| 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 (©) 2009-2014 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: |
| … | |
… | |
| 23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
| 24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
| 25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
| 26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
| 27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
28 | * |
|
|
29 | * Alternatively, the contents of this file may be used under the terms of |
|
|
30 | * the GNU General Public License ("GPL") version 2 or any later version, |
|
|
31 | * in which case the provisions of the GPL are applicable instead of |
|
|
32 | * the above. If you wish to allow the use of your version of this file |
|
|
33 | * only under the terms of the GPL and not to allow others to use your |
|
|
34 | * version of this file under the BSD license, indicate your decision |
|
|
35 | * by deleting the provisions above and replace them with the notice |
|
|
36 | * and other provisions required by the GPL. If you do not delete the |
|
|
37 | * provisions above, a recipient may use your version of this file under |
|
|
38 | * either the BSD or the GPL. |
| 28 | */ |
39 | */ |
| 29 | |
40 | |
| 30 | #ifndef ECB_H |
41 | #ifndef ECB_H |
| 31 | #define ECB_H |
42 | #define ECB_H |
|
|
43 | |
|
|
44 | /* 16 bits major, 16 bits minor */ |
|
|
45 | #define ECB_VERSION 0x00010003 |
| 32 | |
46 | |
| 33 | #ifdef _WIN32 |
47 | #ifdef _WIN32 |
| 34 | typedef signed char int8_t; |
48 | typedef signed char int8_t; |
| 35 | typedef unsigned char uint8_t; |
49 | typedef unsigned char uint8_t; |
| 36 | typedef signed short int16_t; |
50 | typedef signed short int16_t; |
| … | |
… | |
| 42 | typedef unsigned long long uint64_t; |
56 | typedef unsigned long long uint64_t; |
| 43 | #else /* _MSC_VER || __BORLANDC__ */ |
57 | #else /* _MSC_VER || __BORLANDC__ */ |
| 44 | typedef signed __int64 int64_t; |
58 | typedef signed __int64 int64_t; |
| 45 | typedef unsigned __int64 uint64_t; |
59 | typedef unsigned __int64 uint64_t; |
| 46 | #endif |
60 | #endif |
|
|
61 | #ifdef _WIN64 |
|
|
62 | #define ECB_PTRSIZE 8 |
|
|
63 | typedef uint64_t uintptr_t; |
|
|
64 | typedef int64_t intptr_t; |
|
|
65 | #else |
|
|
66 | #define ECB_PTRSIZE 4 |
|
|
67 | typedef uint32_t uintptr_t; |
|
|
68 | typedef int32_t intptr_t; |
|
|
69 | #endif |
| 47 | #else |
70 | #else |
| 48 | #include <inttypes.h> |
71 | #include <inttypes.h> |
|
|
72 | #if UINTMAX_MAX > 0xffffffffU |
|
|
73 | #define ECB_PTRSIZE 8 |
|
|
74 | #else |
|
|
75 | #define ECB_PTRSIZE 4 |
|
|
76 | #endif |
|
|
77 | #endif |
|
|
78 | |
|
|
79 | /* work around x32 idiocy by defining proper macros */ |
|
|
80 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
|
|
81 | #if _ILP32 |
|
|
82 | #define ECB_AMD64_X32 1 |
|
|
83 | #else |
|
|
84 | #define ECB_AMD64 1 |
|
|
85 | #endif |
| 49 | #endif |
86 | #endif |
| 50 | |
87 | |
| 51 | /* many compilers define _GNUC_ to some versions but then only implement |
88 | /* many compilers define _GNUC_ to some versions but then only implement |
| 52 | * what their idiot authors think are the "more important" extensions, |
89 | * what their idiot authors think are the "more important" extensions, |
| 53 | * causing enormous grief in return for some better fake benchmark numbers. |
90 | * causing enormous grief in return for some better fake benchmark numbers. |
| 54 | * or so. |
91 | * or so. |
| 55 | * we try to detect these and simply assume they are not gcc - if they have |
92 | * 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. |
93 | * an issue with that they should have done it right in the first place. |
| 57 | */ |
94 | */ |
| 58 | #ifndef ECB_GCC_VERSION |
95 | #ifndef ECB_GCC_VERSION |
| 59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
96 | #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 |
97 | #define ECB_GCC_VERSION(major,minor) 0 |
| 61 | #else |
98 | #else |
| 62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
99 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 63 | #endif |
100 | #endif |
| 64 | #endif |
101 | #endif |
| 65 | |
102 | |
|
|
103 | #define ECB_CPP (__cplusplus+0) |
|
|
104 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
105 | |
|
|
106 | #if ECB_CPP |
|
|
107 | #define ECB_C 0 |
|
|
108 | #define ECB_STDC_VERSION 0 |
|
|
109 | #else |
|
|
110 | #define ECB_C 1 |
|
|
111 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
112 | #endif |
|
|
113 | |
|
|
114 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
115 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
116 | |
|
|
117 | #if ECB_CPP |
|
|
118 | #define ECB_EXTERN_C extern "C" |
|
|
119 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
|
|
120 | #define ECB_EXTERN_C_END } |
|
|
121 | #else |
|
|
122 | #define ECB_EXTERN_C extern |
|
|
123 | #define ECB_EXTERN_C_BEG |
|
|
124 | #define ECB_EXTERN_C_END |
|
|
125 | #endif |
|
|
126 | |
| 66 | /*****************************************************************************/ |
127 | /*****************************************************************************/ |
| 67 | |
128 | |
| 68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
129 | /* 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 */ |
130 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
| 70 | |
131 | |
| 71 | #if ECB_NO_THREADS || ECB_NO_SMP |
132 | #if ECB_NO_THREADS |
|
|
133 | #define ECB_NO_SMP 1 |
|
|
134 | #endif |
|
|
135 | |
|
|
136 | #if ECB_NO_SMP |
| 72 | #define ECB_MEMORY_FENCE do { } while (0) |
137 | #define ECB_MEMORY_FENCE do { } while (0) |
| 73 | #endif |
138 | #endif |
| 74 | |
139 | |
| 75 | #ifndef ECB_MEMORY_FENCE |
140 | #ifndef ECB_MEMORY_FENCE |
| 76 | #if ECB_GCC_VERSION(2,5) |
141 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 77 | #if __i386__ |
142 | #if __i386 || __i386__ |
| 78 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
143 | #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 */ |
144 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 80 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
145 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 81 | #elif __amd64 |
146 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
| 82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
147 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 83 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
148 | #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 */ |
149 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 85 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
150 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
| 86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
151 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
| 87 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
152 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
| 88 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) |
153 | || 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") |
154 | #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__ ) \ |
155 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
| 91 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
156 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
| 92 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
157 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
158 | #elif __aarch64__ |
|
|
159 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
|
|
160 | #elif (__sparc || __sparc__) && !__sparcv8 |
|
|
161 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
|
|
162 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
163 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
164 | #elif defined __s390__ || defined __s390x__ |
|
|
165 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
166 | #elif defined __mips__ |
|
|
167 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
168 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
|
|
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
|
|
170 | #elif defined __alpha__ |
|
|
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
172 | #elif defined __hppa__ |
|
|
173 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
174 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
175 | #elif defined __ia64__ |
|
|
176 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
177 | #elif defined __m68k__ |
|
|
178 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
179 | #elif defined __m88k__ |
|
|
180 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
181 | #elif defined __sh__ |
|
|
182 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 93 | #endif |
183 | #endif |
| 94 | #endif |
184 | #endif |
| 95 | #endif |
185 | #endif |
| 96 | |
186 | |
| 97 | #ifndef ECB_MEMORY_FENCE |
187 | #ifndef ECB_MEMORY_FENCE |
|
|
188 | #if ECB_GCC_VERSION(4,7) |
|
|
189 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
190 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
191 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
192 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
193 | |
|
|
194 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
|
|
195 | * without risking compile time errors with other compilers. We *could* |
|
|
196 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
|
|
197 | * around this shit time and again. |
|
|
198 | * #elif defined __clang && __has_feature (cxx_atomic) |
|
|
199 | * // see comment below (stdatomic.h) about the C11 memory model. |
|
|
200 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
201 | * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
202 | * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
203 | */ |
|
|
204 | |
| 98 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) |
205 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 99 | #define ECB_MEMORY_FENCE __sync_synchronize () |
206 | #define ECB_MEMORY_FENCE __sync_synchronize () |
| 100 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
207 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 101 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
208 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
209 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
210 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
211 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
212 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 102 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
213 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 103 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
214 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 104 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
215 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 105 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
216 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 106 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
217 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 107 | #elif defined(_WIN32) |
218 | #elif defined _WIN32 |
| 108 | #include <WinNT.h> |
219 | #include <WinNT.h> |
| 109 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
220 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
221 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
222 | #include <mbarrier.h> |
|
|
223 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
224 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
225 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
226 | #elif __xlC__ |
|
|
227 | #define ECB_MEMORY_FENCE __sync () |
|
|
228 | #endif |
|
|
229 | #endif |
|
|
230 | |
|
|
231 | #ifndef ECB_MEMORY_FENCE |
|
|
232 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
233 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
234 | /* not just C11 atomics and atomic accesses */ |
|
|
235 | #include <stdatomic.h> |
|
|
236 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
237 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
238 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
239 | /* for most usages, or gcc and clang have a bug */ |
|
|
240 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
241 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
242 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
243 | /* for all __atomic_thread_fence's except seq_cst */ |
|
|
244 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
| 110 | #endif |
245 | #endif |
| 111 | #endif |
246 | #endif |
| 112 | |
247 | |
| 113 | #ifndef ECB_MEMORY_FENCE |
248 | #ifndef ECB_MEMORY_FENCE |
| 114 | #if !ECB_AVOID_PTHREADS |
249 | #if !ECB_AVOID_PTHREADS |
| … | |
… | |
| 126 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
261 | 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) |
262 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
| 128 | #endif |
263 | #endif |
| 129 | #endif |
264 | #endif |
| 130 | |
265 | |
| 131 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
266 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
| 132 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
267 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
| 133 | #endif |
268 | #endif |
| 134 | |
269 | |
| 135 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
270 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 136 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
271 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 137 | #endif |
272 | #endif |
| 138 | |
273 | |
| 139 | /*****************************************************************************/ |
274 | /*****************************************************************************/ |
| 140 | |
|
|
| 141 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
| 142 | |
275 | |
| 143 | #if __cplusplus |
276 | #if __cplusplus |
| 144 | #define ecb_inline static inline |
277 | #define ecb_inline static inline |
| 145 | #elif ECB_GCC_VERSION(2,5) |
278 | #elif ECB_GCC_VERSION(2,5) |
| 146 | #define ecb_inline static __inline__ |
279 | #define ecb_inline static __inline__ |
| … | |
… | |
| 172 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
305 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
| 173 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
306 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
| 174 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
307 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 175 | #else |
308 | #else |
| 176 | #define ecb_attribute(attrlist) |
309 | #define ecb_attribute(attrlist) |
|
|
310 | |
|
|
311 | /* possible C11 impl for integral types |
|
|
312 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
313 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
314 | |
| 177 | #define ecb_is_constant(expr) 0 |
315 | #define ecb_is_constant(expr) 0 |
| 178 | #define ecb_expect(expr,value) (expr) |
316 | #define ecb_expect(expr,value) (expr) |
| 179 | #define ecb_prefetch(addr,rw,locality) |
317 | #define ecb_prefetch(addr,rw,locality) |
| 180 | #endif |
318 | #endif |
| 181 | |
319 | |
| … | |
… | |
| 184 | #define ecb_decltype(x) __decltype(x) |
322 | #define ecb_decltype(x) __decltype(x) |
| 185 | #elif ECB_GCC_VERSION(3,0) |
323 | #elif ECB_GCC_VERSION(3,0) |
| 186 | #define ecb_decltype(x) __typeof(x) |
324 | #define ecb_decltype(x) __typeof(x) |
| 187 | #endif |
325 | #endif |
| 188 | |
326 | |
|
|
327 | #if _MSC_VER >= 1300 |
|
|
328 | #define ecb_deprecated __declspec(deprecated) |
|
|
329 | #else |
|
|
330 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
|
|
331 | #endif |
|
|
332 | |
| 189 | #define ecb_noinline ecb_attribute ((__noinline__)) |
333 | #define ecb_noinline ecb_attribute ((__noinline__)) |
| 190 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
| 191 | #define ecb_unused ecb_attribute ((__unused__)) |
334 | #define ecb_unused ecb_attribute ((__unused__)) |
| 192 | #define ecb_const ecb_attribute ((__const__)) |
335 | #define ecb_const ecb_attribute ((__const__)) |
| 193 | #define ecb_pure ecb_attribute ((__pure__)) |
336 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
337 | |
|
|
338 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */ |
|
|
339 | #if ECB_C11 |
|
|
340 | #define ecb_noreturn _Noreturn |
|
|
341 | #else |
|
|
342 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
343 | #endif |
| 194 | |
344 | |
| 195 | #if ECB_GCC_VERSION(4,3) |
345 | #if ECB_GCC_VERSION(4,3) |
| 196 | #define ecb_artificial ecb_attribute ((__artificial__)) |
346 | #define ecb_artificial ecb_attribute ((__artificial__)) |
| 197 | #define ecb_hot ecb_attribute ((__hot__)) |
347 | #define ecb_hot ecb_attribute ((__hot__)) |
| 198 | #define ecb_cold ecb_attribute ((__cold__)) |
348 | #define ecb_cold ecb_attribute ((__cold__)) |
| … | |
… | |
| 289 | |
439 | |
| 290 | return r + ecb_ld32 (x); |
440 | return r + ecb_ld32 (x); |
| 291 | } |
441 | } |
| 292 | #endif |
442 | #endif |
| 293 | |
443 | |
|
|
444 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
445 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
446 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
447 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
448 | |
|
|
449 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
450 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
451 | { |
|
|
452 | return ( (x * 0x0802U & 0x22110U) |
|
|
453 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
454 | } |
|
|
455 | |
|
|
456 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
457 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
458 | { |
|
|
459 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
460 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
461 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
462 | x = ( x >> 8 ) | ( x << 8); |
|
|
463 | |
|
|
464 | return x; |
|
|
465 | } |
|
|
466 | |
|
|
467 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
468 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
469 | { |
|
|
470 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
471 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
472 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
473 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
474 | x = ( x >> 16 ) | ( x << 16); |
|
|
475 | |
|
|
476 | return x; |
|
|
477 | } |
|
|
478 | |
| 294 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
479 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 295 | /* so for this version we are lazy */ |
480 | /* so for this version we are lazy */ |
| 296 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
481 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
| 297 | ecb_function_ int |
482 | ecb_function_ int |
| 298 | ecb_popcount64 (uint64_t x) |
483 | ecb_popcount64 (uint64_t x) |
| … | |
… | |
| 347 | |
532 | |
| 348 | #if ECB_GCC_VERSION(4,5) |
533 | #if ECB_GCC_VERSION(4,5) |
| 349 | #define ecb_unreachable() __builtin_unreachable () |
534 | #define ecb_unreachable() __builtin_unreachable () |
| 350 | #else |
535 | #else |
| 351 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
536 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 352 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
537 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
| 353 | ecb_function_ void ecb_unreachable (void) { } |
538 | ecb_inline void ecb_unreachable (void) { } |
| 354 | #endif |
539 | #endif |
| 355 | |
540 | |
| 356 | /* try to tell the compiler that some condition is definitely true */ |
541 | /* try to tell the compiler that some condition is definitely true */ |
| 357 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
542 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 358 | |
543 | |
| 359 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
544 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
| 360 | ecb_function_ unsigned char |
545 | ecb_inline unsigned char |
| 361 | ecb_byteorder_helper (void) |
546 | ecb_byteorder_helper (void) |
| 362 | { |
547 | { |
| 363 | const uint32_t u = 0x11223344; |
548 | /* the union code still generates code under pressure in gcc, */ |
| 364 | return *(unsigned char *)&u; |
549 | /* but less than using pointers, and always seems to */ |
|
|
550 | /* successfully return a constant. */ |
|
|
551 | /* the reason why we have this horrible preprocessor mess */ |
|
|
552 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
553 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
554 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
555 | return 0x44; |
|
|
556 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
557 | return 0x44; |
|
|
558 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
559 | return 0x11; |
|
|
560 | #else |
|
|
561 | union |
|
|
562 | { |
|
|
563 | uint32_t i; |
|
|
564 | uint8_t c; |
|
|
565 | } u = { 0x11223344 }; |
|
|
566 | return u.c; |
|
|
567 | #endif |
| 365 | } |
568 | } |
| 366 | |
569 | |
| 367 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
570 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
| 368 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
571 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
| 369 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
572 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
| 370 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
573 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
| 371 | |
574 | |
| 372 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
575 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
| 373 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
576 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 374 | #else |
577 | #else |
| 375 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
578 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
579 | #endif |
|
|
580 | |
|
|
581 | #if __cplusplus |
|
|
582 | template<typename T> |
|
|
583 | static inline T ecb_div_rd (T val, T div) |
|
|
584 | { |
|
|
585 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
586 | } |
|
|
587 | template<typename T> |
|
|
588 | static inline T ecb_div_ru (T val, T div) |
|
|
589 | { |
|
|
590 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
591 | } |
|
|
592 | #else |
|
|
593 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
594 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 376 | #endif |
595 | #endif |
| 377 | |
596 | |
| 378 | #if ecb_cplusplus_does_not_suck |
597 | #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) */ |
598 | /* 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> |
599 | template<typename T, int N> |
| … | |
… | |
| 384 | } |
603 | } |
| 385 | #else |
604 | #else |
| 386 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
605 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 387 | #endif |
606 | #endif |
| 388 | |
607 | |
|
|
608 | /*******************************************************************************/ |
|
|
609 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
610 | |
|
|
611 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
612 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
613 | #if 0 \ |
|
|
614 | || __i386 || __i386__ \ |
|
|
615 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
616 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
617 | || defined __s390__ || defined __s390x__ \ |
|
|
618 | || defined __mips__ \ |
|
|
619 | || defined __alpha__ \ |
|
|
620 | || defined __hppa__ \ |
|
|
621 | || defined __ia64__ \ |
|
|
622 | || defined __m68k__ \ |
|
|
623 | || defined __m88k__ \ |
|
|
624 | || defined __sh__ \ |
|
|
625 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
626 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
627 | || defined __aarch64__ |
|
|
628 | #define ECB_STDFP 1 |
|
|
629 | #include <string.h> /* for memcpy */ |
|
|
630 | #else |
|
|
631 | #define ECB_STDFP 0 |
|
|
632 | #endif |
|
|
633 | |
|
|
634 | #ifndef ECB_NO_LIBM |
|
|
635 | |
|
|
636 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
637 | |
|
|
638 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
639 | #ifdef INFINITY |
|
|
640 | #define ECB_INFINITY INFINITY |
|
|
641 | #else |
|
|
642 | #define ECB_INFINITY HUGE_VAL |
| 389 | #endif |
643 | #endif |
| 390 | |
644 | |
|
|
645 | #ifdef NAN |
|
|
646 | #define ECB_NAN NAN |
|
|
647 | #else |
|
|
648 | #define ECB_NAN ECB_INFINITY |
|
|
649 | #endif |
|
|
650 | |
|
|
651 | /* converts an ieee half/binary16 to a float */ |
|
|
652 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
653 | ecb_function_ float |
|
|
654 | ecb_binary16_to_float (uint16_t x) |
|
|
655 | { |
|
|
656 | int e = (x >> 10) & 0x1f; |
|
|
657 | int m = x & 0x3ff; |
|
|
658 | float r; |
|
|
659 | |
|
|
660 | if (!e ) r = ldexpf (m , -24); |
|
|
661 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
662 | else if (m ) r = ECB_NAN; |
|
|
663 | else r = ECB_INFINITY; |
|
|
664 | |
|
|
665 | return x & 0x8000 ? -r : r; |
|
|
666 | } |
|
|
667 | |
|
|
668 | /* convert a float to ieee single/binary32 */ |
|
|
669 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
670 | ecb_function_ uint32_t |
|
|
671 | ecb_float_to_binary32 (float x) |
|
|
672 | { |
|
|
673 | uint32_t r; |
|
|
674 | |
|
|
675 | #if ECB_STDFP |
|
|
676 | memcpy (&r, &x, 4); |
|
|
677 | #else |
|
|
678 | /* slow emulation, works for anything but -0 */ |
|
|
679 | uint32_t m; |
|
|
680 | int e; |
|
|
681 | |
|
|
682 | if (x == 0e0f ) return 0x00000000U; |
|
|
683 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
684 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
685 | if (x != x ) return 0x7fbfffffU; |
|
|
686 | |
|
|
687 | m = frexpf (x, &e) * 0x1000000U; |
|
|
688 | |
|
|
689 | r = m & 0x80000000U; |
|
|
690 | |
|
|
691 | if (r) |
|
|
692 | m = -m; |
|
|
693 | |
|
|
694 | if (e <= -126) |
|
|
695 | { |
|
|
696 | m &= 0xffffffU; |
|
|
697 | m >>= (-125 - e); |
|
|
698 | e = -126; |
|
|
699 | } |
|
|
700 | |
|
|
701 | r |= (e + 126) << 23; |
|
|
702 | r |= m & 0x7fffffU; |
|
|
703 | #endif |
|
|
704 | |
|
|
705 | return r; |
|
|
706 | } |
|
|
707 | |
|
|
708 | /* converts an ieee single/binary32 to a float */ |
|
|
709 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
710 | ecb_function_ float |
|
|
711 | ecb_binary32_to_float (uint32_t x) |
|
|
712 | { |
|
|
713 | float r; |
|
|
714 | |
|
|
715 | #if ECB_STDFP |
|
|
716 | memcpy (&r, &x, 4); |
|
|
717 | #else |
|
|
718 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
719 | int neg = x >> 31; |
|
|
720 | int e = (x >> 23) & 0xffU; |
|
|
721 | |
|
|
722 | x &= 0x7fffffU; |
|
|
723 | |
|
|
724 | if (e) |
|
|
725 | x |= 0x800000U; |
|
|
726 | else |
|
|
727 | e = 1; |
|
|
728 | |
|
|
729 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
730 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
731 | |
|
|
732 | r = neg ? -r : r; |
|
|
733 | #endif |
|
|
734 | |
|
|
735 | return r; |
|
|
736 | } |
|
|
737 | |
|
|
738 | /* convert a double to ieee double/binary64 */ |
|
|
739 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
740 | ecb_function_ uint64_t |
|
|
741 | ecb_double_to_binary64 (double x) |
|
|
742 | { |
|
|
743 | uint64_t r; |
|
|
744 | |
|
|
745 | #if ECB_STDFP |
|
|
746 | memcpy (&r, &x, 8); |
|
|
747 | #else |
|
|
748 | /* slow emulation, works for anything but -0 */ |
|
|
749 | uint64_t m; |
|
|
750 | int e; |
|
|
751 | |
|
|
752 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
753 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
754 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
755 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
756 | |
|
|
757 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
758 | |
|
|
759 | r = m & 0x8000000000000000;; |
|
|
760 | |
|
|
761 | if (r) |
|
|
762 | m = -m; |
|
|
763 | |
|
|
764 | if (e <= -1022) |
|
|
765 | { |
|
|
766 | m &= 0x1fffffffffffffU; |
|
|
767 | m >>= (-1021 - e); |
|
|
768 | e = -1022; |
|
|
769 | } |
|
|
770 | |
|
|
771 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
772 | r |= m & 0xfffffffffffffU; |
|
|
773 | #endif |
|
|
774 | |
|
|
775 | return r; |
|
|
776 | } |
|
|
777 | |
|
|
778 | /* converts an ieee double/binary64 to a double */ |
|
|
779 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
780 | ecb_function_ double |
|
|
781 | ecb_binary64_to_double (uint64_t x) |
|
|
782 | { |
|
|
783 | double r; |
|
|
784 | |
|
|
785 | #if ECB_STDFP |
|
|
786 | memcpy (&r, &x, 8); |
|
|
787 | #else |
|
|
788 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
789 | int neg = x >> 63; |
|
|
790 | int e = (x >> 52) & 0x7ffU; |
|
|
791 | |
|
|
792 | x &= 0xfffffffffffffU; |
|
|
793 | |
|
|
794 | if (e) |
|
|
795 | x |= 0x10000000000000U; |
|
|
796 | else |
|
|
797 | e = 1; |
|
|
798 | |
|
|
799 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
800 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
801 | |
|
|
802 | r = neg ? -r : r; |
|
|
803 | #endif |
|
|
804 | |
|
|
805 | return r; |
|
|
806 | } |
|
|
807 | |
|
|
808 | #endif |
|
|
809 | |
|
|
810 | #endif |
|
|
811 | |
