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: |
… | |
… | |
28 | */ |
28 | */ |
29 | |
29 | |
30 | #ifndef ECB_H |
30 | #ifndef ECB_H |
31 | #define ECB_H |
31 | #define ECB_H |
32 | |
32 | |
|
|
33 | /* 16 bits major, 16 bits minor */ |
|
|
34 | #define ECB_VERSION 0x00010003 |
|
|
35 | |
33 | #ifdef _WIN32 |
36 | #ifdef _WIN32 |
34 | typedef signed char int8_t; |
37 | typedef signed char int8_t; |
35 | typedef unsigned char uint8_t; |
38 | typedef unsigned char uint8_t; |
36 | typedef signed short int16_t; |
39 | typedef signed short int16_t; |
37 | typedef unsigned short uint16_t; |
40 | typedef unsigned short uint16_t; |
… | |
… | |
42 | typedef unsigned long long uint64_t; |
45 | typedef unsigned long long uint64_t; |
43 | #else /* _MSC_VER || __BORLANDC__ */ |
46 | #else /* _MSC_VER || __BORLANDC__ */ |
44 | typedef signed __int64 int64_t; |
47 | typedef signed __int64 int64_t; |
45 | typedef unsigned __int64 uint64_t; |
48 | typedef unsigned __int64 uint64_t; |
46 | #endif |
49 | #endif |
|
|
50 | #ifdef _WIN64 |
|
|
51 | #define ECB_PTRSIZE 8 |
|
|
52 | typedef uint64_t uintptr_t; |
|
|
53 | typedef int64_t intptr_t; |
|
|
54 | #else |
|
|
55 | #define ECB_PTRSIZE 4 |
|
|
56 | typedef uint32_t uintptr_t; |
|
|
57 | typedef int32_t intptr_t; |
|
|
58 | #endif |
47 | #else |
59 | #else |
48 | #include <inttypes.h> |
60 | #include <inttypes.h> |
|
|
61 | #if UINTMAX_MAX > 0xffffffffU |
|
|
62 | #define ECB_PTRSIZE 8 |
|
|
63 | #else |
|
|
64 | #define ECB_PTRSIZE 4 |
|
|
65 | #endif |
|
|
66 | #endif |
|
|
67 | |
|
|
68 | /* work around x32 idiocy by defining proper macros */ |
|
|
69 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
|
|
70 | #if _ILP32 |
|
|
71 | #define ECB_AMD64_X32 1 |
|
|
72 | #else |
|
|
73 | #define ECB_AMD64 1 |
|
|
74 | #endif |
49 | #endif |
75 | #endif |
50 | |
76 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
77 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
78 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enormous grief in return for some better fake benchmark numbers. |
79 | * causing enormous grief in return for some better fake benchmark numbers. |
54 | * or so. |
80 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
81 | * we try to detect these and simply assume they are not gcc - if they have |
56 | * an issue with that they should have done it right in the first place. |
82 | * an issue with that they should have done it right in the first place. |
57 | */ |
83 | */ |
58 | #ifndef ECB_GCC_VERSION |
84 | #ifndef ECB_GCC_VERSION |
59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
85 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
60 | #define ECB_GCC_VERSION(major,minor) 0 |
86 | #define ECB_GCC_VERSION(major,minor) 0 |
61 | #else |
87 | #else |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
88 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
63 | #endif |
89 | #endif |
64 | #endif |
90 | #endif |
65 | |
91 | |
|
|
92 | #define ECB_CPP (__cplusplus+0) |
|
|
93 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
94 | |
|
|
95 | #if ECB_CPP |
|
|
96 | #define ECB_C 0 |
|
|
97 | #define ECB_STDC_VERSION 0 |
|
|
98 | #else |
|
|
99 | #define ECB_C 1 |
|
|
100 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
101 | #endif |
|
|
102 | |
|
|
103 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
104 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
105 | |
|
|
106 | #if ECB_CPP |
|
|
107 | #define ECB_EXTERN_C extern "C" |
|
|
108 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
|
|
109 | #define ECB_EXTERN_C_END } |
|
|
110 | #else |
|
|
111 | #define ECB_EXTERN_C extern |
|
|
112 | #define ECB_EXTERN_C_BEG |
|
|
113 | #define ECB_EXTERN_C_END |
|
|
114 | #endif |
|
|
115 | |
66 | /*****************************************************************************/ |
116 | /*****************************************************************************/ |
67 | |
117 | |
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
118 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
69 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
119 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
70 | |
120 | |
71 | #if ECB_NO_THREADS || ECB_NO_SMP |
121 | #if ECB_NO_THREADS |
|
|
122 | #define ECB_NO_SMP 1 |
|
|
123 | #endif |
|
|
124 | |
|
|
125 | #if ECB_NO_SMP |
72 | #define ECB_MEMORY_FENCE do { } while (0) |
126 | #define ECB_MEMORY_FENCE do { } while (0) |
73 | #define ECB_MEMORY_FENCE_ACQUIRE do { } while (0) |
|
|
74 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) |
|
|
75 | #endif |
127 | #endif |
76 | |
128 | |
77 | #ifndef ECB_MEMORY_FENCE |
129 | #ifndef ECB_MEMORY_FENCE |
78 | #if ECB_GCC_VERSION(2,5) |
130 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
79 | #if __x86 |
131 | #if __i386 || __i386__ |
80 | #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") |
81 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
133 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
82 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
134 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
83 | #elif __amd64 |
135 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
84 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
136 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
85 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
137 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
86 | #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") |
87 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
141 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
88 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) \ |
142 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
|
|
143 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
89 | || defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
144 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
90 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
145 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
91 | #define ECB_MEMORY_FENCE \ |
146 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
92 | do { \ |
147 | #elif __aarch64__ |
93 | int null = 0; \ |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
94 | __asm__ __volatile__ ("mcr p15,0,%0,c6,c10,5", : "=&r" (null) : : "memory"); \ |
149 | #elif (__sparc || __sparc__) && !__sparcv8 |
95 | while (0) |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
|
|
151 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
152 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
153 | #elif defined __s390__ || defined __s390x__ |
|
|
154 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
155 | #elif defined __mips__ |
|
|
156 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
157 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
|
|
158 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
|
|
159 | #elif defined __alpha__ |
|
|
160 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
161 | #elif defined __hppa__ |
|
|
162 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
163 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
164 | #elif defined __ia64__ |
|
|
165 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
166 | #elif defined __m68k__ |
|
|
167 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
168 | #elif defined __m88k__ |
|
|
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
170 | #elif defined __sh__ |
|
|
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
96 | #endif |
172 | #endif |
97 | #endif |
173 | #endif |
98 | #endif |
174 | #endif |
99 | |
175 | |
100 | #ifndef ECB_MEMORY_FENCE |
176 | #ifndef ECB_MEMORY_FENCE |
|
|
177 | #if ECB_GCC_VERSION(4,7) |
|
|
178 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
179 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
180 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
181 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
182 | |
|
|
183 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
|
|
184 | * without risking compile time errors with other compilers. We *could* |
|
|
185 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
|
|
186 | * around this shit time and again. |
|
|
187 | * #elif defined __clang && __has_feature (cxx_atomic) |
|
|
188 | * // see comment below (stdatomic.h) about the C11 memory model. |
|
|
189 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
190 | * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
191 | * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
192 | */ |
|
|
193 | |
101 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) |
194 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
102 | #define ECB_MEMORY_FENCE __sync_synchronize () |
195 | #define ECB_MEMORY_FENCE __sync_synchronize () |
103 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
196 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
104 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
197 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
198 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
199 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
200 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
201 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
105 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
202 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
106 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
203 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
107 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
204 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
108 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
205 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
109 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
206 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
110 | #elif defined(_WIN32) |
207 | #elif defined _WIN32 |
111 | #include <WinNT.h> |
208 | #include <WinNT.h> |
112 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
209 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
210 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
211 | #include <mbarrier.h> |
|
|
212 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
213 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
214 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
215 | #elif __xlC__ |
|
|
216 | #define ECB_MEMORY_FENCE __sync () |
|
|
217 | #endif |
|
|
218 | #endif |
|
|
219 | |
|
|
220 | #ifndef ECB_MEMORY_FENCE |
|
|
221 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
222 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
223 | /* not just C11 atomics and atomic accesses */ |
|
|
224 | #include <stdatomic.h> |
|
|
225 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
226 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
227 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
228 | /* for most usages, or gcc and clang have a bug */ |
|
|
229 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
230 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
231 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
232 | /* for all __atomic_thread_fence's except seq_cst */ |
|
|
233 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
113 | #endif |
234 | #endif |
114 | #endif |
235 | #endif |
115 | |
236 | |
116 | #ifndef ECB_MEMORY_FENCE |
237 | #ifndef ECB_MEMORY_FENCE |
117 | #if !ECB_AVOID_PTHREADS |
238 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
129 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
250 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
130 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
251 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
131 | #endif |
252 | #endif |
132 | #endif |
253 | #endif |
133 | |
254 | |
134 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
255 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
135 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
256 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
136 | #endif |
257 | #endif |
137 | |
258 | |
138 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
259 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
139 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
260 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
140 | #endif |
261 | #endif |
141 | |
262 | |
142 | /*****************************************************************************/ |
263 | /*****************************************************************************/ |
143 | |
|
|
144 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
145 | |
264 | |
146 | #if __cplusplus |
265 | #if __cplusplus |
147 | #define ecb_inline static inline |
266 | #define ecb_inline static inline |
148 | #elif ECB_GCC_VERSION(2,5) |
267 | #elif ECB_GCC_VERSION(2,5) |
149 | #define ecb_inline static __inline__ |
268 | #define ecb_inline static __inline__ |
… | |
… | |
175 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
294 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
176 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
295 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
177 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
296 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
178 | #else |
297 | #else |
179 | #define ecb_attribute(attrlist) |
298 | #define ecb_attribute(attrlist) |
|
|
299 | |
|
|
300 | /* possible C11 impl for integral types |
|
|
301 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
302 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
303 | |
180 | #define ecb_is_constant(expr) 0 |
304 | #define ecb_is_constant(expr) 0 |
181 | #define ecb_expect(expr,value) (expr) |
305 | #define ecb_expect(expr,value) (expr) |
182 | #define ecb_prefetch(addr,rw,locality) |
306 | #define ecb_prefetch(addr,rw,locality) |
183 | #endif |
307 | #endif |
184 | |
308 | |
… | |
… | |
188 | #elif ECB_GCC_VERSION(3,0) |
312 | #elif ECB_GCC_VERSION(3,0) |
189 | #define ecb_decltype(x) __typeof(x) |
313 | #define ecb_decltype(x) __typeof(x) |
190 | #endif |
314 | #endif |
191 | |
315 | |
192 | #define ecb_noinline ecb_attribute ((__noinline__)) |
316 | #define ecb_noinline ecb_attribute ((__noinline__)) |
193 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
194 | #define ecb_unused ecb_attribute ((__unused__)) |
317 | #define ecb_unused ecb_attribute ((__unused__)) |
195 | #define ecb_const ecb_attribute ((__const__)) |
318 | #define ecb_const ecb_attribute ((__const__)) |
196 | #define ecb_pure ecb_attribute ((__pure__)) |
319 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
320 | |
|
|
321 | #if ECB_C11 |
|
|
322 | #define ecb_noreturn _Noreturn |
|
|
323 | #else |
|
|
324 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
325 | #endif |
197 | |
326 | |
198 | #if ECB_GCC_VERSION(4,3) |
327 | #if ECB_GCC_VERSION(4,3) |
199 | #define ecb_artificial ecb_attribute ((__artificial__)) |
328 | #define ecb_artificial ecb_attribute ((__artificial__)) |
200 | #define ecb_hot ecb_attribute ((__hot__)) |
329 | #define ecb_hot ecb_attribute ((__hot__)) |
201 | #define ecb_cold ecb_attribute ((__cold__)) |
330 | #define ecb_cold ecb_attribute ((__cold__)) |
… | |
… | |
292 | |
421 | |
293 | return r + ecb_ld32 (x); |
422 | return r + ecb_ld32 (x); |
294 | } |
423 | } |
295 | #endif |
424 | #endif |
296 | |
425 | |
|
|
426 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
427 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
428 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
429 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
430 | |
|
|
431 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
432 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
433 | { |
|
|
434 | return ( (x * 0x0802U & 0x22110U) |
|
|
435 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
436 | } |
|
|
437 | |
|
|
438 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
439 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
440 | { |
|
|
441 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
442 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
443 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
444 | x = ( x >> 8 ) | ( x << 8); |
|
|
445 | |
|
|
446 | return x; |
|
|
447 | } |
|
|
448 | |
|
|
449 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
450 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
451 | { |
|
|
452 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
453 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
454 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
455 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
456 | x = ( x >> 16 ) | ( x << 16); |
|
|
457 | |
|
|
458 | return x; |
|
|
459 | } |
|
|
460 | |
297 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
461 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
298 | /* so for this version we are lazy */ |
462 | /* so for this version we are lazy */ |
299 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
463 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
300 | ecb_function_ int |
464 | ecb_function_ int |
301 | ecb_popcount64 (uint64_t x) |
465 | ecb_popcount64 (uint64_t x) |
… | |
… | |
350 | |
514 | |
351 | #if ECB_GCC_VERSION(4,5) |
515 | #if ECB_GCC_VERSION(4,5) |
352 | #define ecb_unreachable() __builtin_unreachable () |
516 | #define ecb_unreachable() __builtin_unreachable () |
353 | #else |
517 | #else |
354 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
518 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
355 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
519 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
356 | ecb_function_ void ecb_unreachable (void) { } |
520 | ecb_inline void ecb_unreachable (void) { } |
357 | #endif |
521 | #endif |
358 | |
522 | |
359 | /* try to tell the compiler that some condition is definitely true */ |
523 | /* try to tell the compiler that some condition is definitely true */ |
360 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
524 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
361 | |
525 | |
362 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
526 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
363 | ecb_function_ unsigned char |
527 | ecb_inline unsigned char |
364 | ecb_byteorder_helper (void) |
528 | ecb_byteorder_helper (void) |
365 | { |
529 | { |
366 | const uint32_t u = 0x11223344; |
530 | /* the union code still generates code under pressure in gcc, */ |
367 | return *(unsigned char *)&u; |
531 | /* but less than using pointers, and always seems to */ |
|
|
532 | /* successfully return a constant. */ |
|
|
533 | /* the reason why we have this horrible preprocessor mess */ |
|
|
534 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
535 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
536 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
537 | return 0x44; |
|
|
538 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
539 | return 0x44; |
|
|
540 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
541 | return 0x11; |
|
|
542 | #else |
|
|
543 | union |
|
|
544 | { |
|
|
545 | uint32_t i; |
|
|
546 | uint8_t c; |
|
|
547 | } u = { 0x11223344 }; |
|
|
548 | return u.c; |
|
|
549 | #endif |
368 | } |
550 | } |
369 | |
551 | |
370 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
552 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
371 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
553 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
372 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
554 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
373 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
555 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
374 | |
556 | |
375 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
557 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
376 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
558 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
377 | #else |
559 | #else |
378 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
560 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
561 | #endif |
|
|
562 | |
|
|
563 | #if __cplusplus |
|
|
564 | template<typename T> |
|
|
565 | static inline T ecb_div_rd (T val, T div) |
|
|
566 | { |
|
|
567 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
568 | } |
|
|
569 | template<typename T> |
|
|
570 | static inline T ecb_div_ru (T val, T div) |
|
|
571 | { |
|
|
572 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
573 | } |
|
|
574 | #else |
|
|
575 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
576 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
379 | #endif |
577 | #endif |
380 | |
578 | |
381 | #if ecb_cplusplus_does_not_suck |
579 | #if ecb_cplusplus_does_not_suck |
382 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
580 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
383 | template<typename T, int N> |
581 | template<typename T, int N> |
… | |
… | |
387 | } |
585 | } |
388 | #else |
586 | #else |
389 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
587 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
390 | #endif |
588 | #endif |
391 | |
589 | |
|
|
590 | /*******************************************************************************/ |
|
|
591 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
592 | |
|
|
593 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
594 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
595 | #if 0 \ |
|
|
596 | || __i386 || __i386__ \ |
|
|
597 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
598 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
599 | || defined __s390__ || defined __s390x__ \ |
|
|
600 | || defined __mips__ \ |
|
|
601 | || defined __alpha__ \ |
|
|
602 | || defined __hppa__ \ |
|
|
603 | || defined __ia64__ \ |
|
|
604 | || defined __m68k__ \ |
|
|
605 | || defined __m88k__ \ |
|
|
606 | || defined __sh__ \ |
|
|
607 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
608 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
609 | | defined __aarch64__ |
|
|
610 | #define ECB_STDFP 1 |
|
|
611 | #include <string.h> /* for memcpy */ |
|
|
612 | #else |
|
|
613 | #define ECB_STDFP 0 |
|
|
614 | #endif |
|
|
615 | |
|
|
616 | #ifndef ECB_NO_LIBM |
|
|
617 | |
|
|
618 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
619 | |
|
|
620 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
621 | #ifdef INFINITY |
|
|
622 | #define ECB_INFINITY INFINITY |
|
|
623 | #else |
|
|
624 | #define ECB_INFINITY HUGE_VAL |
392 | #endif |
625 | #endif |
393 | |
626 | |
|
|
627 | #ifdef NAN |
|
|
628 | #define ECB_NAN NAN |
|
|
629 | #else |
|
|
630 | #define ECB_NAN ECB_INFINITY |
|
|
631 | #endif |
|
|
632 | |
|
|
633 | /* converts an ieee half/binary16 to a float */ |
|
|
634 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
635 | ecb_function_ float |
|
|
636 | ecb_binary16_to_float (uint16_t x) |
|
|
637 | { |
|
|
638 | int e = (x >> 10) & 0x1f; |
|
|
639 | int m = x & 0x3ff; |
|
|
640 | float r; |
|
|
641 | |
|
|
642 | if (!e ) r = ldexpf (m , -24); |
|
|
643 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
644 | else if (m ) r = ECB_NAN; |
|
|
645 | else r = ECB_INFINITY; |
|
|
646 | |
|
|
647 | return x & 0x8000 ? -r : r; |
|
|
648 | } |
|
|
649 | |
|
|
650 | /* convert a float to ieee single/binary32 */ |
|
|
651 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
652 | ecb_function_ uint32_t |
|
|
653 | ecb_float_to_binary32 (float x) |
|
|
654 | { |
|
|
655 | uint32_t r; |
|
|
656 | |
|
|
657 | #if ECB_STDFP |
|
|
658 | memcpy (&r, &x, 4); |
|
|
659 | #else |
|
|
660 | /* slow emulation, works for anything but -0 */ |
|
|
661 | uint32_t m; |
|
|
662 | int e; |
|
|
663 | |
|
|
664 | if (x == 0e0f ) return 0x00000000U; |
|
|
665 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
666 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
667 | if (x != x ) return 0x7fbfffffU; |
|
|
668 | |
|
|
669 | m = frexpf (x, &e) * 0x1000000U; |
|
|
670 | |
|
|
671 | r = m & 0x80000000U; |
|
|
672 | |
|
|
673 | if (r) |
|
|
674 | m = -m; |
|
|
675 | |
|
|
676 | if (e <= -126) |
|
|
677 | { |
|
|
678 | m &= 0xffffffU; |
|
|
679 | m >>= (-125 - e); |
|
|
680 | e = -126; |
|
|
681 | } |
|
|
682 | |
|
|
683 | r |= (e + 126) << 23; |
|
|
684 | r |= m & 0x7fffffU; |
|
|
685 | #endif |
|
|
686 | |
|
|
687 | return r; |
|
|
688 | } |
|
|
689 | |
|
|
690 | /* converts an ieee single/binary32 to a float */ |
|
|
691 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
692 | ecb_function_ float |
|
|
693 | ecb_binary32_to_float (uint32_t x) |
|
|
694 | { |
|
|
695 | float r; |
|
|
696 | |
|
|
697 | #if ECB_STDFP |
|
|
698 | memcpy (&r, &x, 4); |
|
|
699 | #else |
|
|
700 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
701 | int neg = x >> 31; |
|
|
702 | int e = (x >> 23) & 0xffU; |
|
|
703 | |
|
|
704 | x &= 0x7fffffU; |
|
|
705 | |
|
|
706 | if (e) |
|
|
707 | x |= 0x800000U; |
|
|
708 | else |
|
|
709 | e = 1; |
|
|
710 | |
|
|
711 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
712 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
713 | |
|
|
714 | r = neg ? -r : r; |
|
|
715 | #endif |
|
|
716 | |
|
|
717 | return r; |
|
|
718 | } |
|
|
719 | |
|
|
720 | /* convert a double to ieee double/binary64 */ |
|
|
721 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
722 | ecb_function_ uint64_t |
|
|
723 | ecb_double_to_binary64 (double x) |
|
|
724 | { |
|
|
725 | uint64_t r; |
|
|
726 | |
|
|
727 | #if ECB_STDFP |
|
|
728 | memcpy (&r, &x, 8); |
|
|
729 | #else |
|
|
730 | /* slow emulation, works for anything but -0 */ |
|
|
731 | uint64_t m; |
|
|
732 | int e; |
|
|
733 | |
|
|
734 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
735 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
736 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
737 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
738 | |
|
|
739 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
740 | |
|
|
741 | r = m & 0x8000000000000000;; |
|
|
742 | |
|
|
743 | if (r) |
|
|
744 | m = -m; |
|
|
745 | |
|
|
746 | if (e <= -1022) |
|
|
747 | { |
|
|
748 | m &= 0x1fffffffffffffU; |
|
|
749 | m >>= (-1021 - e); |
|
|
750 | e = -1022; |
|
|
751 | } |
|
|
752 | |
|
|
753 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
754 | r |= m & 0xfffffffffffffU; |
|
|
755 | #endif |
|
|
756 | |
|
|
757 | return r; |
|
|
758 | } |
|
|
759 | |
|
|
760 | /* converts an ieee double/binary64 to a double */ |
|
|
761 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
762 | ecb_function_ double |
|
|
763 | ecb_binary64_to_double (uint64_t x) |
|
|
764 | { |
|
|
765 | double r; |
|
|
766 | |
|
|
767 | #if ECB_STDFP |
|
|
768 | memcpy (&r, &x, 8); |
|
|
769 | #else |
|
|
770 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
771 | int neg = x >> 63; |
|
|
772 | int e = (x >> 52) & 0x7ffU; |
|
|
773 | |
|
|
774 | x &= 0xfffffffffffffU; |
|
|
775 | |
|
|
776 | if (e) |
|
|
777 | x |= 0x10000000000000U; |
|
|
778 | else |
|
|
779 | e = 1; |
|
|
780 | |
|
|
781 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
782 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
783 | |
|
|
784 | r = neg ? -r : r; |
|
|
785 | #endif |
|
|
786 | |
|
|
787 | return r; |
|
|
788 | } |
|
|
789 | |
|
|
790 | #endif |
|
|
791 | |
|
|
792 | #endif |
|
|
793 | |