| 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 (©) 2009-2015 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 0x00010004 |
| 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 enourmous 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 |
|
|
| 59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
95 | #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 |
96 | #define ECB_GCC_VERSION(major,minor) 0 |
| 61 | #else |
97 | #else |
| 62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
98 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 63 | #endif |
99 | #endif |
|
|
100 | |
|
|
101 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
|
|
102 | |
|
|
103 | #if __clang__ && defined __has_builtin |
|
|
104 | #define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
|
|
105 | #else |
|
|
106 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
107 | #endif |
|
|
108 | |
|
|
109 | #if __clang__ && defined __has_extension |
|
|
110 | #define ECB_CLANG_EXTENSION(x) __has_extension (x) |
|
|
111 | #else |
|
|
112 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
113 | #endif |
|
|
114 | |
|
|
115 | #define ECB_CPP (__cplusplus+0) |
|
|
116 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
117 | |
|
|
118 | #if ECB_CPP |
|
|
119 | #define ECB_C 0 |
|
|
120 | #define ECB_STDC_VERSION 0 |
|
|
121 | #else |
|
|
122 | #define ECB_C 1 |
|
|
123 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
124 | #endif |
|
|
125 | |
|
|
126 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
127 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
128 | |
|
|
129 | #if ECB_CPP |
|
|
130 | #define ECB_EXTERN_C extern "C" |
|
|
131 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
|
|
132 | #define ECB_EXTERN_C_END } |
|
|
133 | #else |
|
|
134 | #define ECB_EXTERN_C extern |
|
|
135 | #define ECB_EXTERN_C_BEG |
|
|
136 | #define ECB_EXTERN_C_END |
| 64 | #endif |
137 | #endif |
| 65 | |
138 | |
| 66 | /*****************************************************************************/ |
139 | /*****************************************************************************/ |
| 67 | |
140 | |
|
|
141 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
|
|
142 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
|
|
143 | |
|
|
144 | #if ECB_NO_THREADS |
|
|
145 | #define ECB_NO_SMP 1 |
|
|
146 | #endif |
|
|
147 | |
|
|
148 | #if ECB_NO_SMP |
|
|
149 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
150 | #endif |
|
|
151 | |
| 68 | #ifndef ECB_MEMORY_FENCE |
152 | #ifndef ECB_MEMORY_FENCE |
| 69 | #if ECB_GCC_VERSION(2,5) |
153 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 70 | #if __x86 |
154 | #if __i386 || __i386__ |
| 71 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
155 | #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 */ |
156 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 73 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
157 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 74 | #elif __amd64 |
158 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
| 75 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
159 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 76 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
160 | #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 */ |
161 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
162 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
|
|
163 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
164 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
|
|
165 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
|
|
166 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
|
|
167 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
|
|
168 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
|
|
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
170 | #elif __aarch64__ |
|
|
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
|
|
172 | #elif (__sparc || __sparc__) && !__sparcv8 |
|
|
173 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
|
|
174 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
175 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
176 | #elif defined __s390__ || defined __s390x__ |
|
|
177 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
178 | #elif defined __mips__ |
|
|
179 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
180 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
|
|
181 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
|
|
182 | #elif defined __alpha__ |
|
|
183 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
184 | #elif defined __hppa__ |
|
|
185 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
186 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
187 | #elif defined __ia64__ |
|
|
188 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
189 | #elif defined __m68k__ |
|
|
190 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
191 | #elif defined __m88k__ |
|
|
192 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
193 | #elif defined __sh__ |
|
|
194 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 78 | #endif |
195 | #endif |
| 79 | #endif |
196 | #endif |
| 80 | #endif |
197 | #endif |
| 81 | |
198 | |
| 82 | #ifndef ECB_MEMORY_FENCE |
199 | #ifndef ECB_MEMORY_FENCE |
| 83 | #if ECB_GCC_VERSION(4,4) |
200 | #if ECB_GCC_VERSION(4,7) |
|
|
201 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
202 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
203 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
204 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
205 | |
|
|
206 | #elif ECB_CLANG_EXTENSION(c_atomic) |
|
|
207 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
208 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
209 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
210 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
211 | |
|
|
212 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 84 | #define ECB_MEMORY_FENCE __sync_synchronize () |
213 | #define ECB_MEMORY_FENCE __sync_synchronize () |
| 85 | #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) |
214 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 86 | #define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) |
215 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
216 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
217 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
218 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
219 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 87 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
220 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 88 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
221 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 89 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
222 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 90 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
223 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 91 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
224 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 92 | #elif defined(_WIN32) |
225 | #elif defined _WIN32 |
| 93 | #include <WinNT.h> |
226 | #include <WinNT.h> |
| 94 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
227 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
| 95 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
228 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 96 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
229 | #include <mbarrier.h> |
|
|
230 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
231 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
232 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
233 | #elif __xlC__ |
|
|
234 | #define ECB_MEMORY_FENCE __sync () |
| 97 | #endif |
235 | #endif |
| 98 | #endif |
236 | #endif |
| 99 | |
237 | |
| 100 | #ifndef ECB_MEMORY_FENCE |
238 | #ifndef ECB_MEMORY_FENCE |
|
|
239 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
240 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
241 | /* not just C11 atomics and atomic accesses */ |
|
|
242 | #include <stdatomic.h> |
|
|
243 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
244 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
245 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
246 | /* for most usages, or gcc and clang have a bug */ |
|
|
247 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
248 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
249 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
250 | /* for all __atomic_thread_fence's except seq_cst */ |
|
|
251 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
252 | #endif |
|
|
253 | #endif |
|
|
254 | |
|
|
255 | #ifndef ECB_MEMORY_FENCE |
|
|
256 | #if !ECB_AVOID_PTHREADS |
| 101 | /* |
257 | /* |
| 102 | * if you get undefined symbol references to pthread_mutex_lock, |
258 | * if you get undefined symbol references to pthread_mutex_lock, |
| 103 | * or failure to find pthread.h, then you should implement |
259 | * or failure to find pthread.h, then you should implement |
| 104 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
260 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
| 105 | * OR proide pthread.h and link against the posix thread library |
261 | * OR provide pthread.h and link against the posix thread library |
| 106 | * of your system. |
262 | * of your system. |
| 107 | */ |
263 | */ |
| 108 | #include <pthread.h> |
264 | #include <pthread.h> |
|
|
265 | #define ECB_NEEDS_PTHREADS 1 |
|
|
266 | #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
| 109 | |
267 | |
| 110 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
268 | 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) |
269 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
|
|
270 | #endif |
|
|
271 | #endif |
|
|
272 | |
|
|
273 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
| 112 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
274 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
|
|
275 | #endif |
|
|
276 | |
|
|
277 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 113 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
278 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 114 | #endif |
279 | #endif |
| 115 | |
280 | |
| 116 | /*****************************************************************************/ |
281 | /*****************************************************************************/ |
| 117 | |
282 | |
| 118 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
283 | #if ECB_CPP |
| 119 | |
|
|
| 120 | #if __cplusplus |
|
|
| 121 | #define ecb_inline static inline |
284 | #define ecb_inline static inline |
| 122 | #elif ECB_GCC_VERSION(2,5) |
285 | #elif ECB_GCC_VERSION(2,5) |
| 123 | #define ecb_inline static __inline__ |
286 | #define ecb_inline static __inline__ |
| 124 | #elif ECB_C99 |
287 | #elif ECB_C99 |
| 125 | #define ecb_inline static inline |
288 | #define ecb_inline static inline |
| … | |
… | |
| 139 | |
302 | |
| 140 | #define ECB_CONCAT_(a, b) a ## b |
303 | #define ECB_CONCAT_(a, b) a ## b |
| 141 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
304 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
| 142 | #define ECB_STRINGIFY_(a) # a |
305 | #define ECB_STRINGIFY_(a) # a |
| 143 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
306 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
307 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
| 144 | |
308 | |
| 145 | #define ecb_function_ ecb_inline |
309 | #define ecb_function_ ecb_inline |
| 146 | |
310 | |
| 147 | #if ECB_GCC_VERSION(3,1) |
311 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
| 148 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
312 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
313 | #else |
|
|
314 | #define ecb_attribute(attrlist) |
|
|
315 | #endif |
|
|
316 | |
|
|
317 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
| 149 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
318 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
319 | #else |
|
|
320 | /* possible C11 impl for integral types |
|
|
321 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
322 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
323 | |
|
|
324 | #define ecb_is_constant(expr) 0 |
|
|
325 | #endif |
|
|
326 | |
|
|
327 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
| 150 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
328 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
329 | #else |
|
|
330 | #define ecb_expect(expr,value) (expr) |
|
|
331 | #endif |
|
|
332 | |
|
|
333 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
| 151 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
334 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 152 | #else |
335 | #else |
| 153 | #define ecb_attribute(attrlist) |
|
|
| 154 | #define ecb_is_constant(expr) 0 |
|
|
| 155 | #define ecb_expect(expr,value) (expr) |
|
|
| 156 | #define ecb_prefetch(addr,rw,locality) |
336 | #define ecb_prefetch(addr,rw,locality) |
| 157 | #endif |
337 | #endif |
| 158 | |
338 | |
| 159 | /* no emulation for ecb_decltype */ |
339 | /* no emulation for ecb_decltype */ |
| 160 | #if ECB_GCC_VERSION(4,5) |
340 | #if ECB_CPP11 |
|
|
341 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
342 | template<class T> struct ecb_decltype_t { typedef T type; }; |
| 161 | #define ecb_decltype(x) __decltype(x) |
343 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
| 162 | #elif ECB_GCC_VERSION(3,0) |
344 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
| 163 | #define ecb_decltype(x) __typeof(x) |
345 | #define ecb_decltype(x) __typeof__ (x) |
| 164 | #endif |
346 | #endif |
| 165 | |
347 | |
|
|
348 | #if _MSC_VER >= 1300 |
|
|
349 | #define ecb_deprecated __declspec (deprecated) |
|
|
350 | #else |
|
|
351 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
|
|
352 | #endif |
|
|
353 | |
|
|
354 | #if __MSC_VER >= 1500 |
|
|
355 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
|
|
356 | #elif ECB_GCC_VERSION(4,5) |
|
|
357 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
|
|
358 | #else |
|
|
359 | #define ecb_deprecated_message(msg) ecb_deprecated |
|
|
360 | #endif |
|
|
361 | |
|
|
362 | #if _MSC_VER >= 1400 |
|
|
363 | #define ecb_noinline __declspec (noinline) |
|
|
364 | #else |
| 166 | #define ecb_noinline ecb_attribute ((__noinline__)) |
365 | #define ecb_noinline ecb_attribute ((__noinline__)) |
| 167 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
366 | #endif |
|
|
367 | |
| 168 | #define ecb_unused ecb_attribute ((__unused__)) |
368 | #define ecb_unused ecb_attribute ((__unused__)) |
| 169 | #define ecb_const ecb_attribute ((__const__)) |
369 | #define ecb_const ecb_attribute ((__const__)) |
| 170 | #define ecb_pure ecb_attribute ((__pure__)) |
370 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
371 | |
|
|
372 | /* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */ |
|
|
373 | #if ECB_C11 || __IBMC_NORETURN |
|
|
374 | /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */ |
|
|
375 | #define ecb_noreturn _Noreturn |
|
|
376 | #elif ECB_CPP11 |
|
|
377 | #define ecb_noreturn [[noreturn]] |
|
|
378 | #elif _MSC_VER >= 1200 |
|
|
379 | #define ecb_noreturn __declspec (noreturn) |
|
|
380 | #else |
|
|
381 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
382 | #endif |
| 171 | |
383 | |
| 172 | #if ECB_GCC_VERSION(4,3) |
384 | #if ECB_GCC_VERSION(4,3) |
| 173 | #define ecb_artificial ecb_attribute ((__artificial__)) |
385 | #define ecb_artificial ecb_attribute ((__artificial__)) |
| 174 | #define ecb_hot ecb_attribute ((__hot__)) |
386 | #define ecb_hot ecb_attribute ((__hot__)) |
| 175 | #define ecb_cold ecb_attribute ((__cold__)) |
387 | #define ecb_cold ecb_attribute ((__cold__)) |
| … | |
… | |
| 187 | /* for compatibility to the rest of the world */ |
399 | /* for compatibility to the rest of the world */ |
| 188 | #define ecb_likely(expr) ecb_expect_true (expr) |
400 | #define ecb_likely(expr) ecb_expect_true (expr) |
| 189 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
401 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
| 190 | |
402 | |
| 191 | /* count trailing zero bits and count # of one bits */ |
403 | /* count trailing zero bits and count # of one bits */ |
| 192 | #if ECB_GCC_VERSION(3,4) |
404 | #if ECB_GCC_VERSION(3,4) \ |
|
|
405 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
406 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
407 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
| 193 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
408 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
| 194 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
409 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
| 195 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
410 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
| 196 | #define ecb_ctz32(x) __builtin_ctz (x) |
411 | #define ecb_ctz32(x) __builtin_ctz (x) |
| 197 | #define ecb_ctz64(x) __builtin_ctzll (x) |
412 | #define ecb_ctz64(x) __builtin_ctzll (x) |
| 198 | #define ecb_popcount32(x) __builtin_popcount (x) |
413 | #define ecb_popcount32(x) __builtin_popcount (x) |
| 199 | /* no popcountll */ |
414 | /* no popcountll */ |
| 200 | #else |
415 | #else |
| 201 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
416 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
| 202 | ecb_function_ int |
417 | ecb_function_ ecb_const int |
| 203 | ecb_ctz32 (uint32_t x) |
418 | ecb_ctz32 (uint32_t x) |
| 204 | { |
419 | { |
| 205 | int r = 0; |
420 | int r = 0; |
| 206 | |
421 | |
| 207 | x &= ~x + 1; /* this isolates the lowest bit */ |
422 | x &= ~x + 1; /* this isolates the lowest bit */ |
| … | |
… | |
| 221 | #endif |
436 | #endif |
| 222 | |
437 | |
| 223 | return r; |
438 | return r; |
| 224 | } |
439 | } |
| 225 | |
440 | |
| 226 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
441 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
| 227 | ecb_function_ int |
442 | ecb_function_ ecb_const int |
| 228 | ecb_ctz64 (uint64_t x) |
443 | ecb_ctz64 (uint64_t x) |
| 229 | { |
444 | { |
| 230 | int shift = x & 0xffffffffU ? 0 : 32; |
445 | int shift = x & 0xffffffffU ? 0 : 32; |
| 231 | return ecb_ctz32 (x >> shift) + shift; |
446 | return ecb_ctz32 (x >> shift) + shift; |
| 232 | } |
447 | } |
| 233 | |
448 | |
| 234 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
449 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
| 235 | ecb_function_ int |
450 | ecb_function_ ecb_const int |
| 236 | ecb_popcount32 (uint32_t x) |
451 | ecb_popcount32 (uint32_t x) |
| 237 | { |
452 | { |
| 238 | x -= (x >> 1) & 0x55555555; |
453 | x -= (x >> 1) & 0x55555555; |
| 239 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
454 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
| 240 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
455 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
| 241 | x *= 0x01010101; |
456 | x *= 0x01010101; |
| 242 | |
457 | |
| 243 | return x >> 24; |
458 | return x >> 24; |
| 244 | } |
459 | } |
| 245 | |
460 | |
| 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; |
461 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
| 250 | ecb_function_ int ecb_ld32 (uint32_t x) |
462 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
| 251 | { |
463 | { |
| 252 | int r = 0; |
464 | int r = 0; |
| 253 | |
465 | |
| 254 | if (x >> 16) { x >>= 16; r += 16; } |
466 | if (x >> 16) { x >>= 16; r += 16; } |
| 255 | if (x >> 8) { x >>= 8; r += 8; } |
467 | if (x >> 8) { x >>= 8; r += 8; } |
| … | |
… | |
| 258 | if (x >> 1) { r += 1; } |
470 | if (x >> 1) { r += 1; } |
| 259 | |
471 | |
| 260 | return r; |
472 | return r; |
| 261 | } |
473 | } |
| 262 | |
474 | |
| 263 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
475 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
| 264 | ecb_function_ int ecb_ld64 (uint64_t x) |
476 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
| 265 | { |
477 | { |
| 266 | int r = 0; |
478 | int r = 0; |
| 267 | |
479 | |
| 268 | if (x >> 32) { x >>= 32; r += 32; } |
480 | if (x >> 32) { x >>= 32; r += 32; } |
| 269 | |
481 | |
| 270 | return r + ecb_ld32 (x); |
482 | return r + ecb_ld32 (x); |
| 271 | } |
483 | } |
| 272 | #endif |
484 | #endif |
|
|
485 | |
|
|
486 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
|
|
487 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
488 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
|
|
489 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
490 | |
|
|
491 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
|
|
492 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
|
|
493 | { |
|
|
494 | return ( (x * 0x0802U & 0x22110U) |
|
|
495 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
496 | } |
|
|
497 | |
|
|
498 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
|
|
499 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
|
|
500 | { |
|
|
501 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
502 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
503 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
504 | x = ( x >> 8 ) | ( x << 8); |
|
|
505 | |
|
|
506 | return x; |
|
|
507 | } |
|
|
508 | |
|
|
509 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
|
|
510 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
|
|
511 | { |
|
|
512 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
513 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
514 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
515 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
516 | x = ( x >> 16 ) | ( x << 16); |
|
|
517 | |
|
|
518 | return x; |
|
|
519 | } |
| 273 | |
520 | |
| 274 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
521 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 275 | /* so for this version we are lazy */ |
522 | /* so for this version we are lazy */ |
| 276 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
523 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
| 277 | ecb_function_ int |
524 | ecb_function_ ecb_const int |
| 278 | ecb_popcount64 (uint64_t x) |
525 | ecb_popcount64 (uint64_t x) |
| 279 | { |
526 | { |
| 280 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
527 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
| 281 | } |
528 | } |
| 282 | |
529 | |
| 283 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
530 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
| 284 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
531 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
| 285 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
532 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
| 286 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
533 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
| 287 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
534 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
| 288 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
535 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
| 289 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
536 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
| 290 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
537 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
| 291 | |
538 | |
| 292 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
539 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
| 293 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
540 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
| 294 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
541 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
| 295 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
542 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
| 296 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
543 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
| 297 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
544 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
| 298 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
545 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
| 299 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
546 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
| 300 | |
547 | |
| 301 | #if ECB_GCC_VERSION(4,3) |
548 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
| 302 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
549 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
| 303 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
550 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
| 304 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
551 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
| 305 | #else |
552 | #else |
| 306 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
553 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
| 307 | ecb_function_ uint16_t |
554 | ecb_function_ ecb_const uint16_t |
| 308 | ecb_bswap16 (uint16_t x) |
555 | ecb_bswap16 (uint16_t x) |
| 309 | { |
556 | { |
| 310 | return ecb_rotl16 (x, 8); |
557 | return ecb_rotl16 (x, 8); |
| 311 | } |
558 | } |
| 312 | |
559 | |
| 313 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
560 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
| 314 | ecb_function_ uint32_t |
561 | ecb_function_ ecb_const uint32_t |
| 315 | ecb_bswap32 (uint32_t x) |
562 | ecb_bswap32 (uint32_t x) |
| 316 | { |
563 | { |
| 317 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
564 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
| 318 | } |
565 | } |
| 319 | |
566 | |
| 320 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
567 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
| 321 | ecb_function_ uint64_t |
568 | ecb_function_ ecb_const uint64_t |
| 322 | ecb_bswap64 (uint64_t x) |
569 | ecb_bswap64 (uint64_t x) |
| 323 | { |
570 | { |
| 324 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
571 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 325 | } |
572 | } |
| 326 | #endif |
573 | #endif |
| 327 | |
574 | |
| 328 | #if ECB_GCC_VERSION(4,5) |
575 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
| 329 | #define ecb_unreachable() __builtin_unreachable () |
576 | #define ecb_unreachable() __builtin_unreachable () |
| 330 | #else |
577 | #else |
| 331 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
578 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 332 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
579 | ecb_inline ecb_noreturn void ecb_unreachable (void); |
| 333 | ecb_function_ void ecb_unreachable (void) { } |
580 | ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
| 334 | #endif |
581 | #endif |
| 335 | |
582 | |
| 336 | /* try to tell the compiler that some condition is definitely true */ |
583 | /* try to tell the compiler that some condition is definitely true */ |
| 337 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
584 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 338 | |
585 | |
| 339 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
586 | ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); |
| 340 | ecb_function_ unsigned char |
587 | ecb_inline ecb_const unsigned char |
| 341 | ecb_byteorder_helper (void) |
588 | ecb_byteorder_helper (void) |
| 342 | { |
589 | { |
| 343 | const uint32_t u = 0x11223344; |
590 | /* the union code still generates code under pressure in gcc, */ |
| 344 | return *(unsigned char *)&u; |
591 | /* but less than using pointers, and always seems to */ |
|
|
592 | /* successfully return a constant. */ |
|
|
593 | /* the reason why we have this horrible preprocessor mess */ |
|
|
594 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
595 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
596 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
597 | return 0x44; |
|
|
598 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
599 | return 0x44; |
|
|
600 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
601 | return 0x11; |
|
|
602 | #else |
|
|
603 | union |
|
|
604 | { |
|
|
605 | uint32_t i; |
|
|
606 | uint8_t c; |
|
|
607 | } u = { 0x11223344 }; |
|
|
608 | return u.c; |
|
|
609 | #endif |
| 345 | } |
610 | } |
| 346 | |
611 | |
| 347 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
612 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
| 348 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
613 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
| 349 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
614 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
| 350 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
615 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
| 351 | |
616 | |
| 352 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
617 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
| 353 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
618 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 354 | #else |
619 | #else |
| 355 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
620 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
621 | #endif |
|
|
622 | |
|
|
623 | #if ECB_CPP |
|
|
624 | template<typename T> |
|
|
625 | static inline T ecb_div_rd (T val, T div) |
|
|
626 | { |
|
|
627 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
628 | } |
|
|
629 | template<typename T> |
|
|
630 | static inline T ecb_div_ru (T val, T div) |
|
|
631 | { |
|
|
632 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
633 | } |
|
|
634 | #else |
|
|
635 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
636 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 356 | #endif |
637 | #endif |
| 357 | |
638 | |
| 358 | #if ecb_cplusplus_does_not_suck |
639 | #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) */ |
640 | /* 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> |
641 | template<typename T, int N> |
| … | |
… | |
| 364 | } |
645 | } |
| 365 | #else |
646 | #else |
| 366 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
647 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 367 | #endif |
648 | #endif |
| 368 | |
649 | |
|
|
650 | /*******************************************************************************/ |
|
|
651 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
652 | |
|
|
653 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
654 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
655 | #if 0 \ |
|
|
656 | || __i386 || __i386__ \ |
|
|
657 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
658 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
659 | || defined __s390__ || defined __s390x__ \ |
|
|
660 | || defined __mips__ \ |
|
|
661 | || defined __alpha__ \ |
|
|
662 | || defined __hppa__ \ |
|
|
663 | || defined __ia64__ \ |
|
|
664 | || defined __m68k__ \ |
|
|
665 | || defined __m88k__ \ |
|
|
666 | || defined __sh__ \ |
|
|
667 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
668 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
669 | || defined __aarch64__ |
|
|
670 | #define ECB_STDFP 1 |
|
|
671 | #include <string.h> /* for memcpy */ |
|
|
672 | #else |
|
|
673 | #define ECB_STDFP 0 |
|
|
674 | #endif |
|
|
675 | |
|
|
676 | #ifndef ECB_NO_LIBM |
|
|
677 | |
|
|
678 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
679 | |
|
|
680 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
681 | #ifdef INFINITY |
|
|
682 | #define ECB_INFINITY INFINITY |
|
|
683 | #else |
|
|
684 | #define ECB_INFINITY HUGE_VAL |
| 369 | #endif |
685 | #endif |
| 370 | |
686 | |
|
|
687 | #ifdef NAN |
|
|
688 | #define ECB_NAN NAN |
|
|
689 | #else |
|
|
690 | #define ECB_NAN ECB_INFINITY |
|
|
691 | #endif |
|
|
692 | |
|
|
693 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
|
|
694 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
695 | #else |
|
|
696 | #define ecb_ldexpf(x,e) (float) ldexp ((x), (e)) |
|
|
697 | #endif |
|
|
698 | |
|
|
699 | /* converts an ieee half/binary16 to a float */ |
|
|
700 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
701 | ecb_function_ ecb_const float |
|
|
702 | ecb_binary16_to_float (uint16_t x) |
|
|
703 | { |
|
|
704 | int e = (x >> 10) & 0x1f; |
|
|
705 | int m = x & 0x3ff; |
|
|
706 | float r; |
|
|
707 | |
|
|
708 | if (!e ) r = ecb_ldexpf (m , -24); |
|
|
709 | else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25); |
|
|
710 | else if (m ) r = ECB_NAN; |
|
|
711 | else r = ECB_INFINITY; |
|
|
712 | |
|
|
713 | return x & 0x8000 ? -r : r; |
|
|
714 | } |
|
|
715 | |
|
|
716 | /* convert a float to ieee single/binary32 */ |
|
|
717 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
|
|
718 | ecb_function_ ecb_const uint32_t |
|
|
719 | ecb_float_to_binary32 (float x) |
|
|
720 | { |
|
|
721 | uint32_t r; |
|
|
722 | |
|
|
723 | #if ECB_STDFP |
|
|
724 | memcpy (&r, &x, 4); |
|
|
725 | #else |
|
|
726 | /* slow emulation, works for anything but -0 */ |
|
|
727 | uint32_t m; |
|
|
728 | int e; |
|
|
729 | |
|
|
730 | if (x == 0e0f ) return 0x00000000U; |
|
|
731 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
732 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
733 | if (x != x ) return 0x7fbfffffU; |
|
|
734 | |
|
|
735 | m = frexpf (x, &e) * 0x1000000U; |
|
|
736 | |
|
|
737 | r = m & 0x80000000U; |
|
|
738 | |
|
|
739 | if (r) |
|
|
740 | m = -m; |
|
|
741 | |
|
|
742 | if (e <= -126) |
|
|
743 | { |
|
|
744 | m &= 0xffffffU; |
|
|
745 | m >>= (-125 - e); |
|
|
746 | e = -126; |
|
|
747 | } |
|
|
748 | |
|
|
749 | r |= (e + 126) << 23; |
|
|
750 | r |= m & 0x7fffffU; |
|
|
751 | #endif |
|
|
752 | |
|
|
753 | return r; |
|
|
754 | } |
|
|
755 | |
|
|
756 | /* converts an ieee single/binary32 to a float */ |
|
|
757 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
|
|
758 | ecb_function_ ecb_const float |
|
|
759 | ecb_binary32_to_float (uint32_t x) |
|
|
760 | { |
|
|
761 | float r; |
|
|
762 | |
|
|
763 | #if ECB_STDFP |
|
|
764 | memcpy (&r, &x, 4); |
|
|
765 | #else |
|
|
766 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
767 | int neg = x >> 31; |
|
|
768 | int e = (x >> 23) & 0xffU; |
|
|
769 | |
|
|
770 | x &= 0x7fffffU; |
|
|
771 | |
|
|
772 | if (e) |
|
|
773 | x |= 0x800000U; |
|
|
774 | else |
|
|
775 | e = 1; |
|
|
776 | |
|
|
777 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
778 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
779 | |
|
|
780 | r = neg ? -r : r; |
|
|
781 | #endif |
|
|
782 | |
|
|
783 | return r; |
|
|
784 | } |
|
|
785 | |
|
|
786 | /* convert a double to ieee double/binary64 */ |
|
|
787 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
|
|
788 | ecb_function_ ecb_const uint64_t |
|
|
789 | ecb_double_to_binary64 (double x) |
|
|
790 | { |
|
|
791 | uint64_t r; |
|
|
792 | |
|
|
793 | #if ECB_STDFP |
|
|
794 | memcpy (&r, &x, 8); |
|
|
795 | #else |
|
|
796 | /* slow emulation, works for anything but -0 */ |
|
|
797 | uint64_t m; |
|
|
798 | int e; |
|
|
799 | |
|
|
800 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
801 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
802 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
803 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
804 | |
|
|
805 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
806 | |
|
|
807 | r = m & 0x8000000000000000;; |
|
|
808 | |
|
|
809 | if (r) |
|
|
810 | m = -m; |
|
|
811 | |
|
|
812 | if (e <= -1022) |
|
|
813 | { |
|
|
814 | m &= 0x1fffffffffffffU; |
|
|
815 | m >>= (-1021 - e); |
|
|
816 | e = -1022; |
|
|
817 | } |
|
|
818 | |
|
|
819 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
820 | r |= m & 0xfffffffffffffU; |
|
|
821 | #endif |
|
|
822 | |
|
|
823 | return r; |
|
|
824 | } |
|
|
825 | |
|
|
826 | /* converts an ieee double/binary64 to a double */ |
|
|
827 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
|
|
828 | ecb_function_ ecb_const double |
|
|
829 | ecb_binary64_to_double (uint64_t x) |
|
|
830 | { |
|
|
831 | double r; |
|
|
832 | |
|
|
833 | #if ECB_STDFP |
|
|
834 | memcpy (&r, &x, 8); |
|
|
835 | #else |
|
|
836 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
837 | int neg = x >> 63; |
|
|
838 | int e = (x >> 52) & 0x7ffU; |
|
|
839 | |
|
|
840 | x &= 0xfffffffffffffU; |
|
|
841 | |
|
|
842 | if (e) |
|
|
843 | x |= 0x10000000000000U; |
|
|
844 | else |
|
|
845 | e = 1; |
|
|
846 | |
|
|
847 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
848 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
849 | |
|
|
850 | r = neg ? -r : r; |
|
|
851 | #endif |
|
|
852 | |
|
|
853 | return r; |
|
|
854 | } |
|
|
855 | |
|
|
856 | #endif |
|
|
857 | |
|
|
858 | #endif |
|
|
859 | |
