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 (©) 2009-2012 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: |
… | |
… | |
71 | #if ECB_NO_THREADS || ECB_NO_SMP |
71 | #if ECB_NO_THREADS || ECB_NO_SMP |
72 | #define ECB_MEMORY_FENCE do { } while (0) |
72 | #define ECB_MEMORY_FENCE do { } while (0) |
73 | #endif |
73 | #endif |
74 | |
74 | |
75 | #ifndef ECB_MEMORY_FENCE |
75 | #ifndef ECB_MEMORY_FENCE |
76 | #if ECB_GCC_VERSION(2,5) |
76 | #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
77 | #if __i386__ |
77 | #if __i386__ |
78 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
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 */ |
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 */ |
80 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
81 | #elif __amd64 |
81 | #elif __amd64 |
… | |
… | |
88 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) |
88 | || 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") |
89 | #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__ ) \ |
90 | #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
91 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
91 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
92 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
92 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
93 | #elif defined(__sparc) |
|
|
94 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #StoreLoad | #LoadLoad | #StoreStore" : : : "memory") |
|
|
95 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadLoad" : : : "memory") |
|
|
96 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #StoreStore") |
93 | #endif |
97 | #endif |
94 | #endif |
98 | #endif |
95 | #endif |
99 | #endif |
96 | |
100 | |
97 | #ifndef ECB_MEMORY_FENCE |
101 | #ifndef ECB_MEMORY_FENCE |
98 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) |
102 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__) |
99 | #define ECB_MEMORY_FENCE __sync_synchronize () |
103 | #define ECB_MEMORY_FENCE __sync_synchronize () |
100 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
104 | /*#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 ); }) */ |
105 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
102 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
106 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
103 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
107 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
… | |
… | |
105 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
109 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
106 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
110 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
107 | #elif defined(_WIN32) |
111 | #elif defined(_WIN32) |
108 | #include <WinNT.h> |
112 | #include <WinNT.h> |
109 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
113 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
114 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
115 | #include <mbarrier.h> |
|
|
116 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
117 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
118 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
110 | #endif |
119 | #endif |
111 | #endif |
120 | #endif |
112 | |
121 | |
113 | #ifndef ECB_MEMORY_FENCE |
122 | #ifndef ECB_MEMORY_FENCE |
114 | #if !ECB_AVOID_PTHREADS |
123 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
289 | |
298 | |
290 | return r + ecb_ld32 (x); |
299 | return r + ecb_ld32 (x); |
291 | } |
300 | } |
292 | #endif |
301 | #endif |
293 | |
302 | |
|
|
303 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
304 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
305 | { |
|
|
306 | return ( (x * 0x0802U & 0x22110U) |
|
|
307 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
308 | } |
|
|
309 | |
|
|
310 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
311 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
312 | { |
|
|
313 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
314 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
315 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
316 | x = ( x >> 8 ) | ( x << 8); |
|
|
317 | |
|
|
318 | return x; |
|
|
319 | } |
|
|
320 | |
|
|
321 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
322 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
323 | { |
|
|
324 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
325 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
326 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
327 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
328 | x = ( x >> 16 ) | ( x << 16); |
|
|
329 | |
|
|
330 | return x; |
|
|
331 | } |
|
|
332 | |
294 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
333 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
295 | /* so for this version we are lazy */ |
334 | /* so for this version we are lazy */ |
296 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
335 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
297 | ecb_function_ int |
336 | ecb_function_ int |
298 | ecb_popcount64 (uint64_t x) |
337 | ecb_popcount64 (uint64_t x) |