… | |
… | |
48 | #include <inttypes.h> |
48 | #include <inttypes.h> |
49 | #endif |
49 | #endif |
50 | |
50 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
52 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enourmous grief in return for some better fake benchmark numbers. |
53 | * causing enormous grief in return for some better fake benchmark numbers. |
54 | * or so. |
54 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
55 | * 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. |
56 | * an issue with that they should have done it right in the first place. |
57 | */ |
57 | */ |
58 | #ifndef ECB_GCC_VERSION |
58 | #ifndef ECB_GCC_VERSION |
… | |
… | |
61 | #else |
61 | #else |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
63 | #endif |
63 | #endif |
64 | #endif |
64 | #endif |
65 | |
65 | |
|
|
66 | /*****************************************************************************/ |
|
|
67 | |
|
|
68 | /* 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 */ |
|
|
70 | |
|
|
71 | #if ECB_NO_THREADS || ECB_NO_SMP |
|
|
72 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
73 | #endif |
|
|
74 | |
|
|
75 | #ifndef ECB_MEMORY_FENCE |
|
|
76 | #if ECB_GCC_VERSION(2,5) |
|
|
77 | #if __x86 |
|
|
78 | #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 */ |
|
|
80 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
|
|
81 | #elif __amd64 |
|
|
82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
|
|
83 | #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 */ |
|
|
85 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
|
|
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
87 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
|
|
88 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) |
|
|
89 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,4" : : "r" (0) : "memory") |
|
|
90 | #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
|
|
91 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
|
|
92 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dsb" : : : "memory") |
|
|
93 | #endif |
|
|
94 | #endif |
|
|
95 | #endif |
|
|
96 | |
|
|
97 | #ifndef ECB_MEMORY_FENCE |
|
|
98 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) |
|
|
99 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
100 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
|
|
101 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
|
|
102 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
|
|
103 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
104 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
|
|
105 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
106 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
|
|
107 | #elif defined(_WIN32) |
|
|
108 | #include <WinNT.h> |
|
|
109 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
110 | #endif |
|
|
111 | #endif |
|
|
112 | |
|
|
113 | #ifndef ECB_MEMORY_FENCE |
|
|
114 | #if !ECB_AVOID_PTHREADS |
|
|
115 | /* |
|
|
116 | * if you get undefined symbol references to pthread_mutex_lock, |
|
|
117 | * or failure to find pthread.h, then you should implement |
|
|
118 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
|
|
119 | * OR provide pthread.h and link against the posix thread library |
|
|
120 | * of your system. |
|
|
121 | */ |
|
|
122 | #include <pthread.h> |
|
|
123 | #define ECB_NEEDS_PTHREADS 1 |
|
|
124 | #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
|
|
125 | |
|
|
126 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
|
|
127 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
|
|
128 | #endif |
|
|
129 | #endif |
|
|
130 | |
|
|
131 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
|
|
132 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
|
|
133 | #endif |
|
|
134 | |
|
|
135 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
|
|
136 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
|
|
137 | #endif |
|
|
138 | |
|
|
139 | /*****************************************************************************/ |
|
|
140 | |
66 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
141 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
67 | |
142 | |
68 | #if __cplusplus |
143 | #if __cplusplus |
69 | #define ecb_inline static inline |
144 | #define ecb_inline static inline |
70 | #elif ECB_GCC_VERSION(2,5) |
145 | #elif ECB_GCC_VERSION(2,5) |
… | |
… | |
189 | x *= 0x01010101; |
264 | x *= 0x01010101; |
190 | |
265 | |
191 | return x >> 24; |
266 | return x >> 24; |
192 | } |
267 | } |
193 | |
268 | |
194 | /* you have the choice beetween something with a table lookup, */ |
|
|
195 | /* something using lots of bit arithmetic and a simple loop */ |
|
|
196 | /* we went for the loop */ |
|
|
197 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
269 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
198 | ecb_function_ int ecb_ld32 (uint32_t x) |
270 | ecb_function_ int ecb_ld32 (uint32_t x) |
199 | { |
271 | { |
200 | int r = 0; |
272 | int r = 0; |
201 | |
273 | |