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

Comparing libeio/ecb.h (file contents):
Revision 1.6 by root, Tue Jul 26 11:07:08 2011 UTC vs.
Revision 1.11 by root, Tue Apr 24 18:47:50 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
59 #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) 59 #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 60 #define ECB_GCC_VERSION(major,minor) 0
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/*****************************************************************************/ 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
72# define ECB_NO_SMP 1
73#endif
74
75#if ECB_NO_THREADS || ECB_NO_SMP
76 #define ECB_MEMORY_FENCE do { } while (0)
77#endif
78
68#ifndef ECB_MEMORY_FENCE 79#ifndef ECB_MEMORY_FENCE
69 #if ECB_GCC_VERSION(2,5) 80 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
70 #if __x86 81 #if __i386 || __i386__
71 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
72 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 83 #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 */ 84 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
74 #elif __amd64 85 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
75 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 86 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
76 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 87 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")
88 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
89 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
77 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") 90 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
91 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
92 || 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")
94 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
95 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
96 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
97 #elif __sparc || __sparc__
98 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory")
99 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
100 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
101 #elif defined __s390__ || defined __s390x__
102 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
103 #elif defined __mips__
104 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
105 #elif defined __alpha__
106 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
78 #endif 107 #endif
79 #endif 108 #endif
80#endif 109#endif
81 110
82#ifndef ECB_MEMORY_FENCE 111#ifndef ECB_MEMORY_FENCE
83 #if ECB_GCC_VERSION(4,4) 112 #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
84 #define ECB_MEMORY_FENCE __sync_synchronize () 113 #define ECB_MEMORY_FENCE __sync_synchronize ()
85 #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) 114 /*#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 ); }) 115 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
87 #elif _MSC_VER >= 1400 && 0 /* TODO: only true when using volatiles */ 116 #elif _MSC_VER >= 1400 /* VC++ 2005 */
117 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
88 #define ECB_MEMORY_FENCE do { } while (0) 118 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
89 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 119 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
90 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 120 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
91 #elif defined(_WIN32) 121 #elif defined _WIN32
92 #include <WinNT.h> 122 #include <WinNT.h>
123 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
124 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
125 #include <mbarrier.h>
93 #define ECB_MEMORY_FENCE MemoryBarrier () 126 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
94 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 127 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
95 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 128 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
129 #elif __xlC__
130 #define ECB_MEMORY_FENCE __sync ()
96 #endif 131 #endif
97#endif 132#endif
98 133
99#ifndef ECB_MEMORY_FENCE 134#ifndef ECB_MEMORY_FENCE
135 #if !ECB_AVOID_PTHREADS
136 /*
137 * if you get undefined symbol references to pthread_mutex_lock,
138 * or failure to find pthread.h, then you should implement
139 * the ECB_MEMORY_FENCE operations for your cpu/compiler
140 * OR provide pthread.h and link against the posix thread library
141 * of your system.
142 */
100 #include <pthread.h> 143 #include <pthread.h>
144 #define ECB_NEEDS_PTHREADS 1
145 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
101 146
102 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 147 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) 148 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
149 #endif
150#endif
151
152#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
104 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 153 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
154#endif
155
156#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
105 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 157 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
106#endif 158#endif
107 159
108/*****************************************************************************/ 160/*****************************************************************************/
109 161
… …
233 x *= 0x01010101; 285 x *= 0x01010101;
234 286
235 return x >> 24; 287 return x >> 24;
236 } 288 }
237 289
238 /* you have the choice beetween something with a table lookup, */
239 /* something using lots of bit arithmetic and a simple loop */
240 /* we went for the loop */
241 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 290 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
242 ecb_function_ int ecb_ld32 (uint32_t x) 291 ecb_function_ int ecb_ld32 (uint32_t x)
243 { 292 {
244 int r = 0; 293 int r = 0;
245 294
… …
260 if (x >> 32) { x >>= 32; r += 32; } 309 if (x >> 32) { x >>= 32; r += 32; }
261 310
262 return r + ecb_ld32 (x); 311 return r + ecb_ld32 (x);
263 } 312 }
264#endif 313#endif
314
315ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
316ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
317{
318 return ( (x * 0x0802U & 0x22110U)
319 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
320}
321
322ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
323ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
324{
325 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
326 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
327 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
328 x = ( x >> 8 ) | ( x << 8);
329
330 return x;
331}
332
333ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
334ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
335{
336 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
337 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
338 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
339 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
340 x = ( x >> 16 ) | ( x << 16);
341
342 return x;
343}
265 344
266/* popcount64 is only available on 64 bit cpus as gcc builtin */ 345/* popcount64 is only available on 64 bit cpus as gcc builtin */
267/* so for this version we are lazy */ 346/* so for this version we are lazy */
268ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 347ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
269ecb_function_ int 348ecb_function_ int
… …
319 398
320#if ECB_GCC_VERSION(4,5) 399#if ECB_GCC_VERSION(4,5)
321 #define ecb_unreachable() __builtin_unreachable () 400 #define ecb_unreachable() __builtin_unreachable ()
322#else 401#else
323 /* this seems to work fine, but gcc always emits a warning for it :/ */ 402 /* this seems to work fine, but gcc always emits a warning for it :/ */
324 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 403 ecb_inline void ecb_unreachable (void) ecb_noreturn;
325 ecb_function_ void ecb_unreachable (void) { } 404 ecb_inline void ecb_unreachable (void) { }
326#endif 405#endif
327 406
328/* try to tell the compiler that some condition is definitely true */ 407/* try to tell the compiler that some condition is definitely true */
329#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 408#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
330 409
331ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 410ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
332ecb_function_ unsigned char 411ecb_inline unsigned char
333ecb_byteorder_helper (void) 412ecb_byteorder_helper (void)
334{ 413{
335 const uint32_t u = 0x11223344; 414 const uint32_t u = 0x11223344;
336 return *(unsigned char *)&u; 415 return *(unsigned char *)&u;
337} 416}
338 417
339ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 418ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
340ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 419ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
341ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 420ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
342ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 421ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
343 422
344#if ECB_GCC_VERSION(3,0) || ECB_C99 423#if ECB_GCC_VERSION(3,0) || ECB_C99
345 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 424 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
346#else 425#else
347 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 426 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
427#endif
428
429#if __cplusplus
430 template<typename T>
431 static inline T ecb_div_rd (T val, T div)
432 {
433 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
434 }
435 template<typename T>
436 static inline T ecb_div_ru (T val, T div)
437 {
438 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
439 }
440#else
441 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
442 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
348#endif 443#endif
349 444
350#if ecb_cplusplus_does_not_suck 445#if ecb_cplusplus_does_not_suck
351 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 446 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
352 template<typename T, int N> 447 template<typename T, int N>

Diff Legend

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