ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
(Generate patch)

Comparing libecb/ecb.h (file contents):
Revision 1.56 by sf-exg, Thu Jul 28 13:45:44 2011 UTC vs.
Revision 1.75 by root, Wed Jan 25 19:57:49 2012 UTC

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:
… …
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
… …
63 #endif 63 #endif
64#endif 64#endif
65 65
66/*****************************************************************************/ 66/*****************************************************************************/
67 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
68#ifndef ECB_MEMORY_FENCE 75#ifndef ECB_MEMORY_FENCE
69 #if ECB_GCC_VERSION(2,5) 76 #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
70 #if __x86 77 #if __i386 || __i386__
71 #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")
72 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 79 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */
73 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE /* better be safe than sorry */ 80 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
74 #elif __amd64 81 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
75 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
76 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "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__
77 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") 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,c10,5" : : "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__ ("dmb" : : : "memory")
93 #elif __sparc || __sparc__
94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory")
95 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
96 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
78 #endif 97 #endif
79 #endif 98 #endif
80#endif 99#endif
81 100
82#ifndef ECB_MEMORY_FENCE 101#ifndef ECB_MEMORY_FENCE
83 #if ECB_GCC_VERSION(4,4) 102 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__)
84 #define ECB_MEMORY_FENCE __sync_synchronize () 103 #define ECB_MEMORY_FENCE __sync_synchronize ()
85 #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); }) */
86 #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 ); }) */
87 #elif _MSC_VER >= 1400 && 0 /* TODO: only true when using volatiles */ 106 #elif _MSC_VER >= 1400 /* VC++ 2005 */
107 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
88 #define ECB_MEMORY_FENCE do { } while (0) 108 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
89 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 109 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
90 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 110 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
91 #elif defined(_WIN32) 111 #elif defined(_WIN32)
92 #include <WinNT.h> 112 #include <WinNT.h>
113 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
114 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
115 #include <mbarrier.h>
93 #define ECB_MEMORY_FENCE MemoryBarrier () 116 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
94 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 117 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
95 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 118 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
96 #endif 119 #endif
97#endif 120#endif
98 121
99#ifndef ECB_MEMORY_FENCE 122#ifndef ECB_MEMORY_FENCE
123 #if !ECB_AVOID_PTHREADS
124 /*
125 * if you get undefined symbol references to pthread_mutex_lock,
126 * or failure to find pthread.h, then you should implement
127 * the ECB_MEMORY_FENCE operations for your cpu/compiler
128 * OR provide pthread.h and link against the posix thread library
129 * of your system.
130 */
100 #include <pthread.h> 131 #include <pthread.h>
132 #define ECB_NEEDS_PTHREADS 1
133 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
101 134
102 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 135 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
103 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) 136 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
137 #endif
138#endif
139
140#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE)
104 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 141 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
142#endif
143
144#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE)
105 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 145 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
106#endif 146#endif
107 147
108/*****************************************************************************/ 148/*****************************************************************************/
109 149
… …
258 298
259 return r + ecb_ld32 (x); 299 return r + ecb_ld32 (x);
260 } 300 }
261#endif 301#endif
262 302
303ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
304ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
305{
306 return ( (x * 0x0802U & 0x22110U)
307 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
308}
309
310ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
311ecb_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
321ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
322ecb_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
263/* popcount64 is only available on 64 bit cpus as gcc builtin */ 333/* popcount64 is only available on 64 bit cpus as gcc builtin */
264/* so for this version we are lazy */ 334/* so for this version we are lazy */
265ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 335ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
266ecb_function_ int 336ecb_function_ int
267ecb_popcount64 (uint64_t x) 337ecb_popcount64 (uint64_t x)
… …
316 386
317#if ECB_GCC_VERSION(4,5) 387#if ECB_GCC_VERSION(4,5)
318 #define ecb_unreachable() __builtin_unreachable () 388 #define ecb_unreachable() __builtin_unreachable ()
319#else 389#else
320 /* this seems to work fine, but gcc always emits a warning for it :/ */ 390 /* this seems to work fine, but gcc always emits a warning for it :/ */
321 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 391 ecb_inline void ecb_unreachable (void) ecb_noreturn;
322 ecb_function_ void ecb_unreachable (void) { } 392 ecb_inline void ecb_unreachable (void) { }
323#endif 393#endif
324 394
325/* try to tell the compiler that some condition is definitely true */ 395/* try to tell the compiler that some condition is definitely true */
326#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 396#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
327 397
… …
342 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 412 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
343#else 413#else
344 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 414 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
345#endif 415#endif
346 416
417#if __cplusplus
418 template<typename T>
419 static inline T ecb_div_rd (T val, T div)
420 {
421 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
422 }
423 template<typename T>
424 static inline T ecb_div_ru (T val, T div)
425 {
426 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
427 }
428#else
429 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
430 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
431#endif
432
347#if ecb_cplusplus_does_not_suck 433#if ecb_cplusplus_does_not_suck
348 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 434 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
349 template<typename T, int N> 435 template<typename T, int N>
350 static inline int ecb_array_length (const T (&arr)[N]) 436 static inline int ecb_array_length (const T (&arr)[N])
351 { 437 {

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines