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

Comparing libecb/ecb.h (file contents):
Revision 1.50 by root, Fri Jun 17 21:16:11 2011 UTC vs.
Revision 1.72 by root, Wed Jan 18 12:13:00 2012 UTC

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:
38 typedef signed int int32_t; 38 typedef signed int int32_t;
39 typedef unsigned int uint32_t; 39 typedef unsigned int uint32_t;
40 #if __GNUC__ 40 #if __GNUC__
41 typedef signed long long int64_t; 41 typedef signed long long int64_t;
42 typedef unsigned long long uint64_t; 42 typedef unsigned long long uint64_t;
43 #else 43 #else /* _MSC_VER || __BORLANDC__ */
44 typedef signed __int64 int64_t; 44 typedef signed __int64 int64_t;
45 typedef unsigned __int64 uint64_t; 45 typedef unsigned __int64 uint64_t;
46 #endif 46 #endif
47#else 47#else
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
61 #else 61 #else
62 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 62 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
63 #endif 63 #endif
64#endif 64#endif
65 65
66/*****************************************************************************/
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
75#ifndef ECB_MEMORY_FENCE
76 #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
77 #if __i386__
78 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
79 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */
80 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
81 #elif __amd64
82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "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__
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")
93 #elif defined(__sparc)
94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #StoreLoad | #LoadLoad | #StoreStore" : : : "memory")
95 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadLoad" : : : "memory")
96 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #StoreStore")
97 #endif
98 #endif
99#endif
100
101#ifndef ECB_MEMORY_FENCE
102 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__)
103 #define ECB_MEMORY_FENCE __sync_synchronize ()
104 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
105 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
106 #elif _MSC_VER >= 1400 /* VC++ 2005 */
107 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
108 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
109 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
110 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
111 #elif defined(_WIN32)
112 #include <WinNT.h>
113 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
114 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
115 #include <mbarrier.h>
116 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
117 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
118 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
119 #endif
120#endif
121
122#ifndef ECB_MEMORY_FENCE
123 #if !ECB_AVOID_PTHREADS
124 /*
125 * if you get undefined symbol references to pthread_mutex_lock,
126 * or failure to find pthread.h, then you should implement
127 * the ECB_MEMORY_FENCE operations for your cpu/compiler
128 * OR provide pthread.h and link against the posix thread library
129 * of your system.
130 */
131 #include <pthread.h>
132 #define ECB_NEEDS_PTHREADS 1
133 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
134
135 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
136 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
137 #endif
138#endif
139
140#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE)
141 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
142#endif
143
144#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE)
145 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
146#endif
147
148/*****************************************************************************/
149
66#define ECB_C99 (__STDC_VERSION__ >= 199901L) 150#define ECB_C99 (__STDC_VERSION__ >= 199901L)
67 151
68#if __cplusplus 152#if __cplusplus
69 #define ecb_inline static inline 153 #define ecb_inline static inline
70#elif ECB_GCC_VERSION(2,5) 154#elif ECB_GCC_VERSION(2,5)
189 x *= 0x01010101; 273 x *= 0x01010101;
190 274
191 return x >> 24; 275 return x >> 24;
192 } 276 }
193 277
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; 278 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
198 ecb_function_ int ecb_ld32 (uint32_t x) 279 ecb_function_ int ecb_ld32 (uint32_t x)
199 { 280 {
200 int r = 0; 281 int r = 0;
201 282
216 if (x >> 32) { x >>= 32; r += 32; } 297 if (x >> 32) { x >>= 32; r += 32; }
217 298
218 return r + ecb_ld32 (x); 299 return r + ecb_ld32 (x);
219 } 300 }
220#endif 301#endif
302
303ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
304ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
305{
306 return ( (x * 0x0802U & 0x22110U)
307 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
308}
309
310ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
311ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
312{
313 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
314 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
315 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
316 x = ( x >> 8 ) | ( x << 8);
317
318 return x;
319}
320
321ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
322ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
323{
324 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
325 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
326 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
327 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
328 x = ( x >> 16 ) | ( x << 16);
329
330 return x;
331}
221 332
222/* popcount64 is only available on 64 bit cpus as gcc builtin */ 333/* popcount64 is only available on 64 bit cpus as gcc builtin */
223/* so for this version we are lazy */ 334/* so for this version we are lazy */
224ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 335ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
225ecb_function_ int 336ecb_function_ int
301 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 412 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
302#else 413#else
303 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 414 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
304#endif 415#endif
305 416
417#if __cplusplus
418 template<typename T>
419 static inline T ecb_div_rd (T val, T div)
420 {
421 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
422 }
423 template<typename T>
424 static inline T ecb_div_ru (T val, T div)
425 {
426 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
427 }
428#else
429 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
430 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
431#endif
432
306#if ecb_cplusplus_does_not_suck 433#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) */ 434 /* 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> 435 template<typename T, int N>
309 static inline int ecb_array_length (const T (&arr)[N]) 436 static inline int ecb_array_length (const T (&arr)[N])
310 { 437 {

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines