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

Comparing libecb/ecb.h (file contents):
Revision 1.63 by root, Thu Aug 4 14:37:46 2011 UTC vs.
Revision 1.88 by root, Mon May 28 08:40:25 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:
… …
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
47#else 59#else
48 #include <inttypes.h> 60 #include <inttypes.h>
61 #if UINTMAX_MAX > 0xffffffffU
62 #define ECB_PTRSIZE 8
63 #else
64 #define ECB_PTRSIZE 4
65 #endif
49#endif 66#endif
50 67
51/* many compilers define _GNUC_ to some versions but then only implement 68/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions, 69 * what their idiot authors think are the "more important" extensions,
53 * causing enormous grief in return for some better fake benchmark numbers. 70 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so. 71 * or so.
55 * we try to detect these and simply assume they are not gcc - if they have 72 * 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. 73 * an issue with that they should have done it right in the first place.
57 */ 74 */
58#ifndef ECB_GCC_VERSION 75#ifndef ECB_GCC_VERSION
59 #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) 76 #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 77 #define ECB_GCC_VERSION(major,minor) 0
61 #else 78 #else
62 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 79 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
63 #endif 80 #endif
64#endif 81#endif
… …
66/*****************************************************************************/ 83/*****************************************************************************/
67 84
68/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 85/* 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 */ 86/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
70 87
88#if ECB_NO_THREADS
89# define ECB_NO_SMP 1
90#endif
91
71#if ECB_NO_THREADS || ECB_NO_SMP 92#if ECB_NO_THREADS || ECB_NO_SMP
72 #define ECB_MEMORY_FENCE do { } while (0) 93 #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 94#endif
76 95
77#ifndef ECB_MEMORY_FENCE 96#ifndef ECB_MEMORY_FENCE
78 #if ECB_GCC_VERSION(2,5) 97 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
79 #if __x86 98 #if __i386 || __i386__
80 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 99 #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 */ 100 #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 */ 101 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
83 #elif __amd64 102 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 103 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
85 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 104 #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 */ 105 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
87 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 106 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
88 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 107 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
89 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ 108 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
90 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) \ 109 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
110 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
91 || defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ 111 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
92 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) 112 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
93 #define ECB_MEMORY_FENCE \ 113 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
94 do { \ 114 #elif __sparc || __sparc__
95 int null = 0; \ 115 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory")
96 __asm__ __volatile__ ("mcr p15,0,%0,c6,c10,5", : "=&r" (null) : : "memory"); \ 116 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
97 while (0) 117 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
118 #elif defined __s390__ || defined __s390x__
119 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
120 #elif defined __mips__
121 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
122 #elif defined __alpha__
123 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
98 #endif 124 #endif
99 #endif 125 #endif
100#endif 126#endif
101 127
102#ifndef ECB_MEMORY_FENCE 128#ifndef ECB_MEMORY_FENCE
103 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) 129 #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
104 #define ECB_MEMORY_FENCE __sync_synchronize () 130 #define ECB_MEMORY_FENCE __sync_synchronize ()
105 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ 131 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
106 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ 132 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
107 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 133 #elif _MSC_VER >= 1400 /* VC++ 2005 */
108 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 134 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
109 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 135 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
110 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 136 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
111 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 137 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
112 #elif defined(_WIN32) 138 #elif defined _WIN32
113 #include <WinNT.h> 139 #include <WinNT.h>
114 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 140 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
141 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
142 #include <mbarrier.h>
143 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
144 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
145 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
146 #elif __xlC__
147 #define ECB_MEMORY_FENCE __sync ()
115 #endif 148 #endif
116#endif 149#endif
117 150
118#ifndef ECB_MEMORY_FENCE 151#ifndef ECB_MEMORY_FENCE
119 #if !ECB_AVOID_PTHREADS 152 #if !ECB_AVOID_PTHREADS
… …
131 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 164 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
132 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) 165 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
133 #endif 166 #endif
134#endif 167#endif
135 168
136#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) 169#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
137 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 170 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
138#endif 171#endif
139 172
140#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) 173#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
141 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 174 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
142#endif 175#endif
143 176
144/*****************************************************************************/ 177/*****************************************************************************/
145 178
… …
294 327
295 return r + ecb_ld32 (x); 328 return r + ecb_ld32 (x);
296 } 329 }
297#endif 330#endif
298 331
332ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
333ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
334ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
335ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
336
337ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
338ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
339{
340 return ( (x * 0x0802U & 0x22110U)
341 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
342}
343
344ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
345ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
346{
347 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
348 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
349 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
350 x = ( x >> 8 ) | ( x << 8);
351
352 return x;
353}
354
355ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
356ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
357{
358 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
359 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
360 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
361 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
362 x = ( x >> 16 ) | ( x << 16);
363
364 return x;
365}
366
299/* popcount64 is only available on 64 bit cpus as gcc builtin */ 367/* popcount64 is only available on 64 bit cpus as gcc builtin */
300/* so for this version we are lazy */ 368/* so for this version we are lazy */
301ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 369ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
302ecb_function_ int 370ecb_function_ int
303ecb_popcount64 (uint64_t x) 371ecb_popcount64 (uint64_t x)
… …
352 420
353#if ECB_GCC_VERSION(4,5) 421#if ECB_GCC_VERSION(4,5)
354 #define ecb_unreachable() __builtin_unreachable () 422 #define ecb_unreachable() __builtin_unreachable ()
355#else 423#else
356 /* this seems to work fine, but gcc always emits a warning for it :/ */ 424 /* this seems to work fine, but gcc always emits a warning for it :/ */
357 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 425 ecb_inline void ecb_unreachable (void) ecb_noreturn;
358 ecb_function_ void ecb_unreachable (void) { } 426 ecb_inline void ecb_unreachable (void) { }
359#endif 427#endif
360 428
361/* try to tell the compiler that some condition is definitely true */ 429/* try to tell the compiler that some condition is definitely true */
362#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 430#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
363 431
364ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 432ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
365ecb_function_ unsigned char 433ecb_inline unsigned char
366ecb_byteorder_helper (void) 434ecb_byteorder_helper (void)
367{ 435{
368 const uint32_t u = 0x11223344; 436 const uint32_t u = 0x11223344;
369 return *(unsigned char *)&u; 437 return *(unsigned char *)&u;
370} 438}
371 439
372ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 440ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
373ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 441ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
374ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 442ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
375ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 443ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
376 444
377#if ECB_GCC_VERSION(3,0) || ECB_C99 445#if ECB_GCC_VERSION(3,0) || ECB_C99
378 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 446 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
379#else 447#else
380 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 448 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
449#endif
450
451#if __cplusplus
452 template<typename T>
453 static inline T ecb_div_rd (T val, T div)
454 {
455 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
456 }
457 template<typename T>
458 static inline T ecb_div_ru (T val, T div)
459 {
460 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
461 }
462#else
463 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
464 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
381#endif 465#endif
382 466
383#if ecb_cplusplus_does_not_suck 467#if ecb_cplusplus_does_not_suck
384 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 468 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
385 template<typename T, int N> 469 template<typename T, int N>

Diff Legend

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