… | |
… | |
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 |
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 |
49 | #endif |
66 | #endif |
50 | |
67 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
68 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
69 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enormous grief in return for some better fake benchmark numbers. |
70 | * causing enormous grief in return for some better fake benchmark numbers. |
54 | * or so. |
71 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
72 | * 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. |
73 | * an issue with that they should have done it right in the first place. |
57 | */ |
74 | */ |
58 | #ifndef ECB_GCC_VERSION |
75 | #ifndef ECB_GCC_VERSION |
59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
76 | #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 |
77 | #define ECB_GCC_VERSION(major,minor) 0 |
61 | #else |
78 | #else |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
79 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
63 | #endif |
80 | #endif |
64 | #endif |
81 | #endif |
… | |
… | |
66 | /*****************************************************************************/ |
83 | /*****************************************************************************/ |
67 | |
84 | |
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
85 | /* 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 */ |
86 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
70 | |
87 | |
|
|
88 | #if ECB_NO_THREADS |
|
|
89 | # define ECB_NO_SMP 1 |
|
|
90 | #endif |
|
|
91 | |
71 | #if ECB_NO_THREADS || ECB_NO_SMP |
92 | #if ECB_NO_THREADS || ECB_NO_SMP |
72 | #define ECB_MEMORY_FENCE do { } while (0) |
93 | #define ECB_MEMORY_FENCE do { } while (0) |
73 | #endif |
94 | #endif |
74 | |
95 | |
75 | #ifndef ECB_MEMORY_FENCE |
96 | #ifndef ECB_MEMORY_FENCE |
76 | #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
97 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
77 | #if __i386 || __i386__ |
98 | #if __i386 || __i386__ |
78 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
99 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
79 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
100 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
80 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
101 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
81 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
102 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
103 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
83 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
104 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
84 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
105 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
85 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
106 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
107 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
87 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
108 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
88 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) |
109 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
89 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
110 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
90 | #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
111 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
91 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
112 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
92 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
113 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
93 | #elif __sparc || __sparc__ |
114 | #elif __sparc || __sparc__ |
94 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
115 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
95 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
116 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
96 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
117 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
97 | #elif defined(__s390__) || defined(__s390x__) |
118 | #elif defined __s390__ || defined __s390x__ |
98 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
119 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
120 | #elif defined __mips__ |
|
|
121 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
122 | #elif defined __alpha__ |
|
|
123 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
99 | #endif |
124 | #endif |
100 | #endif |
125 | #endif |
101 | #endif |
126 | #endif |
102 | |
127 | |
103 | #ifndef ECB_MEMORY_FENCE |
128 | #ifndef ECB_MEMORY_FENCE |
104 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__) |
129 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
105 | #define ECB_MEMORY_FENCE __sync_synchronize () |
130 | #define ECB_MEMORY_FENCE __sync_synchronize () |
106 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
131 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
107 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
132 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
108 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
133 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
109 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
134 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
110 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
135 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
111 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
136 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
112 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
137 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
113 | #elif defined(_WIN32) |
138 | #elif defined _WIN32 |
114 | #include <WinNT.h> |
139 | #include <WinNT.h> |
115 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
140 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
116 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
141 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
117 | #include <mbarrier.h> |
142 | #include <mbarrier.h> |
118 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
143 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
119 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
144 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
120 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
145 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
146 | #elif __xlC__ |
|
|
147 | #define ECB_MEMORY_FENCE __sync () |
121 | #endif |
148 | #endif |
122 | #endif |
149 | #endif |
123 | |
150 | |
124 | #ifndef ECB_MEMORY_FENCE |
151 | #ifndef ECB_MEMORY_FENCE |
125 | #if !ECB_AVOID_PTHREADS |
152 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
137 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
164 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
138 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
165 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
139 | #endif |
166 | #endif |
140 | #endif |
167 | #endif |
141 | |
168 | |
142 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
169 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
143 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
170 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
144 | #endif |
171 | #endif |
145 | |
172 | |
146 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
173 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
147 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
174 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
148 | #endif |
175 | #endif |
149 | |
176 | |
150 | /*****************************************************************************/ |
177 | /*****************************************************************************/ |
151 | |
178 | |
… | |
… | |
300 | |
327 | |
301 | return r + ecb_ld32 (x); |
328 | return r + ecb_ld32 (x); |
302 | } |
329 | } |
303 | #endif |
330 | #endif |
304 | |
331 | |
|
|
332 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
333 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
334 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
335 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
336 | |
305 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
337 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
306 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
338 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
307 | { |
339 | { |
308 | return ( (x * 0x0802U & 0x22110U) |
340 | return ( (x * 0x0802U & 0x22110U) |
309 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
341 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |