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

Comparing libeio/ecb.h (file contents):
Revision 1.1 by root, Fri Jun 10 12:35:18 2011 UTC vs.
Revision 1.14 by root, Sun Jun 17 17:07:25 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
36#ifdef _WIN32
37 typedef signed char int8_t;
38 typedef unsigned char uint8_t;
39 typedef signed short int16_t;
40 typedef unsigned short uint16_t;
41 typedef signed int int32_t;
42 typedef unsigned int uint32_t;
43 #if __GNUC__
44 typedef signed long long int64_t;
45 typedef unsigned long long uint64_t;
46 #else /* _MSC_VER || __BORLANDC__ */
47 typedef signed __int64 int64_t;
48 typedef unsigned __int64 uint64_t;
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;
60#else
33#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
67#endif
34 68
35/* many compilers define _GNUC_ to some versions but then only implement 69/* many compilers define _GNUC_ to some versions but then only implement
36 * what their idiot authors think are the "more important" extensions, 70 * what their idiot authors think are the "more important" extensions,
37 * causing enourmous grief in return for some better fake benchmark numbers. 71 * causing enormous grief in return for some better fake benchmark numbers.
38 * or so. 72 * or so.
39 * 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
40 * 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.
41 */ 75 */
42#ifndef ECB_GCC_VERSION 76#ifndef ECB_GCC_VERSION
43# if defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__llvm__) 77 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
44# define ECB_GCC_VERSION(major,minor) 0 78 #define ECB_GCC_VERSION(major,minor) 0
45# else 79 #else
46# 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)))
81 #endif
47# endif 82#endif
83
84#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
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
48#endif 137 #endif
138#endif
49 139
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/*****************************************************************************/
209
50#ifndef __cplusplus 210#if __cplusplus
51# if __STDC_VERSION__ >= 199901L 211 #define ecb_inline static inline
52# define ECB_INLINE static inline 212#elif ECB_GCC_VERSION(2,5)
53typedef _Bool ecb_bool; 213 #define ecb_inline static __inline__
214#elif ECB_C99
215 #define ecb_inline static inline
54# else 216#else
55# define ECB_INLINE static inline /* yeah! */ 217 #define ecb_inline static
218#endif
219
220#if ECB_GCC_VERSION(3,3)
221 #define ecb_restrict __restrict__
222#elif ECB_C99
223 #define ecb_restrict restrict
224#else
225 #define ecb_restrict
226#endif
227
56typedef int ecb_bool; 228typedef int ecb_bool;
57# endif
58#else
59# define ECB_INLINE static inline
60typedef bool ecb_bool;
61#endif
62 229
63#define ECB_CONCAT_(a, b) a ## b 230#define ECB_CONCAT_(a, b) a ## b
64#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 231#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
65#define ECB_STRINGIFY_(a) # a 232#define ECB_STRINGIFY_(a) # a
66#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 233#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
67 234
68#define ECB_HEADER_INLINE ECB_INLINE 235#define ecb_function_ ecb_inline
69 236
70#if ECB_GCC_VERSION(3,1) 237#if ECB_GCC_VERSION(3,1)
71# define ecb_attribute(attrlist) __attribute__(attrlist) 238 #define ecb_attribute(attrlist) __attribute__(attrlist)
72# define ecb_is_constant(expr) __builtin_constant_p (expr) 239 #define ecb_is_constant(expr) __builtin_constant_p (expr)
73# define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 240 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
74# define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 241 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
75#else 242#else
76# define ecb_attribute(attrlist) 243 #define ecb_attribute(attrlist)
77# define ecb_is_constant(expr) 0 244 #define ecb_is_constant(expr) 0
78# define ecb_expect(expr,value) (expr) 245 #define ecb_expect(expr,value) (expr)
79# define ecb_prefetch(addr,rw,locality) 246 #define ecb_prefetch(addr,rw,locality)
80#endif 247#endif
81 248
82/* no emulation for ecb_decltype */ 249/* no emulation for ecb_decltype */
83#if ECB_GCC_VERSION(4,5) 250#if ECB_GCC_VERSION(4,5)
84# define ecb_decltype(x) __decltype(x) 251 #define ecb_decltype(x) __decltype(x)
85#elif ECB_GCC_VERSION(3,0) 252#elif ECB_GCC_VERSION(3,0)
86# define ecb_decltype(x) typeof(x) 253 #define ecb_decltype(x) __typeof(x)
87#endif 254#endif
88 255
89#define ecb_artificial ecb_attribute ((__artificial__)) /* 4.3 */
90#define ecb_noinline ecb_attribute ((__noinline__)) 256#define ecb_noinline ecb_attribute ((__noinline__))
91#define ecb_noreturn ecb_attribute ((__noreturn__))
92#define ecb_unused ecb_attribute ((__unused__)) 257#define ecb_unused ecb_attribute ((__unused__))
93#define ecb_const ecb_attribute ((__const__)) 258#define ecb_const ecb_attribute ((__const__))
94#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
266
267#if ECB_GCC_VERSION(4,3)
268 #define ecb_artificial ecb_attribute ((__artificial__))
95#define ecb_hot ecb_attribute ((__hot__)) /* 4.3 */ 269 #define ecb_hot ecb_attribute ((__hot__))
96#define ecb_cold ecb_attribute ((__cold__)) /* 4.3 */ 270 #define ecb_cold ecb_attribute ((__cold__))
271#else
272 #define ecb_artificial
273 #define ecb_hot
274 #define ecb_cold
275#endif
97 276
98/* put into if's if you are very sure that the expression */ 277/* put around conditional expressions if you are very sure that the */
99/* is mostly true or mostly false. note that these return */ 278/* expression is mostly true or mostly false. note that these return */
100/* booleans, not the expression. */ 279/* booleans, not the expression. */
101#define ecb_expect_false(expr) ecb_expect (!!(expr), 0) 280#define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
102#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) 281#define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
282/* for compatibility to the rest of the world */
103#define ecb_likely(expr) ecb_expect_true (expr) 283#define ecb_likely(expr) ecb_expect_true (expr)
104#define ecb_unlikely(expr) ecb_expect_false (expr) 284#define ecb_unlikely(expr) ecb_expect_false (expr)
105 285
106/* try to tell the compiler that some condition is definitely true */
107#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
108
109/* count trailing zero bits and count # of one bits */ 286/* count trailing zero bits and count # of one bits */
110#if ECB_GCC_VERSION(3,4) 287#if ECB_GCC_VERSION(3,4)
288 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
289 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
290 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
111#define ecb_ctz32(x) __builtin_ctz (x) 291 #define ecb_ctz32(x) __builtin_ctz (x)
292 #define ecb_ctz64(x) __builtin_ctzll (x)
112#define ecb_popcount32(x) __builtin_popcount (x) 293 #define ecb_popcount32(x) __builtin_popcount (x)
294 /* no popcountll */
113#else 295#else
114ECB_HEADER_INLINE int ecb_ctz32 (uint32_t x) ecb_const; 296 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
115ECB_HEADER_INLINE int 297 ecb_function_ int
116ecb_ctz32 (uint32_t x) 298 ecb_ctz32 (uint32_t x)
299 {
300 int r = 0;
301
302 x &= ~x + 1; /* this isolates the lowest bit */
303
304#if ECB_branchless_on_i386
305 r += !!(x & 0xaaaaaaaa) << 0;
306 r += !!(x & 0xcccccccc) << 1;
307 r += !!(x & 0xf0f0f0f0) << 2;
308 r += !!(x & 0xff00ff00) << 3;
309 r += !!(x & 0xffff0000) << 4;
310#else
311 if (x & 0xaaaaaaaa) r += 1;
312 if (x & 0xcccccccc) r += 2;
313 if (x & 0xf0f0f0f0) r += 4;
314 if (x & 0xff00ff00) r += 8;
315 if (x & 0xffff0000) r += 16;
316#endif
317
318 return r;
319 }
320
321 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
322 ecb_function_ int
323 ecb_ctz64 (uint64_t x)
324 {
325 int shift = x & 0xffffffffU ? 0 : 32;
326 return ecb_ctz32 (x >> shift) + shift;
327 }
328
329 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
330 ecb_function_ int
331 ecb_popcount32 (uint32_t x)
332 {
333 x -= (x >> 1) & 0x55555555;
334 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
335 x = ((x >> 4) + x) & 0x0f0f0f0f;
336 x *= 0x01010101;
337
338 return x >> 24;
339 }
340
341 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
342 ecb_function_ int ecb_ld32 (uint32_t x)
343 {
344 int r = 0;
345
346 if (x >> 16) { x >>= 16; r += 16; }
347 if (x >> 8) { x >>= 8; r += 8; }
348 if (x >> 4) { x >>= 4; r += 4; }
349 if (x >> 2) { x >>= 2; r += 2; }
350 if (x >> 1) { r += 1; }
351
352 return r;
353 }
354
355 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
356 ecb_function_ int ecb_ld64 (uint64_t x)
357 {
358 int r = 0;
359
360 if (x >> 32) { x >>= 32; r += 32; }
361
362 return r + ecb_ld32 (x);
363 }
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)
117{ 373{
118 int r = 0; 374 return ( (x * 0x0802U & 0x22110U)
119 375 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
120 x &= -x; /* this isolates the lowest bit */
121
122 if (x & 0xaaaaaaaa) r += 1;
123 if (x & 0xcccccccc) r += 2;
124 if (x & 0xf0f0f0f0) r += 4;
125 if (x & 0xff00ff00) r += 8;
126 if (x & 0xffff0000) r += 16;
127
128 return r;
129} 376}
130 377
131ECB_HEADER_INLINE int ecb_popcount32 (uint32_t x) ecb_const; 378ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
132ECB_HEADER_INLINE int 379ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
133ecb_popcount32 (uint32_t x)
134{ 380{
135 x -= (x >> 1) & 0x55555555; 381 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
136 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 382 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
137 x = ((x >> 4) + x) & 0x0f0f0f0f; 383 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
138 x *= 0x01010101; 384 x = ( x >> 8 ) | ( x << 8);
139 385
140 return x >> 24; 386 return x;
141} 387}
142#endif 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}
400
401/* popcount64 is only available on 64 bit cpus as gcc builtin */
402/* so for this version we are lazy */
403ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
404ecb_function_ int
405ecb_popcount64 (uint64_t x)
406{
407 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
408}
409
410ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
411ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
412ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
413ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
414ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
415ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
416ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
417ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
418
419ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
420ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
421ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
422ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
423ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
424ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
425ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
426ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
143 427
144#if ECB_GCC_VERSION(4,3) 428#if ECB_GCC_VERSION(4,3)
145# define ecb_bswap32(x) __builtin_bswap32 (x)
146# define ecb_bswap16(x) (__builtin_bswap32(x) >> 16) 429 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
430 #define ecb_bswap32(x) __builtin_bswap32 (x)
431 #define ecb_bswap64(x) __builtin_bswap64 (x)
147#else 432#else
433 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
434 ecb_function_ uint16_t
435 ecb_bswap16 (uint16_t x)
436 {
437 return ecb_rotl16 (x, 8);
438 }
439
148ECB_HEADER_INLINE uint32_t ecb_bswap32 (uint32_t x) ecb_const; 440 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
149ECB_HEADER_INLINE uint32_t 441 ecb_function_ uint32_t
150ecb_bswap32 (uint32_t x) 442 ecb_bswap32 (uint32_t x)
151{ 443 {
152 return (x >> 24) 444 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
153 | ((x >> 8) & 0x0000ff00) 445 }
154 | ((x << 8) & 0x00ff0000)
155 | (x << 24);
156}
157 446
158ECB_HEADER_INLINE uint32_t ecb_bswap16 (uint32_t x) ecb_const; 447 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
159ECB_HEADER_INLINE uint32_t 448 ecb_function_ uint64_t
160ecb_bswap16 (uint32_t x) 449 ecb_bswap64 (uint64_t x)
161{ 450 {
162 return ((x >> 8) & 0xff) 451 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
163 | ((x << 8) & 0x00ff0000) 452 }
164 | (x << 24);
165}
166#endif 453#endif
167 454
168#if ECB_GCC_VERSION(4,5) 455#if ECB_GCC_VERSION(4,5)
169# define ecb_unreachable() __builtin_unreachable () 456 #define ecb_unreachable() __builtin_unreachable ()
170#else 457#else
171/* 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 :/ */
172ECB_HEADER_INLINE void ecb_unreachable (void) ecb_noreturn; 459 ecb_inline void ecb_unreachable (void) ecb_noreturn;
173ECB_HEADER_INLINE void ecb_unreachable (void) { } 460 ecb_inline void ecb_unreachable (void) { }
174#endif 461#endif
175 462
463/* try to tell the compiler that some condition is definitely true */
464#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
465
176ECB_HEADER_INLINE unsigned char ecb_byteorder_helper (void) ecb_const; 466ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
177ECB_HEADER_INLINE unsigned char 467ecb_inline unsigned char
178ecb_byteorder_helper (void) 468ecb_byteorder_helper (void)
179{ 469{
180 const uint32_t u = 0x11223344; 470 /* the union code still generates code under pressure in gcc, */
181 return *(unsigned char *)&u; 471 /* but less than using pointers, and always seem to */
472 /* successfully return a constant. */
473 /* the reason why we have this horrible preprocessor mess */
474 /* is to avoid it in all cases, at least on common architectures */
475 /* and yes, gcc defines __BYTE_ORDER__, g++ does not */
476#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
477 return 0x44;
478#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
479 return 0x44;
480#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
481 retrurn 0x11;
482#else
483 union
484 {
485 uint32_t i;
486 uint8_t c;
487 } u = { 0x11223344 };
488 return u.c;
489#endif
182} 490}
183 491
184ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) ecb_const; 492ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
185ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }; 493ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
186ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) ecb_const; 494ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
187ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }; 495ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
188 496
189#if ECB_GCC_VERSION(3,0) 497#if ECB_GCC_VERSION(3,0) || ECB_C99
190# define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 498 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
191#else 499#else
192# define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 500 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
501#endif
502
503#if __cplusplus
504 template<typename T>
505 static inline T ecb_div_rd (T val, T div)
506 {
507 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
508 }
509 template<typename T>
510 static inline T ecb_div_ru (T val, T div)
511 {
512 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
513 }
514#else
515 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
516 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
193#endif 517#endif
194 518
195#if ecb_cplusplus_does_not_suck 519#if ecb_cplusplus_does_not_suck
196// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) 520 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
197template<typename T, int N> 521 template<typename T, int N>
198static inline int ecb_array_length (const T (&arr)[N]) 522 static inline int ecb_array_length (const T (&arr)[N])
199{ 523 {
200 return N; 524 return N;
201} 525 }
202#else 526#else
203#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 527 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
204#endif 528#endif
205 529
206ECB_INLINE uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
207ECB_INLINE uint32_t
208ecb_rotr32 (uint32_t x, unsigned int count)
209{
210 return (x << (32 - count)) | (x >> count);
211}
212
213ECB_INLINE uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
214ECB_INLINE uint32_t
215ecb_rotl32 (uint32_t x, unsigned int count)
216{
217 return (x >> (32 - count)) | (x << count);
218}
219
220#endif 530#endif
221 531

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines