Sat Dec 10 11:58:38 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
… …
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 || ECB_NO_SMP
72 #define ECB_MEMORY_FENCE do { } while (0)
73#endif
74
68#ifndef ECB_MEMORY_FENCE 75#ifndef ECB_MEMORY_FENCE
69 #if ECB_GCC_VERSION(2,5) 76 #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__)
70 #if __x86 77 #if __i386__
71 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; or.b $0, -1(%%esp)" : : : "memory") 78 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
72 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 79 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */
73 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENC /* better be safe than sorry */ 80 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
74 #elif __amd64 81 #elif __amd64
75 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
76 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 83 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")
84 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
85 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
77 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") 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")
78 #endif 93 #endif
79 #endif 94 #endif
80#endif 95#endif
81 96
82#ifndef ECB_MEMORY_FENCE 97#ifndef ECB_MEMORY_FENCE
83 #if ECB_GCC_VERSION(4,4) 98 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__)
84 #define ECB_MEMORY_FENCE __sync_synchronize () 99 #define ECB_MEMORY_FENCE __sync_synchronize ()
85 #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); }) */
86 #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 ); }) */
87 #elif _MSC_VER >= 1400 102 #elif _MSC_VER >= 1400 /* VC++ 2005 */
88 #define ECB_MEMORY_FENCE do { } while (0) 103 #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 () 104 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
93 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 105 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
94 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 106 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
107 #elif defined(_WIN32)
108 #include <WinNT.h>
109 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
95 #endif 110 #endif
96#endif 111#endif
97 112
98#ifndef ECB_MEMORY_FENCE_ACQ 113#ifndef ECB_MEMORY_FENCE
114 #if !ECB_AVOID_PTHREADS
115 /*
116 * if you get undefined symbol references to pthread_mutex_lock,
117 * or failure to find pthread.h, then you should implement
118 * the ECB_MEMORY_FENCE operations for your cpu/compiler
119 * OR provide pthread.h and link against the posix thread library
120 * of your system.
121 */
99 #include <pthread.h> 122 #include <pthread.h>
123 #define ECB_NEEDS_PTHREADS 1
124 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
100 125
101 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 126 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) 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)
103 #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)
104 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 136 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
105#endif 137#endif
106 138
107/*****************************************************************************/ 139/*****************************************************************************/
108 140
… …
232 x *= 0x01010101; 264 x *= 0x01010101;
233 265
234 return x >> 24; 266 return x >> 24;
235 } 267 }
236 268
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; 269 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
241 ecb_function_ int ecb_ld32 (uint32_t x) 270 ecb_function_ int ecb_ld32 (uint32_t x)
242 { 271 {
243 int r = 0; 272 int r = 0;
244 273
… …
259 if (x >> 32) { x >>= 32; r += 32; } 288 if (x >> 32) { x >>= 32; r += 32; }
260 289
261 return r + ecb_ld32 (x); 290 return r + ecb_ld32 (x);
262 } 291 }
263#endif 292#endif
293
294ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
295{
296 return ( (x * 0x0802U & 0x22110U)
297 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
298}
299
300ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
301{
302 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
303 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
304 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
305 x = ( x >> 8 ) | ( x << 8);
306
307 return x;
308}
309
310ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
311{
312 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
313 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
314 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
315 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
316 x = ( x >> 16 ) | ( x << 16);
317
318 return x;
319}
264 320
265/* popcount64 is only available on 64 bit cpus as gcc builtin */ 321/* popcount64 is only available on 64 bit cpus as gcc builtin */
266/* so for this version we are lazy */ 322/* so for this version we are lazy */
267ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 323ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
268ecb_function_ int 324ecb_function_ int
… …
344 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 400 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
345#else 401#else
346 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 402 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
347#endif 403#endif
348 404
405#if __cplusplus
406 template<typename T>
407 static inline T ecb_div_rd (T val, T div)
408 {
409 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
410 }
411 template<typename T>
412 static inline T ecb_div_ru (T val, T div)
413 {
414 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
415 }
416#else
417 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
418 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
419#endif
420
349#if ecb_cplusplus_does_not_suck 421#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) */ 422 /* 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> 423 template<typename T, int N>
352 static inline int ecb_array_length (const T (&arr)[N]) 424 static inline int ecb_array_length (const T (&arr)[N])
353 { 425 {

Diff Legend

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