| 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 (©) 2009-2013 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 enourmous 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 | |
|
|
118 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
|
|
119 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
|
|
120 | |
|
|
121 | #if ECB_NO_THREADS |
|
|
122 | #define ECB_NO_SMP 1 |
|
|
123 | #endif |
|
|
124 | |
|
|
125 | #if ECB_NO_SMP |
|
|
126 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
127 | #endif |
|
|
128 | |
| 68 | #ifndef ECB_MEMORY_FENCE |
129 | #ifndef ECB_MEMORY_FENCE |
| 69 | #if ECB_GCC_VERSION(2,5) |
130 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 70 | #if __x86 |
131 | #if __i386 || __i386__ |
| 71 | #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") |
| 72 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
133 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 73 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
134 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 74 | #elif __amd64 |
135 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
| 75 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
136 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 76 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
137 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 77 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
138 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
139 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
|
|
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
141 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
|
|
142 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
|
|
143 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
|
|
144 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
|
|
145 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
|
|
146 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
147 | #elif (__sparc || __sparc__) && !__sparcv8 |
|
|
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
|
|
149 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
150 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
151 | #elif defined __s390__ || defined __s390x__ |
|
|
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") |
| 78 | #endif |
170 | #endif |
| 79 | #endif |
171 | #endif |
| 80 | #endif |
172 | #endif |
| 81 | |
173 | |
| 82 | #ifndef ECB_MEMORY_FENCE |
174 | #ifndef ECB_MEMORY_FENCE |
| 83 | #if ECB_GCC_VERSION(4,4) |
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 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
179 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
180 | |
|
|
181 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
|
|
182 | * without risking compile time errors with other compilers. We *could* |
|
|
183 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
|
|
184 | * around this shit time and again. |
|
|
185 | * #elif defined __clang && __has_feature (cxx_atomic) |
|
|
186 | * // see comment below (stdatomic.h) about the C11 memory model. |
|
|
187 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
188 | * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
189 | * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
190 | */ |
|
|
191 | |
|
|
192 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 84 | #define ECB_MEMORY_FENCE __sync_synchronize () |
193 | #define ECB_MEMORY_FENCE __sync_synchronize () |
| 85 | #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) |
194 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 86 | #define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) |
195 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
196 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
197 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
198 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
199 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 87 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
200 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 88 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
201 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 89 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
202 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 90 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
203 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 91 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
204 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 92 | #elif defined(_WIN32) |
205 | #elif defined _WIN32 |
| 93 | #include <WinNT.h> |
206 | #include <WinNT.h> |
| 94 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
207 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
| 95 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
208 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 96 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
209 | #include <mbarrier.h> |
|
|
210 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
211 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
212 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
213 | #elif __xlC__ |
|
|
214 | #define ECB_MEMORY_FENCE __sync () |
| 97 | #endif |
215 | #endif |
| 98 | #endif |
216 | #endif |
| 99 | |
217 | |
| 100 | #ifndef ECB_MEMORY_FENCE |
218 | #ifndef ECB_MEMORY_FENCE |
|
|
219 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
220 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
221 | /* not just C11 atomics and atomic accesses */ |
|
|
222 | #include <stdatomic.h> |
|
|
223 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
224 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
225 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
226 | /* for most usages, or gcc and clang have a bug */ |
|
|
227 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
228 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
229 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
230 | /* for all __atomic_thread_fence's except seq_cst */ |
|
|
231 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
232 | #endif |
|
|
233 | #endif |
|
|
234 | |
|
|
235 | #ifndef ECB_MEMORY_FENCE |
|
|
236 | #if !ECB_AVOID_PTHREADS |
| 101 | /* |
237 | /* |
| 102 | * if you get undefined symbol references to pthread_mutex_lock, |
238 | * if you get undefined symbol references to pthread_mutex_lock, |
| 103 | * or failure to find pthread.h, then you should implement |
239 | * or failure to find pthread.h, then you should implement |
| 104 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
240 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
| 105 | * OR proide pthread.h and link against the posix thread library |
241 | * OR provide pthread.h and link against the posix thread library |
| 106 | * of your system. |
242 | * of your system. |
| 107 | */ |
243 | */ |
| 108 | #include <pthread.h> |
244 | #include <pthread.h> |
|
|
245 | #define ECB_NEEDS_PTHREADS 1 |
|
|
246 | #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
| 109 | |
247 | |
| 110 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
248 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
| 111 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
249 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
|
|
250 | #endif |
|
|
251 | #endif |
|
|
252 | |
|
|
253 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
| 112 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
254 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
|
|
255 | #endif |
|
|
256 | |
|
|
257 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 113 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
258 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 114 | #endif |
259 | #endif |
| 115 | |
260 | |
| 116 | /*****************************************************************************/ |
261 | /*****************************************************************************/ |
| 117 | |
|
|
| 118 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
| 119 | |
262 | |
| 120 | #if __cplusplus |
263 | #if __cplusplus |
| 121 | #define ecb_inline static inline |
264 | #define ecb_inline static inline |
| 122 | #elif ECB_GCC_VERSION(2,5) |
265 | #elif ECB_GCC_VERSION(2,5) |
| 123 | #define ecb_inline static __inline__ |
266 | #define ecb_inline static __inline__ |
| … | |
… | |
| 149 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
292 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
| 150 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
293 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
| 151 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
294 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 152 | #else |
295 | #else |
| 153 | #define ecb_attribute(attrlist) |
296 | #define ecb_attribute(attrlist) |
|
|
297 | |
|
|
298 | /* possible C11 impl for integral types |
|
|
299 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
300 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
301 | |
| 154 | #define ecb_is_constant(expr) 0 |
302 | #define ecb_is_constant(expr) 0 |
| 155 | #define ecb_expect(expr,value) (expr) |
303 | #define ecb_expect(expr,value) (expr) |
| 156 | #define ecb_prefetch(addr,rw,locality) |
304 | #define ecb_prefetch(addr,rw,locality) |
| 157 | #endif |
305 | #endif |
| 158 | |
306 | |
| … | |
… | |
| 162 | #elif ECB_GCC_VERSION(3,0) |
310 | #elif ECB_GCC_VERSION(3,0) |
| 163 | #define ecb_decltype(x) __typeof(x) |
311 | #define ecb_decltype(x) __typeof(x) |
| 164 | #endif |
312 | #endif |
| 165 | |
313 | |
| 166 | #define ecb_noinline ecb_attribute ((__noinline__)) |
314 | #define ecb_noinline ecb_attribute ((__noinline__)) |
| 167 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
| 168 | #define ecb_unused ecb_attribute ((__unused__)) |
315 | #define ecb_unused ecb_attribute ((__unused__)) |
| 169 | #define ecb_const ecb_attribute ((__const__)) |
316 | #define ecb_const ecb_attribute ((__const__)) |
| 170 | #define ecb_pure ecb_attribute ((__pure__)) |
317 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
318 | |
|
|
319 | #if ECB_C11 |
|
|
320 | #define ecb_noreturn _Noreturn |
|
|
321 | #else |
|
|
322 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
323 | #endif |
| 171 | |
324 | |
| 172 | #if ECB_GCC_VERSION(4,3) |
325 | #if ECB_GCC_VERSION(4,3) |
| 173 | #define ecb_artificial ecb_attribute ((__artificial__)) |
326 | #define ecb_artificial ecb_attribute ((__artificial__)) |
| 174 | #define ecb_hot ecb_attribute ((__hot__)) |
327 | #define ecb_hot ecb_attribute ((__hot__)) |
| 175 | #define ecb_cold ecb_attribute ((__cold__)) |
328 | #define ecb_cold ecb_attribute ((__cold__)) |
| … | |
… | |
| 241 | x *= 0x01010101; |
394 | x *= 0x01010101; |
| 242 | |
395 | |
| 243 | return x >> 24; |
396 | return x >> 24; |
| 244 | } |
397 | } |
| 245 | |
398 | |
| 246 | /* you have the choice beetween something with a table lookup, */ |
|
|
| 247 | /* something using lots of bit arithmetic and a simple loop */ |
|
|
| 248 | /* we went for the loop */ |
|
|
| 249 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
399 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
| 250 | ecb_function_ int ecb_ld32 (uint32_t x) |
400 | ecb_function_ int ecb_ld32 (uint32_t x) |
| 251 | { |
401 | { |
| 252 | int r = 0; |
402 | int r = 0; |
| 253 | |
403 | |
| … | |
… | |
| 268 | if (x >> 32) { x >>= 32; r += 32; } |
418 | if (x >> 32) { x >>= 32; r += 32; } |
| 269 | |
419 | |
| 270 | return r + ecb_ld32 (x); |
420 | return r + ecb_ld32 (x); |
| 271 | } |
421 | } |
| 272 | #endif |
422 | #endif |
|
|
423 | |
|
|
424 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
425 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
426 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
427 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
428 | |
|
|
429 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
430 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
431 | { |
|
|
432 | return ( (x * 0x0802U & 0x22110U) |
|
|
433 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
434 | } |
|
|
435 | |
|
|
436 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
437 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
438 | { |
|
|
439 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
440 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
441 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
442 | x = ( x >> 8 ) | ( x << 8); |
|
|
443 | |
|
|
444 | return x; |
|
|
445 | } |
|
|
446 | |
|
|
447 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
448 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
449 | { |
|
|
450 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
451 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
452 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
453 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
454 | x = ( x >> 16 ) | ( x << 16); |
|
|
455 | |
|
|
456 | return x; |
|
|
457 | } |
| 273 | |
458 | |
| 274 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
459 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 275 | /* so for this version we are lazy */ |
460 | /* so for this version we are lazy */ |
| 276 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
461 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
| 277 | ecb_function_ int |
462 | ecb_function_ int |
| … | |
… | |
| 327 | |
512 | |
| 328 | #if ECB_GCC_VERSION(4,5) |
513 | #if ECB_GCC_VERSION(4,5) |
| 329 | #define ecb_unreachable() __builtin_unreachable () |
514 | #define ecb_unreachable() __builtin_unreachable () |
| 330 | #else |
515 | #else |
| 331 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
516 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 332 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
517 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
| 333 | ecb_function_ void ecb_unreachable (void) { } |
518 | ecb_inline void ecb_unreachable (void) { } |
| 334 | #endif |
519 | #endif |
| 335 | |
520 | |
| 336 | /* try to tell the compiler that some condition is definitely true */ |
521 | /* try to tell the compiler that some condition is definitely true */ |
| 337 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
522 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 338 | |
523 | |
| 339 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
524 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
| 340 | ecb_function_ unsigned char |
525 | ecb_inline unsigned char |
| 341 | ecb_byteorder_helper (void) |
526 | ecb_byteorder_helper (void) |
| 342 | { |
527 | { |
| 343 | const uint32_t u = 0x11223344; |
528 | /* the union code still generates code under pressure in gcc, */ |
| 344 | return *(unsigned char *)&u; |
529 | /* but less than using pointers, and always seems to */ |
|
|
530 | /* successfully return a constant. */ |
|
|
531 | /* the reason why we have this horrible preprocessor mess */ |
|
|
532 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
533 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
534 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
535 | return 0x44; |
|
|
536 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
537 | return 0x44; |
|
|
538 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
539 | return 0x11; |
|
|
540 | #else |
|
|
541 | union |
|
|
542 | { |
|
|
543 | uint32_t i; |
|
|
544 | uint8_t c; |
|
|
545 | } u = { 0x11223344 }; |
|
|
546 | return u.c; |
|
|
547 | #endif |
| 345 | } |
548 | } |
| 346 | |
549 | |
| 347 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
550 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
| 348 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
551 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
| 349 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
552 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
| 350 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
553 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
| 351 | |
554 | |
| 352 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
555 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
| 353 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
556 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 354 | #else |
557 | #else |
| 355 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
558 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
559 | #endif |
|
|
560 | |
|
|
561 | #if __cplusplus |
|
|
562 | template<typename T> |
|
|
563 | static inline T ecb_div_rd (T val, T div) |
|
|
564 | { |
|
|
565 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
566 | } |
|
|
567 | template<typename T> |
|
|
568 | static inline T ecb_div_ru (T val, T div) |
|
|
569 | { |
|
|
570 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
571 | } |
|
|
572 | #else |
|
|
573 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
574 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 356 | #endif |
575 | #endif |
| 357 | |
576 | |
| 358 | #if ecb_cplusplus_does_not_suck |
577 | #if ecb_cplusplus_does_not_suck |
| 359 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
578 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
| 360 | template<typename T, int N> |
579 | template<typename T, int N> |
| … | |
… | |
| 364 | } |
583 | } |
| 365 | #else |
584 | #else |
| 366 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
585 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 367 | #endif |
586 | #endif |
| 368 | |
587 | |
|
|
588 | /*******************************************************************************/ |
|
|
589 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
590 | |
|
|
591 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
592 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
593 | #if 0 \ |
|
|
594 | || __i386 || __i386__ \ |
|
|
595 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
596 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
597 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
598 | || defined __s390__ || defined __s390x__ \ |
|
|
599 | || defined __mips__ \ |
|
|
600 | || defined __alpha__ \ |
|
|
601 | || defined __hppa__ \ |
|
|
602 | || defined __ia64__ \ |
|
|
603 | || defined __m68k__ \ |
|
|
604 | || defined __m88k__ \ |
|
|
605 | || defined __sh__ \ |
|
|
606 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
|
|
607 | #define ECB_STDFP 1 |
|
|
608 | #include <string.h> /* for memcpy */ |
|
|
609 | #else |
|
|
610 | #define ECB_STDFP 0 |
|
|
611 | #endif |
|
|
612 | |
|
|
613 | #ifndef ECB_NO_LIBM |
|
|
614 | |
|
|
615 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
616 | |
|
|
617 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
618 | #ifdef INFINITY |
|
|
619 | #define ECB_INFINITY INFINITY |
|
|
620 | #else |
|
|
621 | #define ECB_INFINITY HUGE_VAL |
| 369 | #endif |
622 | #endif |
| 370 | |
623 | |
|
|
624 | #ifdef NAN |
|
|
625 | #define ECB_NAN NAN |
|
|
626 | #else |
|
|
627 | #define ECB_NAN ECB_INFINITY |
|
|
628 | #endif |
|
|
629 | |
|
|
630 | /* converts an ieee half/binary16 to a float */ |
|
|
631 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
632 | ecb_function_ float |
|
|
633 | ecb_binary16_to_float (uint16_t x) |
|
|
634 | { |
|
|
635 | int e = (x >> 10) & 0x1f; |
|
|
636 | int m = x & 0x3ff; |
|
|
637 | float r; |
|
|
638 | |
|
|
639 | if (!e ) r = ldexpf (m , -24); |
|
|
640 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
641 | else if (m ) r = ECB_NAN; |
|
|
642 | else r = ECB_INFINITY; |
|
|
643 | |
|
|
644 | return x & 0x8000 ? -r : r; |
|
|
645 | } |
|
|
646 | |
|
|
647 | /* convert a float to ieee single/binary32 */ |
|
|
648 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
649 | ecb_function_ uint32_t |
|
|
650 | ecb_float_to_binary32 (float x) |
|
|
651 | { |
|
|
652 | uint32_t r; |
|
|
653 | |
|
|
654 | #if ECB_STDFP |
|
|
655 | memcpy (&r, &x, 4); |
|
|
656 | #else |
|
|
657 | /* slow emulation, works for anything but -0 */ |
|
|
658 | uint32_t m; |
|
|
659 | int e; |
|
|
660 | |
|
|
661 | if (x == 0e0f ) return 0x00000000U; |
|
|
662 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
663 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
664 | if (x != x ) return 0x7fbfffffU; |
|
|
665 | |
|
|
666 | m = frexpf (x, &e) * 0x1000000U; |
|
|
667 | |
|
|
668 | r = m & 0x80000000U; |
|
|
669 | |
|
|
670 | if (r) |
|
|
671 | m = -m; |
|
|
672 | |
|
|
673 | if (e <= -126) |
|
|
674 | { |
|
|
675 | m &= 0xffffffU; |
|
|
676 | m >>= (-125 - e); |
|
|
677 | e = -126; |
|
|
678 | } |
|
|
679 | |
|
|
680 | r |= (e + 126) << 23; |
|
|
681 | r |= m & 0x7fffffU; |
|
|
682 | #endif |
|
|
683 | |
|
|
684 | return r; |
|
|
685 | } |
|
|
686 | |
|
|
687 | /* converts an ieee single/binary32 to a float */ |
|
|
688 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
689 | ecb_function_ float |
|
|
690 | ecb_binary32_to_float (uint32_t x) |
|
|
691 | { |
|
|
692 | float r; |
|
|
693 | |
|
|
694 | #if ECB_STDFP |
|
|
695 | memcpy (&r, &x, 4); |
|
|
696 | #else |
|
|
697 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
698 | int neg = x >> 31; |
|
|
699 | int e = (x >> 23) & 0xffU; |
|
|
700 | |
|
|
701 | x &= 0x7fffffU; |
|
|
702 | |
|
|
703 | if (e) |
|
|
704 | x |= 0x800000U; |
|
|
705 | else |
|
|
706 | e = 1; |
|
|
707 | |
|
|
708 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
709 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
710 | |
|
|
711 | r = neg ? -r : r; |
|
|
712 | #endif |
|
|
713 | |
|
|
714 | return r; |
|
|
715 | } |
|
|
716 | |
|
|
717 | /* convert a double to ieee double/binary64 */ |
|
|
718 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
719 | ecb_function_ uint64_t |
|
|
720 | ecb_double_to_binary64 (double x) |
|
|
721 | { |
|
|
722 | uint64_t r; |
|
|
723 | |
|
|
724 | #if ECB_STDFP |
|
|
725 | memcpy (&r, &x, 8); |
|
|
726 | #else |
|
|
727 | /* slow emulation, works for anything but -0 */ |
|
|
728 | uint64_t m; |
|
|
729 | int e; |
|
|
730 | |
|
|
731 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
732 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
733 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
734 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
735 | |
|
|
736 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
737 | |
|
|
738 | r = m & 0x8000000000000000;; |
|
|
739 | |
|
|
740 | if (r) |
|
|
741 | m = -m; |
|
|
742 | |
|
|
743 | if (e <= -1022) |
|
|
744 | { |
|
|
745 | m &= 0x1fffffffffffffU; |
|
|
746 | m >>= (-1021 - e); |
|
|
747 | e = -1022; |
|
|
748 | } |
|
|
749 | |
|
|
750 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
751 | r |= m & 0xfffffffffffffU; |
|
|
752 | #endif |
|
|
753 | |
|
|
754 | return r; |
|
|
755 | } |
|
|
756 | |
|
|
757 | /* converts an ieee double/binary64 to a double */ |
|
|
758 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
759 | ecb_function_ double |
|
|
760 | ecb_binary64_to_double (uint64_t x) |
|
|
761 | { |
|
|
762 | double r; |
|
|
763 | |
|
|
764 | #if ECB_STDFP |
|
|
765 | memcpy (&r, &x, 8); |
|
|
766 | #else |
|
|
767 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
768 | int neg = x >> 63; |
|
|
769 | int e = (x >> 52) & 0x7ffU; |
|
|
770 | |
|
|
771 | x &= 0xfffffffffffffU; |
|
|
772 | |
|
|
773 | if (e) |
|
|
774 | x |= 0x10000000000000U; |
|
|
775 | else |
|
|
776 | e = 1; |
|
|
777 | |
|
|
778 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
779 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
780 | |
|
|
781 | r = neg ? -r : r; |
|
|
782 | #endif |
|
|
783 | |
|
|
784 | return r; |
|
|
785 | } |
|
|
786 | |
|
|
787 | #endif |
|
|
788 | |
|
|
789 | #endif |
|
|
790 | |
