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

Comparing libecb/ecb.h (file contents):
Revision 1.66 by root, Wed Aug 24 16:07:55 2011 UTC vs.
Revision 1.92 by root, Tue May 29 17:17:56 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
59 typedef intptr_t ptrdiff_t;
47#else 60#else
48 #include <inttypes.h> 61 #include <inttypes.h>
62 #if UINTMAX_MAX > 0xffffffffU
63 #define ECB_PTRSIZE 8
64 #else
65 #define ECB_PTRSIZE 4
66 #endif
49#endif 67#endif
50 68
51/* many compilers define _GNUC_ to some versions but then only implement 69/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions, 70 * what their idiot authors think are the "more important" extensions,
53 * causing enormous grief in return for some better fake benchmark numbers. 71 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so. 72 * or so.
55 * we try to detect these and simply assume they are not gcc - if they have 73 * 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. 74 * an issue with that they should have done it right in the first place.
57 */ 75 */
58#ifndef ECB_GCC_VERSION 76#ifndef ECB_GCC_VERSION
59 #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) 77 #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 78 #define ECB_GCC_VERSION(major,minor) 0
61 #else 79 #else
62 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 80 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
63 #endif 81 #endif
64#endif 82#endif
65 83
84#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
85#define ECB_C99 (__STDC_VERSION__ >= 199901L)
86#define ECB_C11 (__STDC_VERSION__ >= 201112L)
87#define ECB_CPP (__cplusplus+0)
88#define ECB_CPP98 (__cplusplus >= 199711L)
89#define ECB_CPP11 (__cplusplus >= 201103L)
90
66/*****************************************************************************/ 91/*****************************************************************************/
67 92
68/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 93/* 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 */ 94/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
70 95
96#if ECB_NO_THREADS
97# define ECB_NO_SMP 1
98#endif
99
71#if ECB_NO_THREADS || ECB_NO_SMP 100#if ECB_NO_THREADS || ECB_NO_SMP
72 #define ECB_MEMORY_FENCE do { } while (0) 101 #define ECB_MEMORY_FENCE do { } while (0)
73#endif 102#endif
74 103
75#ifndef ECB_MEMORY_FENCE 104#ifndef ECB_MEMORY_FENCE
105 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
106 /* we assume that these memory fences work on all variables/all memory accesses, */
107 /* not just C11 atomics and atomic accesses */
108 #include <stdatomic.h>
109 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_acq_rel)
110 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
111 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
112 #endif
113#endif
114
115#ifndef ECB_MEMORY_FENCE_RELEASE
76 #if ECB_GCC_VERSION(2,5) 116 #if ECB_GCC_VERSION(4,7)
117 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_ACQ_REL)
118 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
119 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
120 #endif
121#endif
122
123#ifndef ECB_MEMORY_FENCE
124 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
77 #if __i386__ 125 #if __i386 || __i386__
78 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 126 #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 */ 127 #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 */ 128 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
81 #elif __amd64 129 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 130 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
83 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 131 #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 */ 132 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
85 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 133 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
86 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 134 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
87 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ 135 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
88 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) 136 || 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") 137 #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__ ) \ 138 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
91 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) 139 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
92 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
141 #elif __sparc || __sparc__
142 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory")
143 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
144 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
145 #elif defined __s390__ || defined __s390x__
146 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
147 #elif defined __mips__
148 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
149 #elif defined __alpha__
150 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
93 #endif 151 #endif
94 #endif 152 #endif
95#endif 153#endif
96 154
97#ifndef ECB_MEMORY_FENCE 155#ifndef ECB_MEMORY_FENCE
98 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) 156 #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
99 #define ECB_MEMORY_FENCE __sync_synchronize () 157 #define ECB_MEMORY_FENCE __sync_synchronize ()
100 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ 158 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
101 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ 159 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
102 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 160 #elif _MSC_VER >= 1400 /* VC++ 2005 */
103 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 161 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
104 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 162 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
105 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 163 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
106 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 164 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
107 #elif defined(_WIN32) 165 #elif defined _WIN32
108 #include <WinNT.h> 166 #include <WinNT.h>
109 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 167 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
168 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
169 #include <mbarrier.h>
170 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
171 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
172 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
173 #elif __xlC__
174 #define ECB_MEMORY_FENCE __sync ()
110 #endif 175 #endif
111#endif 176#endif
112 177
113#ifndef ECB_MEMORY_FENCE 178#ifndef ECB_MEMORY_FENCE
114 #if !ECB_AVOID_PTHREADS 179 #if !ECB_AVOID_PTHREADS
126 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 191 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
127 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) 192 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
128 #endif 193 #endif
129#endif 194#endif
130 195
131#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) 196#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
132 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 197 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
133#endif 198#endif
134 199
135#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) 200#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
136 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 201 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
137#endif 202#endif
138 203
139/*****************************************************************************/ 204/*****************************************************************************/
140
141#define ECB_C99 (__STDC_VERSION__ >= 199901L)
142 205
143#if __cplusplus 206#if __cplusplus
144 #define ecb_inline static inline 207 #define ecb_inline static inline
145#elif ECB_GCC_VERSION(2,5) 208#elif ECB_GCC_VERSION(2,5)
146 #define ecb_inline static __inline__ 209 #define ecb_inline static __inline__
185#elif ECB_GCC_VERSION(3,0) 248#elif ECB_GCC_VERSION(3,0)
186 #define ecb_decltype(x) __typeof(x) 249 #define ecb_decltype(x) __typeof(x)
187#endif 250#endif
188 251
189#define ecb_noinline ecb_attribute ((__noinline__)) 252#define ecb_noinline ecb_attribute ((__noinline__))
190#define ecb_noreturn ecb_attribute ((__noreturn__))
191#define ecb_unused ecb_attribute ((__unused__)) 253#define ecb_unused ecb_attribute ((__unused__))
192#define ecb_const ecb_attribute ((__const__)) 254#define ecb_const ecb_attribute ((__const__))
193#define ecb_pure ecb_attribute ((__pure__)) 255#define ecb_pure ecb_attribute ((__pure__))
256
257#if ECB_C11
258 #define ecb_noreturn _Noreturn
259#else
260 #define ecb_noreturn ecb_attribute ((__noreturn__))
261#endif
194 262
195#if ECB_GCC_VERSION(4,3) 263#if ECB_GCC_VERSION(4,3)
196 #define ecb_artificial ecb_attribute ((__artificial__)) 264 #define ecb_artificial ecb_attribute ((__artificial__))
197 #define ecb_hot ecb_attribute ((__hot__)) 265 #define ecb_hot ecb_attribute ((__hot__))
198 #define ecb_cold ecb_attribute ((__cold__)) 266 #define ecb_cold ecb_attribute ((__cold__))
289 357
290 return r + ecb_ld32 (x); 358 return r + ecb_ld32 (x);
291 } 359 }
292#endif 360#endif
293 361
362ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
363ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
364ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
365ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
366
367ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
368ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
369{
370 return ( (x * 0x0802U & 0x22110U)
371 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
372}
373
374ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
375ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
376{
377 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
378 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
379 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
380 x = ( x >> 8 ) | ( x << 8);
381
382 return x;
383}
384
385ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
386ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
387{
388 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
389 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
390 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
391 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
392 x = ( x >> 16 ) | ( x << 16);
393
394 return x;
395}
396
294/* popcount64 is only available on 64 bit cpus as gcc builtin */ 397/* popcount64 is only available on 64 bit cpus as gcc builtin */
295/* so for this version we are lazy */ 398/* so for this version we are lazy */
296ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 399ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
297ecb_function_ int 400ecb_function_ int
298ecb_popcount64 (uint64_t x) 401ecb_popcount64 (uint64_t x)
347 450
348#if ECB_GCC_VERSION(4,5) 451#if ECB_GCC_VERSION(4,5)
349 #define ecb_unreachable() __builtin_unreachable () 452 #define ecb_unreachable() __builtin_unreachable ()
350#else 453#else
351 /* this seems to work fine, but gcc always emits a warning for it :/ */ 454 /* this seems to work fine, but gcc always emits a warning for it :/ */
352 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 455 ecb_inline void ecb_unreachable (void) ecb_noreturn;
353 ecb_function_ void ecb_unreachable (void) { } 456 ecb_inline void ecb_unreachable (void) { }
354#endif 457#endif
355 458
356/* try to tell the compiler that some condition is definitely true */ 459/* try to tell the compiler that some condition is definitely true */
357#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 460#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
358 461
359ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 462ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
360ecb_function_ unsigned char 463ecb_inline unsigned char
361ecb_byteorder_helper (void) 464ecb_byteorder_helper (void)
362{ 465{
363 const uint32_t u = 0x11223344; 466 const uint32_t u = 0x11223344;
364 return *(unsigned char *)&u; 467 return *(unsigned char *)&u;
365} 468}
366 469
367ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 470ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
368ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 471ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
369ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 472ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
370ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 473ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
371 474
372#if ECB_GCC_VERSION(3,0) || ECB_C99 475#if ECB_GCC_VERSION(3,0) || ECB_C99
373 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 476 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
374#else 477#else
375 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 478 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
479#endif
480
481#if __cplusplus
482 template<typename T>
483 static inline T ecb_div_rd (T val, T div)
484 {
485 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
486 }
487 template<typename T>
488 static inline T ecb_div_ru (T val, T div)
489 {
490 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
491 }
492#else
493 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
494 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
376#endif 495#endif
377 496
378#if ecb_cplusplus_does_not_suck 497#if ecb_cplusplus_does_not_suck
379 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 498 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
380 template<typename T, int N> 499 template<typename T, int N>

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines