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

Comparing libeio/ecb.h (file contents):
Revision 1.3 by root, Mon Jun 20 07:28:14 2011 UTC vs.
Revision 1.13 by root, Sat Jun 2 20:13:26 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:
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
59 typedef intptr_t ptrdiff_t;
47#else 60#else
48 #include <inttypes.h> 61 #include <inttypes.h>
62 #if UINTMAX_MAX > 0xffffffffU
63 #define ECB_PTRSIZE 8
64 #else
65 #define ECB_PTRSIZE 4
66 #endif
49#endif 67#endif
50 68
51/* many compilers define _GNUC_ to some versions but then only implement 69/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions, 70 * what their idiot authors think are the "more important" extensions,
53 * causing enourmous grief in return for some better fake benchmark numbers. 71 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so. 72 * or so.
55 * we try to detect these and simply assume they are not gcc - if they have 73 * 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. 74 * an issue with that they should have done it right in the first place.
57 */ 75 */
58#ifndef ECB_GCC_VERSION 76#ifndef ECB_GCC_VERSION
59 #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) 77 #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 78 #define ECB_GCC_VERSION(major,minor) 0
61 #else 79 #else
62 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 80 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
63 #endif 81 #endif
64#endif 82#endif
65 83
84#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
66#define ECB_C99 (__STDC_VERSION__ >= 199901L) 85#define ECB_C99 (__STDC_VERSION__ >= 199901L)
86#define ECB_C11 (__STDC_VERSION__ >= 201112L)
87#define ECB_CPP (__cplusplus+0)
88#define ECB_CPP11 (__cplusplus >= 201103L)
89
90/*****************************************************************************/
91
92/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
93/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
94
95#if ECB_NO_THREADS
96 #define ECB_NO_SMP 1
97#endif
98
99#if ECB_NO_SMP
100 #define ECB_MEMORY_FENCE do { } while (0)
101#endif
102
103#ifndef ECB_MEMORY_FENCE
104 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
105 #if __i386 || __i386__
106 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
107 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
108 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
109 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
110 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
111 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
112 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
113 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
114 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
115 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
116 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
117 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
118 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
119 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
120 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
121 #elif __sparc || __sparc__
122 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
123 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
124 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
125 #elif defined __s390__ || defined __s390x__
126 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
127 #elif defined __mips__
128 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
129 #elif defined __alpha__
130 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
131 #elif defined __hppa__
132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
133 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
134 #elif defined __ia64__
135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
136 #endif
137 #endif
138#endif
139
140#ifndef ECB_MEMORY_FENCE
141 #if ECB_GCC_VERSION(4,7)
142 /* see comment below (stdatomic.h) about the C11 memory model. */
143 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
144 #elif defined __clang && __has_feature (cxx_atomic)
145 /* see comment below (stdatomic.h) about the C11 memory model. */
146 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
147 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
148 #define ECB_MEMORY_FENCE __sync_synchronize ()
149 #elif _MSC_VER >= 1400 /* VC++ 2005 */
150 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
151 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
152 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
153 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
154 #elif defined _WIN32
155 #include <WinNT.h>
156 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
157 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
158 #include <mbarrier.h>
159 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
160 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
161 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
162 #elif __xlC__
163 #define ECB_MEMORY_FENCE __sync ()
164 #endif
165#endif
166
167#ifndef ECB_MEMORY_FENCE
168 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
169 /* we assume that these memory fences work on all variables/all memory accesses, */
170 /* not just C11 atomics and atomic accesses */
171 #include <stdatomic.h>
172 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
173 /* any fence other than seq_cst, which isn't very efficient for us. */
174 /* Why that is, we don't know - either the C11 memory model is quite useless */
175 /* for most usages, or gcc and clang have a bug */
176 /* I *currently* lean towards the latter, and inefficiently implement */
177 /* all three of ecb's fences as a seq_cst fence */
178 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
179 #endif
180#endif
181
182#ifndef ECB_MEMORY_FENCE
183 #if !ECB_AVOID_PTHREADS
184 /*
185 * if you get undefined symbol references to pthread_mutex_lock,
186 * or failure to find pthread.h, then you should implement
187 * the ECB_MEMORY_FENCE operations for your cpu/compiler
188 * OR provide pthread.h and link against the posix thread library
189 * of your system.
190 */
191 #include <pthread.h>
192 #define ECB_NEEDS_PTHREADS 1
193 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
194
195 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
196 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
197 #endif
198#endif
199
200#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
201 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
202#endif
203
204#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
205 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
206#endif
207
208/*****************************************************************************/
67 209
68#if __cplusplus 210#if __cplusplus
69 #define ecb_inline static inline 211 #define ecb_inline static inline
70#elif ECB_GCC_VERSION(2,5) 212#elif ECB_GCC_VERSION(2,5)
71 #define ecb_inline static __inline__ 213 #define ecb_inline static __inline__
110#elif ECB_GCC_VERSION(3,0) 252#elif ECB_GCC_VERSION(3,0)
111 #define ecb_decltype(x) __typeof(x) 253 #define ecb_decltype(x) __typeof(x)
112#endif 254#endif
113 255
114#define ecb_noinline ecb_attribute ((__noinline__)) 256#define ecb_noinline ecb_attribute ((__noinline__))
115#define ecb_noreturn ecb_attribute ((__noreturn__))
116#define ecb_unused ecb_attribute ((__unused__)) 257#define ecb_unused ecb_attribute ((__unused__))
117#define ecb_const ecb_attribute ((__const__)) 258#define ecb_const ecb_attribute ((__const__))
118#define ecb_pure ecb_attribute ((__pure__)) 259#define ecb_pure ecb_attribute ((__pure__))
260
261#if ECB_C11
262 #define ecb_noreturn _Noreturn
263#else
264 #define ecb_noreturn ecb_attribute ((__noreturn__))
265#endif
119 266
120#if ECB_GCC_VERSION(4,3) 267#if ECB_GCC_VERSION(4,3)
121 #define ecb_artificial ecb_attribute ((__artificial__)) 268 #define ecb_artificial ecb_attribute ((__artificial__))
122 #define ecb_hot ecb_attribute ((__hot__)) 269 #define ecb_hot ecb_attribute ((__hot__))
123 #define ecb_cold ecb_attribute ((__cold__)) 270 #define ecb_cold ecb_attribute ((__cold__))
189 x *= 0x01010101; 336 x *= 0x01010101;
190 337
191 return x >> 24; 338 return x >> 24;
192 } 339 }
193 340
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; 341 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
198 ecb_function_ int ecb_ld32 (uint32_t x) 342 ecb_function_ int ecb_ld32 (uint32_t x)
199 { 343 {
200 int r = 0; 344 int r = 0;
201 345
216 if (x >> 32) { x >>= 32; r += 32; } 360 if (x >> 32) { x >>= 32; r += 32; }
217 361
218 return r + ecb_ld32 (x); 362 return r + ecb_ld32 (x);
219 } 363 }
220#endif 364#endif
365
366ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
367ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
368ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
369ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
370
371ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
372ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
373{
374 return ( (x * 0x0802U & 0x22110U)
375 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
376}
377
378ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
379ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
380{
381 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
382 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
383 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
384 x = ( x >> 8 ) | ( x << 8);
385
386 return x;
387}
388
389ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
390ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
391{
392 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
393 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
394 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
395 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
396 x = ( x >> 16 ) | ( x << 16);
397
398 return x;
399}
221 400
222/* popcount64 is only available on 64 bit cpus as gcc builtin */ 401/* popcount64 is only available on 64 bit cpus as gcc builtin */
223/* so for this version we are lazy */ 402/* so for this version we are lazy */
224ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 403ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
225ecb_function_ int 404ecb_function_ int
275 454
276#if ECB_GCC_VERSION(4,5) 455#if ECB_GCC_VERSION(4,5)
277 #define ecb_unreachable() __builtin_unreachable () 456 #define ecb_unreachable() __builtin_unreachable ()
278#else 457#else
279 /* this seems to work fine, but gcc always emits a warning for it :/ */ 458 /* this seems to work fine, but gcc always emits a warning for it :/ */
280 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 459 ecb_inline void ecb_unreachable (void) ecb_noreturn;
281 ecb_function_ void ecb_unreachable (void) { } 460 ecb_inline void ecb_unreachable (void) { }
282#endif 461#endif
283 462
284/* try to tell the compiler that some condition is definitely true */ 463/* try to tell the compiler that some condition is definitely true */
285#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 464#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
286 465
287ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 466ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
288ecb_function_ unsigned char 467ecb_inline unsigned char
289ecb_byteorder_helper (void) 468ecb_byteorder_helper (void)
290{ 469{
291 const uint32_t u = 0x11223344; 470 const uint32_t u = 0x11223344;
292 return *(unsigned char *)&u; 471 return *(unsigned char *)&u;
293} 472}
294 473
295ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 474ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
296ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 475ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
297ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 476ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
298ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 477ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
299 478
300#if ECB_GCC_VERSION(3,0) || ECB_C99 479#if ECB_GCC_VERSION(3,0) || ECB_C99
301 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 480 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
302#else 481#else
303 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 482 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
483#endif
484
485#if __cplusplus
486 template<typename T>
487 static inline T ecb_div_rd (T val, T div)
488 {
489 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
490 }
491 template<typename T>
492 static inline T ecb_div_ru (T val, T div)
493 {
494 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
495 }
496#else
497 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
498 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
304#endif 499#endif
305 500
306#if ecb_cplusplus_does_not_suck 501#if ecb_cplusplus_does_not_suck
307 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 502 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
308 template<typename T, int N> 503 template<typename T, int N>

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines