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

Comparing libecb/ecb.h (file contents):
Revision 1.73 by root, Wed Jan 18 12:44:23 2012 UTC vs.
Revision 1.129 by root, Thu Mar 20 22:04:29 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 (©) 2009-2014 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 || ECB_NO_SMP 121#if ECB_NO_THREADS
122 #define ECB_NO_SMP 1
123#endif
124
125#if ECB_NO_SMP
72 #define ECB_MEMORY_FENCE do { } while (0) 126 #define ECB_MEMORY_FENCE do { } while (0)
73#endif 127#endif
74 128
75#ifndef ECB_MEMORY_FENCE 129#ifndef ECB_MEMORY_FENCE
76 #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 130 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
77 #if __i386 || __i386__ 131 #if __i386 || __i386__
78 #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")
79 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ 133 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
80 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ 134 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
81 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 135 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 136 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
83 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 137 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
84 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ 138 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
85 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 139 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
86 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
87 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ 141 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
88 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) 142 || 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") 143 #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__ ) \ 144 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
91 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) 145 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
92 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 146 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
147 #elif __aarch64__
148 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
93 #elif __sparc || __sparc__ 149 #elif (__sparc || __sparc__) && !__sparcv8
94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #StoreLoad | #LoadLoad | #StoreStore" : : : "memory") 150 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
95 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadLoad" : : : "memory") 151 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
152 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
153 #elif defined __s390__ || defined __s390x__
154 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
155 #elif defined __mips__
156 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
157 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
158 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
159 #elif defined __alpha__
160 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
161 #elif defined __hppa__
162 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
96 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #StoreStore") 163 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
164 #elif defined __ia64__
165 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
166 #elif defined __m68k__
167 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
168 #elif defined __m88k__
169 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
170 #elif defined __sh__
171 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
97 #endif 172 #endif
98 #endif 173 #endif
99#endif 174#endif
100 175
101#ifndef ECB_MEMORY_FENCE 176#ifndef ECB_MEMORY_FENCE
177 #if ECB_GCC_VERSION(4,7)
178 /* see comment below (stdatomic.h) about the C11 memory model. */
179 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
180 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
181 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
182
183 /* The __has_feature syntax from clang is so misdesigned that we cannot use it
184 * without risking compile time errors with other compilers. We *could*
185 * define our own ecb_clang_has_feature, but I just can't be bothered to work
186 * around this shit time and again.
187 * #elif defined __clang && __has_feature (cxx_atomic)
188 * // see comment below (stdatomic.h) about the C11 memory model.
189 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
190 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
191 * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
192 */
193
102 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__) 194 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
103 #define ECB_MEMORY_FENCE __sync_synchronize () 195 #define ECB_MEMORY_FENCE __sync_synchronize ()
104 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ 196 #elif _MSC_VER >= 1500 /* VC++ 2008 */
105 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ 197 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
198 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
199 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
200 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
201 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
106 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 202 #elif _MSC_VER >= 1400 /* VC++ 2005 */
107 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 203 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
108 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 204 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
109 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 205 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
110 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 206 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
111 #elif defined(_WIN32) 207 #elif defined _WIN32
112 #include <WinNT.h> 208 #include <WinNT.h>
113 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 209 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
114 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 210 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
115 #include <mbarrier.h> 211 #include <mbarrier.h>
116 #define ECB_MEMORY_FENCE __machine_rw_barrier () 212 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
117 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 213 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
118 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () 214 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
215 #elif __xlC__
216 #define ECB_MEMORY_FENCE __sync ()
217 #endif
218#endif
219
220#ifndef ECB_MEMORY_FENCE
221 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
222 /* we assume that these memory fences work on all variables/all memory accesses, */
223 /* not just C11 atomics and atomic accesses */
224 #include <stdatomic.h>
225 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
226 /* any fence other than seq_cst, which isn't very efficient for us. */
227 /* Why that is, we don't know - either the C11 memory model is quite useless */
228 /* for most usages, or gcc and clang have a bug */
229 /* I *currently* lean towards the latter, and inefficiently implement */
230 /* all three of ecb's fences as a seq_cst fence */
231 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
232 /* for all __atomic_thread_fence's except seq_cst */
233 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
119 #endif 234 #endif
120#endif 235#endif
121 236
122#ifndef ECB_MEMORY_FENCE 237#ifndef ECB_MEMORY_FENCE
123 #if !ECB_AVOID_PTHREADS 238 #if !ECB_AVOID_PTHREADS
… …
135 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 250 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) 251 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
137 #endif 252 #endif
138#endif 253#endif
139 254
140#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) 255#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
141 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 256 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
142#endif 257#endif
143 258
144#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) 259#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
145 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 260 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
146#endif 261#endif
147 262
148/*****************************************************************************/ 263/*****************************************************************************/
149
150#define ECB_C99 (__STDC_VERSION__ >= 199901L)
151 264
152#if __cplusplus 265#if __cplusplus
153 #define ecb_inline static inline 266 #define ecb_inline static inline
154#elif ECB_GCC_VERSION(2,5) 267#elif ECB_GCC_VERSION(2,5)
155 #define ecb_inline static __inline__ 268 #define ecb_inline static __inline__
… …
181 #define ecb_is_constant(expr) __builtin_constant_p (expr) 294 #define ecb_is_constant(expr) __builtin_constant_p (expr)
182 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 295 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
183 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 296 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
184#else 297#else
185 #define ecb_attribute(attrlist) 298 #define ecb_attribute(attrlist)
299
300 /* possible C11 impl for integral types
301 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
302 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
303
186 #define ecb_is_constant(expr) 0 304 #define ecb_is_constant(expr) 0
187 #define ecb_expect(expr,value) (expr) 305 #define ecb_expect(expr,value) (expr)
188 #define ecb_prefetch(addr,rw,locality) 306 #define ecb_prefetch(addr,rw,locality)
189#endif 307#endif
190 308
… …
194#elif ECB_GCC_VERSION(3,0) 312#elif ECB_GCC_VERSION(3,0)
195 #define ecb_decltype(x) __typeof(x) 313 #define ecb_decltype(x) __typeof(x)
196#endif 314#endif
197 315
198#define ecb_noinline ecb_attribute ((__noinline__)) 316#define ecb_noinline ecb_attribute ((__noinline__))
199#define ecb_noreturn ecb_attribute ((__noreturn__))
200#define ecb_unused ecb_attribute ((__unused__)) 317#define ecb_unused ecb_attribute ((__unused__))
201#define ecb_const ecb_attribute ((__const__)) 318#define ecb_const ecb_attribute ((__const__))
202#define ecb_pure ecb_attribute ((__pure__)) 319#define ecb_pure ecb_attribute ((__pure__))
320
321#if ECB_C11
322 #define ecb_noreturn _Noreturn
323#else
324 #define ecb_noreturn ecb_attribute ((__noreturn__))
325#endif
203 326
204#if ECB_GCC_VERSION(4,3) 327#if ECB_GCC_VERSION(4,3)
205 #define ecb_artificial ecb_attribute ((__artificial__)) 328 #define ecb_artificial ecb_attribute ((__artificial__))
206 #define ecb_hot ecb_attribute ((__hot__)) 329 #define ecb_hot ecb_attribute ((__hot__))
207 #define ecb_cold ecb_attribute ((__cold__)) 330 #define ecb_cold ecb_attribute ((__cold__))
… …
298 421
299 return r + ecb_ld32 (x); 422 return r + ecb_ld32 (x);
300 } 423 }
301#endif 424#endif
302 425
426ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
427ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
428ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
429ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
430
303ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 431ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
304ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 432ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
305{ 433{
306 return ( (x * 0x0802U & 0x22110U) 434 return ( (x * 0x0802U & 0x22110U)
307 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 435 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
… …
386 514
387#if ECB_GCC_VERSION(4,5) 515#if ECB_GCC_VERSION(4,5)
388 #define ecb_unreachable() __builtin_unreachable () 516 #define ecb_unreachable() __builtin_unreachable ()
389#else 517#else
390 /* this seems to work fine, but gcc always emits a warning for it :/ */ 518 /* this seems to work fine, but gcc always emits a warning for it :/ */
391 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 519 ecb_inline void ecb_unreachable (void) ecb_noreturn;
392 ecb_function_ void ecb_unreachable (void) { } 520 ecb_inline void ecb_unreachable (void) { }
393#endif 521#endif
394 522
395/* try to tell the compiler that some condition is definitely true */ 523/* try to tell the compiler that some condition is definitely true */
396#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 524#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
397 525
398ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 526ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
399ecb_function_ unsigned char 527ecb_inline unsigned char
400ecb_byteorder_helper (void) 528ecb_byteorder_helper (void)
401{ 529{
402 const uint32_t u = 0x11223344; 530 /* the union code still generates code under pressure in gcc, */
403 return *(unsigned char *)&u; 531 /* but less than using pointers, and always seems to */
532 /* successfully return a constant. */
533 /* the reason why we have this horrible preprocessor mess */
534 /* is to avoid it in all cases, at least on common architectures */
535 /* or when using a recent enough gcc version (>= 4.6) */
536#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
537 return 0x44;
538#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
539 return 0x44;
540#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
541 return 0x11;
542#else
543 union
544 {
545 uint32_t i;
546 uint8_t c;
547 } u = { 0x11223344 };
548 return u.c;
549#endif
404} 550}
405 551
406ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 552ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
407ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 553ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
408ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 554ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
409ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 555ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
410 556
411#if ECB_GCC_VERSION(3,0) || ECB_C99 557#if ECB_GCC_VERSION(3,0) || ECB_C99
412 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 558 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
413#else 559#else
414 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 560 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
… …
439 } 585 }
440#else 586#else
441 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 587 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
442#endif 588#endif
443 589
590/*******************************************************************************/
591/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
592
593/* basically, everything uses "ieee pure-endian" floating point numbers */
594/* the only noteworthy exception is ancient armle, which uses order 43218765 */
595#if 0 \
596 || __i386 || __i386__ \
597 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
598 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
599 || defined __arm__ && defined __ARM_EABI__ \
600 || defined __s390__ || defined __s390x__ \
601 || defined __mips__ \
602 || defined __alpha__ \
603 || defined __hppa__ \
604 || defined __ia64__ \
605 || defined __m68k__ \
606 || defined __m88k__ \
607 || defined __sh__ \
608 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
609 #define ECB_STDFP 1
610 #include <string.h> /* for memcpy */
611#else
612 #define ECB_STDFP 0
613#endif
614
615#ifndef ECB_NO_LIBM
616
617 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
618
619 /* only the oldest of old doesn't have this one. solaris. */
620 #ifdef INFINITY
621 #define ECB_INFINITY INFINITY
622 #else
623 #define ECB_INFINITY HUGE_VAL
444#endif 624 #endif
445 625
626 #ifdef NAN
627 #define ECB_NAN NAN
628 #else
629 #define ECB_NAN ECB_INFINITY
630 #endif
631
632 /* converts an ieee half/binary16 to a float */
633 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
634 ecb_function_ float
635 ecb_binary16_to_float (uint16_t x)
636 {
637 int e = (x >> 10) & 0x1f;
638 int m = x & 0x3ff;
639 float r;
640
641 if (!e ) r = ldexpf (m , -24);
642 else if (e != 31) r = ldexpf (m + 0x400, e - 25);
643 else if (m ) r = ECB_NAN;
644 else r = ECB_INFINITY;
645
646 return x & 0x8000 ? -r : r;
647 }
648
649 /* convert a float to ieee single/binary32 */
650 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
651 ecb_function_ uint32_t
652 ecb_float_to_binary32 (float x)
653 {
654 uint32_t r;
655
656 #if ECB_STDFP
657 memcpy (&r, &x, 4);
658 #else
659 /* slow emulation, works for anything but -0 */
660 uint32_t m;
661 int e;
662
663 if (x == 0e0f ) return 0x00000000U;
664 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
665 if (x < -3.40282346638528860e+38f) return 0xff800000U;
666 if (x != x ) return 0x7fbfffffU;
667
668 m = frexpf (x, &e) * 0x1000000U;
669
670 r = m & 0x80000000U;
671
672 if (r)
673 m = -m;
674
675 if (e <= -126)
676 {
677 m &= 0xffffffU;
678 m >>= (-125 - e);
679 e = -126;
680 }
681
682 r |= (e + 126) << 23;
683 r |= m & 0x7fffffU;
684 #endif
685
686 return r;
687 }
688
689 /* converts an ieee single/binary32 to a float */
690 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
691 ecb_function_ float
692 ecb_binary32_to_float (uint32_t x)
693 {
694 float r;
695
696 #if ECB_STDFP
697 memcpy (&r, &x, 4);
698 #else
699 /* emulation, only works for normals and subnormals and +0 */
700 int neg = x >> 31;
701 int e = (x >> 23) & 0xffU;
702
703 x &= 0x7fffffU;
704
705 if (e)
706 x |= 0x800000U;
707 else
708 e = 1;
709
710 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
711 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
712
713 r = neg ? -r : r;
714 #endif
715
716 return r;
717 }
718
719 /* convert a double to ieee double/binary64 */
720 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
721 ecb_function_ uint64_t
722 ecb_double_to_binary64 (double x)
723 {
724 uint64_t r;
725
726 #if ECB_STDFP
727 memcpy (&r, &x, 8);
728 #else
729 /* slow emulation, works for anything but -0 */
730 uint64_t m;
731 int e;
732
733 if (x == 0e0 ) return 0x0000000000000000U;
734 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
735 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
736 if (x != x ) return 0X7ff7ffffffffffffU;
737
738 m = frexp (x, &e) * 0x20000000000000U;
739
740 r = m & 0x8000000000000000;;
741
742 if (r)
743 m = -m;
744
745 if (e <= -1022)
746 {
747 m &= 0x1fffffffffffffU;
748 m >>= (-1021 - e);
749 e = -1022;
750 }
751
752 r |= ((uint64_t)(e + 1022)) << 52;
753 r |= m & 0xfffffffffffffU;
754 #endif
755
756 return r;
757 }
758
759 /* converts an ieee double/binary64 to a double */
760 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
761 ecb_function_ double
762 ecb_binary64_to_double (uint64_t x)
763 {
764 double r;
765
766 #if ECB_STDFP
767 memcpy (&r, &x, 8);
768 #else
769 /* emulation, only works for normals and subnormals and +0 */
770 int neg = x >> 63;
771 int e = (x >> 52) & 0x7ffU;
772
773 x &= 0xfffffffffffffU;
774
775 if (e)
776 x |= 0x10000000000000U;
777 else
778 e = 1;
779
780 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
781 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
782
783 r = neg ? -r : r;
784 #endif
785
786 return r;
787 }
788
789#endif
790
791#endif
792

Diff Legend

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