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

Comparing libecb/ecb.h (file contents):
Revision 1.23 by root, Thu May 26 19:40:09 2011 UTC vs.
Revision 1.101 by root, Thu Jun 28 14:50:01 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 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#else
33#include <inttypes.h> 60 #include <inttypes.h>
61 #if UINTMAX_MAX > 0xffffffffU
62 #define ECB_PTRSIZE 8
63 #else
64 #define ECB_PTRSIZE 4
65 #endif
66#endif
34 67
35/* many compilers define _GNUC_ to some versions but then only implement 68/* many compilers define _GNUC_ to some versions but then only implement
36 * what their idiot authors think are the "more important" extensions, 69 * what their idiot authors think are the "more important" extensions,
37 * causing enourmous grief in return for some better fake benchmark numbers. 70 * causing enormous grief in return for some better fake benchmark numbers.
38 * or so. 71 * or so.
39 * we try to detect these and simply assume they are not gcc - if they have 72 * 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. 73 * an issue with that they should have done it right in the first place.
41 */ 74 */
42#ifndef ECB_GCC_VERSION 75#ifndef ECB_GCC_VERSION
43# if defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__llvm__) 76 #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 77 #define ECB_GCC_VERSION(major,minor) 0
45# else 78 #else
46# define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 79 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
80 #endif
47# endif 81#endif
82
83#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
84#define ECB_C99 (__STDC_VERSION__ >= 199901L)
85#define ECB_C11 (__STDC_VERSION__ >= 201112L)
86#define ECB_CPP (__cplusplus+0)
87#define ECB_CPP11 (__cplusplus >= 201103L)
88
89/*****************************************************************************/
90
91/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
92/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
93
94#if ECB_NO_THREADS
95 #define ECB_NO_SMP 1
96#endif
97
98#if ECB_NO_SMP
99 #define ECB_MEMORY_FENCE do { } while (0)
100#endif
101
102#ifndef ECB_MEMORY_FENCE
103 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
104 #if __i386 || __i386__
105 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
106 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
107 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
108 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
109 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
110 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
111 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
112 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
113 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
114 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
115 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
116 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
117 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
118 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
119 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
120 #elif __sparc || __sparc__
121 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
122 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
123 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
124 #elif defined __s390__ || defined __s390x__
125 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
126 #elif defined __mips__
127 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
128 #elif defined __alpha__
129 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
130 #elif defined __hppa__
131 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
132 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
133 #elif defined __ia64__
134 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
135 #endif
48#endif 136 #endif
137#endif
49 138
139#ifndef ECB_MEMORY_FENCE
140 #if ECB_GCC_VERSION(4,7)
141 /* see comment below (stdatomic.h) about the C11 memory model. */
142 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
143 #elif defined __clang && __has_feature (cxx_atomic)
144 /* see comment below (stdatomic.h) about the C11 memory model. */
145 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
146 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
147 #define ECB_MEMORY_FENCE __sync_synchronize ()
148 #elif _MSC_VER >= 1400 /* VC++ 2005 */
149 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
150 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
151 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
152 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
153 #elif defined _WIN32
154 #include <WinNT.h>
155 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
156 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
157 #include <mbarrier.h>
158 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
159 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
160 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
161 #elif __xlC__
162 #define ECB_MEMORY_FENCE __sync ()
163 #endif
164#endif
165
166#ifndef ECB_MEMORY_FENCE
167 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
168 /* we assume that these memory fences work on all variables/all memory accesses, */
169 /* not just C11 atomics and atomic accesses */
170 #include <stdatomic.h>
171 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
172 /* any fence other than seq_cst, which isn't very efficient for us. */
173 /* Why that is, we don't know - either the C11 memory model is quite useless */
174 /* for most usages, or gcc and clang have a bug */
175 /* I *currently* lean towards the latter, and inefficiently implement */
176 /* all three of ecb's fences as a seq_cst fence */
177 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
178 #endif
179#endif
180
181#ifndef ECB_MEMORY_FENCE
182 #if !ECB_AVOID_PTHREADS
183 /*
184 * if you get undefined symbol references to pthread_mutex_lock,
185 * or failure to find pthread.h, then you should implement
186 * the ECB_MEMORY_FENCE operations for your cpu/compiler
187 * OR provide pthread.h and link against the posix thread library
188 * of your system.
189 */
190 #include <pthread.h>
191 #define ECB_NEEDS_PTHREADS 1
192 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
193
194 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
195 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
196 #endif
197#endif
198
199#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
200 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
201#endif
202
203#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
204 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
205#endif
206
207/*****************************************************************************/
208
50#ifndef __cplusplus 209#if __cplusplus
51# if __STDC_VERSION__ >= 199901L 210 #define ecb_inline static inline
52# define ECB_INLINE inline 211#elif ECB_GCC_VERSION(2,5)
53typedef _Bool ecb_bool; 212 #define ecb_inline static __inline__
213#elif ECB_C99
214 #define ecb_inline static inline
54# else 215#else
55# define ECB_INLINE inline /* yeah! */ 216 #define ecb_inline static
217#endif
218
219#if ECB_GCC_VERSION(3,3)
220 #define ecb_restrict __restrict__
221#elif ECB_C99
222 #define ecb_restrict restrict
223#else
224 #define ecb_restrict
225#endif
226
56typedef int ecb_bool; 227typedef int ecb_bool;
57# endif
58#endif
59 228
60#define ECB_CONCAT_(a, b) a ## b 229#define ECB_CONCAT_(a, b) a ## b
61#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 230#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
62#define ECB_STRINGIFY_(a) # a 231#define ECB_STRINGIFY_(a) # a
63#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 232#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
64 233
65#define ECB_HEADER_INLINE static ECB_INLINE 234#define ecb_function_ ecb_inline
66 235
67#if ECB_GCC_VERSION(3,1) 236#if ECB_GCC_VERSION(3,1)
68# define ecb_attribute(attrlist) __attribute__(attrlist) 237 #define ecb_attribute(attrlist) __attribute__(attrlist)
69# define ecb_is_constant(expr) __builtin_constant_p (expr) 238 #define ecb_is_constant(expr) __builtin_constant_p (expr)
70# define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 239 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
71# define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 240 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
72#else 241#else
73# define ecb_attribute(attrlist) 242 #define ecb_attribute(attrlist)
74# define ecb_is_constant(expr) 0 243 #define ecb_is_constant(expr) 0
75# define ecb_expect(expr,value) (expr) 244 #define ecb_expect(expr,value) (expr)
76# define ecb_prefetch(addr,rw,locality) 245 #define ecb_prefetch(addr,rw,locality)
77#endif 246#endif
78 247
79/* no emulation for ecb_decltype */ 248/* no emulation for ecb_decltype */
80#if ECB_GCC_VERSION(4,5) 249#if ECB_GCC_VERSION(4,5)
81# define ecb_decltype(x) __decltype(x) 250 #define ecb_decltype(x) __decltype(x)
82#elif ECB_GCC_VERSION(3,0) 251#elif ECB_GCC_VERSION(3,0)
83# define ecb_decltype(x) typeof(x) 252 #define ecb_decltype(x) __typeof(x)
84#endif 253#endif
85 254
86#define ecb_noinline ecb_attribute ((noinline)) 255#define ecb_noinline ecb_attribute ((__noinline__))
87#define ecb_noreturn ecb_attribute ((noreturn))
88#define ecb_unused ecb_attribute ((unused)) 256#define ecb_unused ecb_attribute ((__unused__))
89#define ecb_const ecb_attribute ((const)) 257#define ecb_const ecb_attribute ((__const__))
90#define ecb_pure ecb_attribute ((pure)) 258#define ecb_pure ecb_attribute ((__pure__))
259
260#if ECB_C11
261 #define ecb_noreturn _Noreturn
262#else
263 #define ecb_noreturn ecb_attribute ((__noreturn__))
264#endif
265
266#if ECB_GCC_VERSION(4,3)
267 #define ecb_artificial ecb_attribute ((__artificial__))
91#define ecb_hot ecb_attribute ((hot)) /* 4.3 */ 268 #define ecb_hot ecb_attribute ((__hot__))
92#define ecb_cold ecb_attribute ((cold)) /* 4.3 */ 269 #define ecb_cold ecb_attribute ((__cold__))
270#else
271 #define ecb_artificial
272 #define ecb_hot
273 #define ecb_cold
274#endif
93 275
94/* put into if's if you are very sure that the expression */ 276/* put around conditional expressions if you are very sure that the */
95/* is mostly true or mosty false. note that these return */ 277/* expression is mostly true or mostly false. note that these return */
96/* booleans, not the expression. */ 278/* booleans, not the expression. */
97#define ecb_unlikely(expr) ecb_expect (!!(expr), 0) 279#define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
280#define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
281/* for compatibility to the rest of the world */
98#define ecb_likely(expr) ecb_expect (!!(expr), 1) 282#define ecb_likely(expr) ecb_expect_true (expr)
99 283#define ecb_unlikely(expr) ecb_expect_false (expr)
100/* try to tell the compiler that some condition is definitely true */
101#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
102 284
103/* count trailing zero bits and count # of one bits */ 285/* count trailing zero bits and count # of one bits */
104#if ECB_GCC_VERSION(3,4) 286#if ECB_GCC_VERSION(3,4)
287 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
288 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
289 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
105#define ecb_ctz32 (x) __builtin_ctz (x) 290 #define ecb_ctz32(x) __builtin_ctz (x)
291 #define ecb_ctz64(x) __builtin_ctzll (x)
106#define ecb_popcount32 (x) __builtin_popcount (x) 292 #define ecb_popcount32(x) __builtin_popcount (x)
293 /* no popcountll */
107#else 294#else
108ECB_HEADER_INLINE int ecb_ctz32 (uint32_t x) ecb_const; 295 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
109ECB_HEADER_INLINE 296 ecb_function_ int
110ecb_ctz32 (uint32_t x) 297 ecb_ctz32 (uint32_t x)
298 {
299 int r = 0;
300
301 x &= ~x + 1; /* this isolates the lowest bit */
302
303#if ECB_branchless_on_i386
304 r += !!(x & 0xaaaaaaaa) << 0;
305 r += !!(x & 0xcccccccc) << 1;
306 r += !!(x & 0xf0f0f0f0) << 2;
307 r += !!(x & 0xff00ff00) << 3;
308 r += !!(x & 0xffff0000) << 4;
309#else
310 if (x & 0xaaaaaaaa) r += 1;
311 if (x & 0xcccccccc) r += 2;
312 if (x & 0xf0f0f0f0) r += 4;
313 if (x & 0xff00ff00) r += 8;
314 if (x & 0xffff0000) r += 16;
315#endif
316
317 return r;
318 }
319
320 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
321 ecb_function_ int
322 ecb_ctz64 (uint64_t x)
323 {
324 int shift = x & 0xffffffffU ? 0 : 32;
325 return ecb_ctz32 (x >> shift) + shift;
326 }
327
328 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
329 ecb_function_ int
330 ecb_popcount32 (uint32_t x)
331 {
332 x -= (x >> 1) & 0x55555555;
333 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
334 x = ((x >> 4) + x) & 0x0f0f0f0f;
335 x *= 0x01010101;
336
337 return x >> 24;
338 }
339
340 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
341 ecb_function_ int ecb_ld32 (uint32_t x)
342 {
343 int r = 0;
344
345 if (x >> 16) { x >>= 16; r += 16; }
346 if (x >> 8) { x >>= 8; r += 8; }
347 if (x >> 4) { x >>= 4; r += 4; }
348 if (x >> 2) { x >>= 2; r += 2; }
349 if (x >> 1) { r += 1; }
350
351 return r;
352 }
353
354 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
355 ecb_function_ int ecb_ld64 (uint64_t x)
356 {
357 int r = 0;
358
359 if (x >> 32) { x >>= 32; r += 32; }
360
361 return r + ecb_ld32 (x);
362 }
363#endif
364
365ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
366ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
367ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
368ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
369
370ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
371ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
111{ 372{
112 int r = 0; 373 return ( (x * 0x0802U & 0x22110U)
113 374 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
114 x &= -x; /* this isolates the lowest bit */
115
116 if (x & 0xaaaaaaaa) r += 1;
117 if (x & 0xcccccccc) r += 2;
118 if (x & 0xf0f0f0f0) r += 4;
119 if (x & 0xff00ff00) r += 8;
120 if (x & 0xffff0000) r += 16;
121
122 return r;
123} 375}
124 376
125ECB_HEADER_INLINE int ecb_popcount32 (uint32_t x) ecb_const; 377ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
126ECB_HEADER_INLINE 378ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
127ecb_popcount32 (uint32_t x)
128{ 379{
129 x -= (x >> 1) & 0x55555555; 380 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
130 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 381 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
131 x = ((x >> 4) + x) & 0x0f0f0f0f; 382 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
132 x *= 0x01010101; 383 x = ( x >> 8 ) | ( x << 8);
133 384
134 return x >> 24; 385 return x;
135} 386}
136#endif 387
388ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
389ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
390{
391 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
392 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
393 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
394 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
395 x = ( x >> 16 ) | ( x << 16);
396
397 return x;
398}
399
400/* popcount64 is only available on 64 bit cpus as gcc builtin */
401/* so for this version we are lazy */
402ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
403ecb_function_ int
404ecb_popcount64 (uint64_t x)
405{
406 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
407}
408
409ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
410ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
411ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
412ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
413ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
414ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
415ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
416ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
417
418ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
419ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
420ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
421ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
422ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
423ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
424ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
425ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
137 426
138#if ECB_GCC_VERSION(4,3) 427#if ECB_GCC_VERSION(4,3)
139# define ecb_bswap32(x) __builtin_bswap32 (x)
140# define ecb_bswap16(x) (__builtin_bswap32(x) >> 16) 428 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
429 #define ecb_bswap32(x) __builtin_bswap32 (x)
430 #define ecb_bswap64(x) __builtin_bswap64 (x)
141#else 431#else
432 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
433 ecb_function_ uint16_t
434 ecb_bswap16 (uint16_t x)
435 {
436 return ecb_rotl16 (x, 8);
437 }
438
142ECB_HEADER_INLINE uint32_t ecb_bswap32 (uint32_t x) ecb_const; 439 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
143ECB_HEADER_INLINE uint32_t 440 ecb_function_ uint32_t
144ecb_bswap32 (uint32_t x) 441 ecb_bswap32 (uint32_t x)
145{ 442 {
146 return (x >> 24) 443 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
147 | ((x >> 8) & 0x0000ff00) 444 }
148 | ((x << 8) & 0x00ff0000)
149 | (x << 24);
150}
151 445
152ECB_HEADER_INLINE uint32_t ecb_bswap16 (uint32_t x) ecb_const; 446 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
153ECB_HEADER_INLINE uint32_t 447 ecb_function_ uint64_t
154ecb_bswap16 (uint32_t x) 448 ecb_bswap64 (uint64_t x)
155{ 449 {
156 return ((x >> 8) & 0xff) 450 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
157 | ((x << 8) & 0x00ff0000) 451 }
158 | (x << 24);
159}
160#endif 452#endif
161 453
162#if ECB_GCC_VERSION(4,5) 454#if ECB_GCC_VERSION(4,5)
163# define ecb_unreachable() __builtin_unreachable () 455 #define ecb_unreachable() __builtin_unreachable ()
164#else 456#else
165/* this seems to work fine, but gcc always emits a warning for it :/ */ 457 /* this seems to work fine, but gcc always emits a warning for it :/ */
166ECB_HEADER_INLINE void ecb_unreachable (void) ecb_noreturn; 458 ecb_inline void ecb_unreachable (void) ecb_noreturn;
167ECB_HEADER_INLINE void ecb_unreachable (void) { } 459 ecb_inline void ecb_unreachable (void) { }
168#endif 460#endif
169 461
462/* try to tell the compiler that some condition is definitely true */
463#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
464
170ECB_HEADER_INLINE unsigned char ecb_byteorder_helper (void) ecb_const; 465ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
171ECB_HEADER_INLINE unsigned char 466ecb_inline unsigned char
172ecb_byteorder_helper (void) 467ecb_byteorder_helper (void)
173{ 468{
174 const uint32_t u = 0x11223344; 469 /* the union code still generates code under pressure in gcc, */
175 return *(unsigned char *)&u; 470 /* but less than using pointers, and always seem to */
471 /* successfully return a constant. */
472 /* the reason why we have this horrible preprocessor mess */
473 /* is to avoid it in all cases, at least on common architectures */
474 /* and yes, gcc defines __BYTE_ORDER__, g++ does not */
475#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
476 return 0x44;
477#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
478 return 0x44;
479#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
480 retrurn 0x11;
481#else
482 union
483 {
484 uint32_t i;
485 uint8_t c;
486 } u = { 0x11223344 };
487 return u.c;
488#endif
176} 489}
177 490
178ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) ecb_const; 491ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
179ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }; 492ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
180ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) ecb_const; 493ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
181ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }; 494ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
182 495
496#if ECB_GCC_VERSION(3,0) || ECB_C99
497 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
498#else
183#define ecb_mod(m, n) ((m) < 0 ? ((n) - (-(m) % (n))) : ((m) % (n))) 499 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
500#endif
501
502#if __cplusplus
503 template<typename T>
504 static inline T ecb_div_rd (T val, T div)
505 {
506 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
507 }
508 template<typename T>
509 static inline T ecb_div_ru (T val, T div)
510 {
511 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
512 }
513#else
514 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
515 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
516#endif
184 517
185#if ecb_cplusplus_does_not_suck 518#if ecb_cplusplus_does_not_suck
186// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) 519 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
187template<typename T, int N> 520 template<typename T, int N>
188static inline int ecb_array_length (const T (&arr)[N]) 521 static inline int ecb_array_length (const T (&arr)[N])
189{ 522 {
190 return N; 523 return N;
191} 524 }
192#else 525#else
193#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 526 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
194#endif 527#endif
195 528
196ECB_INLINE uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
197ECB_INLINE uint32_t
198ecb_rotr32 (uint32_t x, unsigned int count)
199{
200 return (x << (32 - count)) | (x >> count);
201}
202
203ECB_INLINE uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
204ECB_INLINE uint32_t
205ecb_rotl32 (uint32_t x, unsigned int count)
206{
207 return (x >> (32 - count)) | (x << count);
208}
209
210#endif 529#endif
211 530

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines