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

Comparing libecb/ecb.h (file contents):
Revision 1.79 by root, Sat Feb 4 17:57:47 2012 UTC vs.
Revision 1.127 by root, Sun Jan 5 15:54:09 2014 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2013 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 0x00010003
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
66#endif
67
68/* work around x32 idiocy by defining proper macros */
69#if __amd64 || __x86_64 || _M_AMD64 || _M_X64
70 #if _ILP32
71 #define ECB_AMD64_X32 1
72 #else
73 #define ECB_AMD64 1
74 #endif
49#endif 75#endif
50 76
51/* many compilers define _GNUC_ to some versions but then only implement 77/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions, 78 * what their idiot authors think are the "more important" extensions,
53 * causing enormous grief in return for some better fake benchmark numbers. 79 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so. 80 * or so.
55 * we try to detect these and simply assume they are not gcc - if they have 81 * 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. 82 * an issue with that they should have done it right in the first place.
57 */ 83 */
58#ifndef ECB_GCC_VERSION 84#ifndef ECB_GCC_VERSION
59 #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) 85 #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 86 #define ECB_GCC_VERSION(major,minor) 0
61 #else 87 #else
62 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 88 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
63 #endif 89 #endif
64#endif 90#endif
65 91
92#define ECB_CPP (__cplusplus+0)
93#define ECB_CPP11 (__cplusplus >= 201103L)
94
95#if ECB_CPP
96 #define ECB_C 0
97 #define ECB_STDC_VERSION 0
98#else
99 #define ECB_C 1
100 #define ECB_STDC_VERSION __STDC_VERSION__
101#endif
102
103#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
104#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
105
106#if ECB_CPP
107 #define ECB_EXTERN_C extern "C"
108 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
109 #define ECB_EXTERN_C_END }
110#else
111 #define ECB_EXTERN_C extern
112 #define ECB_EXTERN_C_BEG
113 #define ECB_EXTERN_C_END
114#endif
115
66/*****************************************************************************/ 116/*****************************************************************************/
67 117
68/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 118/* 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 */ 119/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
70 120
71#if ECB_NO_THREADS 121#if ECB_NO_THREADS
72# define ECB_NO_SMP 1 122 #define ECB_NO_SMP 1
73#endif 123#endif
74 124
75#if ECB_NO_THREADS || ECB_NO_SMP 125#if ECB_NO_SMP
76 #define ECB_MEMORY_FENCE do { } while (0) 126 #define ECB_MEMORY_FENCE do { } while (0)
77#endif 127#endif
78 128
79#ifndef ECB_MEMORY_FENCE 129#ifndef ECB_MEMORY_FENCE
80 #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 130 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
81 #if __i386 || __i386__ 131 #if __i386 || __i386__
82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
83 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ 133 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
84 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ 134 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
85 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 135 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
86 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 136 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
87 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 137 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
88 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ 138 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
89 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 139 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
90 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
91 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ 141 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
92 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) 142 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
93 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 143 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
94 #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ 144 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
95 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) 145 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
96 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 146 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
97 #elif __sparc || __sparc__ 147 #elif (__sparc || __sparc__) && !__sparcv8
98 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") 148 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
99 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 149 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
100 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 150 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
101 #elif defined(__s390__) || defined(__s390x__) 151 #elif defined __s390__ || defined __s390x__
102 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 152 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
153 #elif defined __mips__
154 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
155 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
156 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
157 #elif defined __alpha__
158 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
159 #elif defined __hppa__
160 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
161 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
162 #elif defined __ia64__
163 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
164 #elif defined __m68k__
165 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
166 #elif defined __m88k__
167 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
168 #elif defined __sh__
169 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
103 #endif 170 #endif
104 #endif 171 #endif
105#endif 172#endif
106 173
107#ifndef ECB_MEMORY_FENCE 174#ifndef ECB_MEMORY_FENCE
175 #if ECB_GCC_VERSION(4,7)
176 /* see comment below (stdatomic.h) about the C11 memory model. */
177 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
178
179 /* The __has_feature syntax from clang is so misdesigned that we cannot use it
180 * without risking compile time errors with other compilers. We *could*
181 * define our own ecb_clang_has_feature, but I just can't be bothered to work
182 * around this shit time and again.
183 * #elif defined __clang && __has_feature (cxx_atomic)
184 * // see comment below (stdatomic.h) about the C11 memory model.
185 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
186 */
187
108 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__) 188 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
109 #define ECB_MEMORY_FENCE __sync_synchronize () 189 #define ECB_MEMORY_FENCE __sync_synchronize ()
110 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ 190 #elif _MSC_VER >= 1500 /* VC++ 2008 */
111 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ 191 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
192 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
193 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
194 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
195 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
112 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 196 #elif _MSC_VER >= 1400 /* VC++ 2005 */
113 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 197 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
114 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 198 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
115 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 199 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
116 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 200 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
117 #elif defined(_WIN32) 201 #elif defined _WIN32
118 #include <WinNT.h> 202 #include <WinNT.h>
119 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 203 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
120 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 204 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
121 #include <mbarrier.h> 205 #include <mbarrier.h>
122 #define ECB_MEMORY_FENCE __machine_rw_barrier () 206 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
123 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 207 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
124 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () 208 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
209 #elif __xlC__
210 #define ECB_MEMORY_FENCE __sync ()
211 #endif
212#endif
213
214#ifndef ECB_MEMORY_FENCE
215 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
216 /* we assume that these memory fences work on all variables/all memory accesses, */
217 /* not just C11 atomics and atomic accesses */
218 #include <stdatomic.h>
219 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
220 /* any fence other than seq_cst, which isn't very efficient for us. */
221 /* Why that is, we don't know - either the C11 memory model is quite useless */
222 /* for most usages, or gcc and clang have a bug */
223 /* I *currently* lean towards the latter, and inefficiently implement */
224 /* all three of ecb's fences as a seq_cst fence */
225 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
125 #endif 226 #endif
126#endif 227#endif
127 228
128#ifndef ECB_MEMORY_FENCE 229#ifndef ECB_MEMORY_FENCE
129 #if !ECB_AVOID_PTHREADS 230 #if !ECB_AVOID_PTHREADS
141 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 242 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
142 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) 243 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
143 #endif 244 #endif
144#endif 245#endif
145 246
146#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) 247#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
147 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 248 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
148#endif 249#endif
149 250
150#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) 251#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
151 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 252 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
152#endif 253#endif
153 254
154/*****************************************************************************/ 255/*****************************************************************************/
155
156#define ECB_C99 (__STDC_VERSION__ >= 199901L)
157 256
158#if __cplusplus 257#if __cplusplus
159 #define ecb_inline static inline 258 #define ecb_inline static inline
160#elif ECB_GCC_VERSION(2,5) 259#elif ECB_GCC_VERSION(2,5)
161 #define ecb_inline static __inline__ 260 #define ecb_inline static __inline__
187 #define ecb_is_constant(expr) __builtin_constant_p (expr) 286 #define ecb_is_constant(expr) __builtin_constant_p (expr)
188 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 287 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
189 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 288 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
190#else 289#else
191 #define ecb_attribute(attrlist) 290 #define ecb_attribute(attrlist)
291
292 /* possible C11 impl for integral types
293 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
294 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
295
192 #define ecb_is_constant(expr) 0 296 #define ecb_is_constant(expr) 0
193 #define ecb_expect(expr,value) (expr) 297 #define ecb_expect(expr,value) (expr)
194 #define ecb_prefetch(addr,rw,locality) 298 #define ecb_prefetch(addr,rw,locality)
195#endif 299#endif
196 300
200#elif ECB_GCC_VERSION(3,0) 304#elif ECB_GCC_VERSION(3,0)
201 #define ecb_decltype(x) __typeof(x) 305 #define ecb_decltype(x) __typeof(x)
202#endif 306#endif
203 307
204#define ecb_noinline ecb_attribute ((__noinline__)) 308#define ecb_noinline ecb_attribute ((__noinline__))
205#define ecb_noreturn ecb_attribute ((__noreturn__))
206#define ecb_unused ecb_attribute ((__unused__)) 309#define ecb_unused ecb_attribute ((__unused__))
207#define ecb_const ecb_attribute ((__const__)) 310#define ecb_const ecb_attribute ((__const__))
208#define ecb_pure ecb_attribute ((__pure__)) 311#define ecb_pure ecb_attribute ((__pure__))
312
313#if ECB_C11
314 #define ecb_noreturn _Noreturn
315#else
316 #define ecb_noreturn ecb_attribute ((__noreturn__))
317#endif
209 318
210#if ECB_GCC_VERSION(4,3) 319#if ECB_GCC_VERSION(4,3)
211 #define ecb_artificial ecb_attribute ((__artificial__)) 320 #define ecb_artificial ecb_attribute ((__artificial__))
212 #define ecb_hot ecb_attribute ((__hot__)) 321 #define ecb_hot ecb_attribute ((__hot__))
213 #define ecb_cold ecb_attribute ((__cold__)) 322 #define ecb_cold ecb_attribute ((__cold__))
304 413
305 return r + ecb_ld32 (x); 414 return r + ecb_ld32 (x);
306 } 415 }
307#endif 416#endif
308 417
418ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
419ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
420ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
421ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
422
309ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 423ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
310ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 424ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
311{ 425{
312 return ( (x * 0x0802U & 0x22110U) 426 return ( (x * 0x0802U & 0x22110U)
313 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 427 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
397 ecb_inline void ecb_unreachable (void) ecb_noreturn; 511 ecb_inline void ecb_unreachable (void) ecb_noreturn;
398 ecb_inline void ecb_unreachable (void) { } 512 ecb_inline void ecb_unreachable (void) { }
399#endif 513#endif
400 514
401/* try to tell the compiler that some condition is definitely true */ 515/* try to tell the compiler that some condition is definitely true */
402#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 516#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
403 517
404ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 518ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
405ecb_inline unsigned char 519ecb_inline unsigned char
406ecb_byteorder_helper (void) 520ecb_byteorder_helper (void)
407{ 521{
408 const uint32_t u = 0x11223344; 522 /* the union code still generates code under pressure in gcc, */
409 return *(unsigned char *)&u; 523 /* but less than using pointers, and always seems to */
524 /* successfully return a constant. */
525 /* the reason why we have this horrible preprocessor mess */
526 /* is to avoid it in all cases, at least on common architectures */
527 /* or when using a recent enough gcc version (>= 4.6) */
528#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
529 return 0x44;
530#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
531 return 0x44;
532#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
533 return 0x11;
534#else
535 union
536 {
537 uint32_t i;
538 uint8_t c;
539 } u = { 0x11223344 };
540 return u.c;
541#endif
410} 542}
411 543
412ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 544ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
413ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 545ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
414ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 546ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
445 } 577 }
446#else 578#else
447 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 579 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
448#endif 580#endif
449 581
582/*******************************************************************************/
583/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
584
585/* basically, everything uses "ieee pure-endian" floating point numbers */
586/* the only noteworthy exception is ancient armle, which uses order 43218765 */
587#if 0 \
588 || __i386 || __i386__ \
589 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
590 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
591 || defined __arm__ && defined __ARM_EABI__ \
592 || defined __s390__ || defined __s390x__ \
593 || defined __mips__ \
594 || defined __alpha__ \
595 || defined __hppa__ \
596 || defined __ia64__ \
597 || defined __m68k__ \
598 || defined __m88k__ \
599 || defined __sh__ \
600 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
601 #define ECB_STDFP 1
602 #include <string.h> /* for memcpy */
603#else
604 #define ECB_STDFP 0
605#endif
606
607#ifndef ECB_NO_LIBM
608
609 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
610
611 /* only the oldest of old doesn't have this one. solaris. */
612 #ifdef INFINITY
613 #define ECB_INFINITY INFINITY
614 #else
615 #define ECB_INFINITY HUGE_VAL
450#endif 616 #endif
451 617
618 #ifdef NAN
619 #define ECB_NAN NAN
620 #else
621 #define ECB_NAN ECB_INFINITY
622 #endif
623
624 /* converts an ieee half/binary16 to a float */
625 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
626 ecb_function_ float
627 ecb_binary16_to_float (uint16_t x)
628 {
629 int e = (x >> 10) & 0x1f;
630 int m = x & 0x3ff;
631 float r;
632
633 if (!e ) r = ldexpf (m , -24);
634 else if (e != 31) r = ldexpf (m + 0x400, e - 25);
635 else if (m ) r = ECB_NAN;
636 else r = ECB_INFINITY;
637
638 return x & 0x8000 ? -r : r;
639 }
640
641 /* convert a float to ieee single/binary32 */
642 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
643 ecb_function_ uint32_t
644 ecb_float_to_binary32 (float x)
645 {
646 uint32_t r;
647
648 #if ECB_STDFP
649 memcpy (&r, &x, 4);
650 #else
651 /* slow emulation, works for anything but -0 */
652 uint32_t m;
653 int e;
654
655 if (x == 0e0f ) return 0x00000000U;
656 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
657 if (x < -3.40282346638528860e+38f) return 0xff800000U;
658 if (x != x ) return 0x7fbfffffU;
659
660 m = frexpf (x, &e) * 0x1000000U;
661
662 r = m & 0x80000000U;
663
664 if (r)
665 m = -m;
666
667 if (e <= -126)
668 {
669 m &= 0xffffffU;
670 m >>= (-125 - e);
671 e = -126;
672 }
673
674 r |= (e + 126) << 23;
675 r |= m & 0x7fffffU;
676 #endif
677
678 return r;
679 }
680
681 /* converts an ieee single/binary32 to a float */
682 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
683 ecb_function_ float
684 ecb_binary32_to_float (uint32_t x)
685 {
686 float r;
687
688 #if ECB_STDFP
689 memcpy (&r, &x, 4);
690 #else
691 /* emulation, only works for normals and subnormals and +0 */
692 int neg = x >> 31;
693 int e = (x >> 23) & 0xffU;
694
695 x &= 0x7fffffU;
696
697 if (e)
698 x |= 0x800000U;
699 else
700 e = 1;
701
702 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
703 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
704
705 r = neg ? -r : r;
706 #endif
707
708 return r;
709 }
710
711 /* convert a double to ieee double/binary64 */
712 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
713 ecb_function_ uint64_t
714 ecb_double_to_binary64 (double x)
715 {
716 uint64_t r;
717
718 #if ECB_STDFP
719 memcpy (&r, &x, 8);
720 #else
721 /* slow emulation, works for anything but -0 */
722 uint64_t m;
723 int e;
724
725 if (x == 0e0 ) return 0x0000000000000000U;
726 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
727 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
728 if (x != x ) return 0X7ff7ffffffffffffU;
729
730 m = frexp (x, &e) * 0x20000000000000U;
731
732 r = m & 0x8000000000000000;;
733
734 if (r)
735 m = -m;
736
737 if (e <= -1022)
738 {
739 m &= 0x1fffffffffffffU;
740 m >>= (-1021 - e);
741 e = -1022;
742 }
743
744 r |= ((uint64_t)(e + 1022)) << 52;
745 r |= m & 0xfffffffffffffU;
746 #endif
747
748 return r;
749 }
750
751 /* converts an ieee double/binary64 to a double */
752 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
753 ecb_function_ double
754 ecb_binary64_to_double (uint64_t x)
755 {
756 double r;
757
758 #if ECB_STDFP
759 memcpy (&r, &x, 8);
760 #else
761 /* emulation, only works for normals and subnormals and +0 */
762 int neg = x >> 63;
763 int e = (x >> 52) & 0x7ffU;
764
765 x &= 0xfffffffffffffU;
766
767 if (e)
768 x |= 0x10000000000000U;
769 else
770 e = 1;
771
772 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
773 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
774
775 r = neg ? -r : r;
776 #endif
777
778 return r;
779 }
780
781#endif
782
783#endif
784

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines