… | |
… | |
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 0x00010001 |
|
|
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 |
|
|
59 | typedef intptr_t ptrdiff_t; |
47 | #else |
60 | #else |
48 | #include <inttypes.h> |
61 | #include <inttypes.h> |
|
|
62 | #if UINTMAX_MAX > 0xffffffffU |
|
|
63 | #define ECB_PTRSIZE 8 |
|
|
64 | #else |
|
|
65 | #define ECB_PTRSIZE 4 |
|
|
66 | #endif |
49 | #endif |
67 | #endif |
50 | |
68 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
69 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
70 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enormous grief in return for some better fake benchmark numbers. |
71 | * causing enormous grief in return for some better fake benchmark numbers. |
54 | * or so. |
72 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
73 | * 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. |
74 | * an issue with that they should have done it right in the first place. |
57 | */ |
75 | */ |
58 | #ifndef ECB_GCC_VERSION |
76 | #ifndef ECB_GCC_VERSION |
59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
77 | #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 |
78 | #define ECB_GCC_VERSION(major,minor) 0 |
61 | #else |
79 | #else |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
80 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
63 | #endif |
81 | #endif |
64 | #endif |
82 | #endif |
65 | |
83 | |
|
|
84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
|
|
85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
|
|
87 | #define ECB_CPP (__cplusplus+0) |
|
|
88 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
89 | |
66 | /*****************************************************************************/ |
90 | /*****************************************************************************/ |
67 | |
91 | |
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
92 | /* 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 */ |
93 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
70 | |
94 | |
71 | #if ECB_NO_THREADS |
95 | #if ECB_NO_THREADS |
72 | # define ECB_NO_SMP 1 |
96 | #define ECB_NO_SMP 1 |
73 | #endif |
97 | #endif |
74 | |
98 | |
75 | #if ECB_NO_THREADS || ECB_NO_SMP |
99 | #if ECB_NO_SMP |
76 | #define ECB_MEMORY_FENCE do { } while (0) |
100 | #define ECB_MEMORY_FENCE do { } while (0) |
77 | #endif |
101 | #endif |
78 | |
102 | |
79 | #ifndef ECB_MEMORY_FENCE |
103 | #ifndef ECB_MEMORY_FENCE |
80 | #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
104 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
81 | #if __i386 || __i386__ |
105 | #if __i386 || __i386__ |
82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
106 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
83 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
107 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
84 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
108 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
85 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
109 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
110 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
87 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
111 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
88 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
112 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
89 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
113 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
90 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
114 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
91 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
115 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
92 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) |
116 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
93 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
117 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
94 | #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
118 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
95 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
119 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
96 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
120 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
97 | #elif __sparc || __sparc__ |
121 | #elif __sparc || __sparc__ |
98 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
122 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
99 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
123 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
100 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
124 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
101 | #elif defined(__s390__) || defined(__s390x__) |
125 | #elif defined __s390__ || defined __s390x__ |
102 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
103 | #elif defined(__mips__) |
127 | #elif defined __mips__ |
104 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
128 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
105 | #elif __xlC__ |
129 | #elif defined __alpha__ |
106 | #define ECB_MEMORY_FENCE __lwsync () |
130 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
131 | #elif defined __hppa__ |
|
|
132 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
134 | #elif defined __ia64__ |
|
|
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
107 | #endif |
136 | #endif |
108 | #endif |
137 | #endif |
109 | #endif |
138 | #endif |
110 | |
139 | |
111 | #ifndef ECB_MEMORY_FENCE |
140 | #ifndef ECB_MEMORY_FENCE |
|
|
141 | #if ECB_GCC_VERSION(4,7) |
|
|
142 | /* see comment below about the C11 memory model. in short - avoid */ |
|
|
143 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
144 | #elif defined __clang && __has_feature (cxx_atomic) |
|
|
145 | /* see above */ |
|
|
146 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
112 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__) |
147 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
113 | #define ECB_MEMORY_FENCE __sync_synchronize () |
148 | #define ECB_MEMORY_FENCE __sync_synchronize () |
114 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
|
|
115 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
|
|
116 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
149 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
117 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
150 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
118 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
151 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
119 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
152 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
120 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
153 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
121 | #elif defined(_WIN32) |
154 | #elif defined _WIN32 |
122 | #include <WinNT.h> |
155 | #include <WinNT.h> |
123 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
156 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
124 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
157 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
125 | #include <mbarrier.h> |
158 | #include <mbarrier.h> |
126 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
159 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
127 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
160 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
128 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
161 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
162 | #elif __xlC__ |
|
|
163 | #define ECB_MEMORY_FENCE __sync () |
|
|
164 | #endif |
|
|
165 | #endif |
|
|
166 | |
|
|
167 | #ifndef ECB_MEMORY_FENCE |
|
|
168 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
169 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
170 | /* not just C11 atomics and atomic accesses */ |
|
|
171 | #include <stdatomic.h> |
|
|
172 | /* unfortunately, the C11 memory model seems to be very limited, and unable to express */ |
|
|
173 | /* simple barrier semantics. That means we need to take out thor's hammer. */ |
|
|
174 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
129 | #endif |
175 | #endif |
130 | #endif |
176 | #endif |
131 | |
177 | |
132 | #ifndef ECB_MEMORY_FENCE |
178 | #ifndef ECB_MEMORY_FENCE |
133 | #if !ECB_AVOID_PTHREADS |
179 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
145 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
191 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
146 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
192 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
147 | #endif |
193 | #endif |
148 | #endif |
194 | #endif |
149 | |
195 | |
150 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
196 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
151 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
197 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
152 | #endif |
198 | #endif |
153 | |
199 | |
154 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
200 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
155 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
201 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
156 | #endif |
202 | #endif |
157 | |
203 | |
158 | /*****************************************************************************/ |
204 | /*****************************************************************************/ |
159 | |
|
|
160 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
161 | |
205 | |
162 | #if __cplusplus |
206 | #if __cplusplus |
163 | #define ecb_inline static inline |
207 | #define ecb_inline static inline |
164 | #elif ECB_GCC_VERSION(2,5) |
208 | #elif ECB_GCC_VERSION(2,5) |
165 | #define ecb_inline static __inline__ |
209 | #define ecb_inline static __inline__ |
… | |
… | |
204 | #elif ECB_GCC_VERSION(3,0) |
248 | #elif ECB_GCC_VERSION(3,0) |
205 | #define ecb_decltype(x) __typeof(x) |
249 | #define ecb_decltype(x) __typeof(x) |
206 | #endif |
250 | #endif |
207 | |
251 | |
208 | #define ecb_noinline ecb_attribute ((__noinline__)) |
252 | #define ecb_noinline ecb_attribute ((__noinline__)) |
209 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
210 | #define ecb_unused ecb_attribute ((__unused__)) |
253 | #define ecb_unused ecb_attribute ((__unused__)) |
211 | #define ecb_const ecb_attribute ((__const__)) |
254 | #define ecb_const ecb_attribute ((__const__)) |
212 | #define ecb_pure ecb_attribute ((__pure__)) |
255 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
256 | |
|
|
257 | #if ECB_C11 |
|
|
258 | #define ecb_noreturn _Noreturn |
|
|
259 | #else |
|
|
260 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
261 | #endif |
213 | |
262 | |
214 | #if ECB_GCC_VERSION(4,3) |
263 | #if ECB_GCC_VERSION(4,3) |
215 | #define ecb_artificial ecb_attribute ((__artificial__)) |
264 | #define ecb_artificial ecb_attribute ((__artificial__)) |
216 | #define ecb_hot ecb_attribute ((__hot__)) |
265 | #define ecb_hot ecb_attribute ((__hot__)) |
217 | #define ecb_cold ecb_attribute ((__cold__)) |
266 | #define ecb_cold ecb_attribute ((__cold__)) |
… | |
… | |
308 | |
357 | |
309 | return r + ecb_ld32 (x); |
358 | return r + ecb_ld32 (x); |
310 | } |
359 | } |
311 | #endif |
360 | #endif |
312 | |
361 | |
|
|
362 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
363 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
364 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
365 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
366 | |
313 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
367 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
314 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
368 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
315 | { |
369 | { |
316 | return ( (x * 0x0802U & 0x22110U) |
370 | return ( (x * 0x0802U & 0x22110U) |
317 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
371 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |