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

Comparing libecb/ecb.h (file contents):
Revision 1.11 by root, Thu May 26 18:00:42 2011 UTC vs.
Revision 1.63 by root, Thu Aug 4 14:37:46 2011 UTC

1/* 1/*
2 * libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2011 Marc Alexander Lehmann 4 * Copyright (©) 2009-2011 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#ifdef _WIN32
34 typedef signed char int8_t;
35 typedef unsigned char uint8_t;
36 typedef signed short int16_t;
37 typedef unsigned short uint16_t;
38 typedef signed int int32_t;
39 typedef unsigned int uint32_t;
40 #if __GNUC__
41 typedef signed long long int64_t;
42 typedef unsigned long long uint64_t;
43 #else /* _MSC_VER || __BORLANDC__ */
44 typedef signed __int64 int64_t;
45 typedef unsigned __int64 uint64_t;
46 #endif
47#else
33#include <inttypes.h> 48 #include <inttypes.h>
49#endif
34 50
51/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions,
53 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so.
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.
57 */
58#ifndef ECB_GCC_VERSION
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
61 #else
35#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
64#endif
36 65
66/*****************************************************************************/
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 #define ECB_MEMORY_FENCE_ACQUIRE do { } while (0)
74 #define ECB_MEMORY_FENCE_RELEASE do { } while (0)
75#endif
76
77#ifndef ECB_MEMORY_FENCE
78 #if ECB_GCC_VERSION(2,5)
79 #if __x86
80 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
81 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */
82 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
83 #elif __amd64
84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
85 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")
86 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
87 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
88 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
89 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \
90 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) \
91 || defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \
92 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ )
93 #define ECB_MEMORY_FENCE \
94 do { \
95 int null = 0; \
96 __asm__ __volatile__ ("mcr p15,0,%0,c6,c10,5", : "=&r" (null) : : "memory"); \
97 while (0)
98 #endif
99 #endif
100#endif
101
102#ifndef ECB_MEMORY_FENCE
103 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER)
104 #define ECB_MEMORY_FENCE __sync_synchronize ()
105 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
106 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
107 #elif _MSC_VER >= 1400 /* VC++ 2005 */
108 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
109 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
110 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
111 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
112 #elif defined(_WIN32)
113 #include <WinNT.h>
114 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
115 #endif
116#endif
117
118#ifndef ECB_MEMORY_FENCE
119 #if !ECB_AVOID_PTHREADS
120 /*
121 * if you get undefined symbol references to pthread_mutex_lock,
122 * or failure to find pthread.h, then you should implement
123 * the ECB_MEMORY_FENCE operations for your cpu/compiler
124 * OR provide pthread.h and link against the posix thread library
125 * of your system.
126 */
127 #include <pthread.h>
128 #define ECB_NEEDS_PTHREADS 1
129 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
130
131 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
132 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
133 #endif
134#endif
135
136#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE)
137 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
138#endif
139
140#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE)
141 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
142#endif
143
144/*****************************************************************************/
145
146#define ECB_C99 (__STDC_VERSION__ >= 199901L)
147
37#ifndef __cplusplus 148#if __cplusplus
38# if __STDC_VERSION__ >= 199901L 149 #define ecb_inline static inline
39# define ECB_INLINE inline 150#elif ECB_GCC_VERSION(2,5)
40typedef _Bool ecb_bool; 151 #define ecb_inline static __inline__
152#elif ECB_C99
153 #define ecb_inline static inline
41# else 154#else
42# define ECB_INLINE inline /* yeah! */ 155 #define ecb_inline static
156#endif
157
158#if ECB_GCC_VERSION(3,3)
159 #define ecb_restrict __restrict__
160#elif ECB_C99
161 #define ecb_restrict restrict
162#else
163 #define ecb_restrict
164#endif
165
43typedef int ecb_bool; 166typedef int ecb_bool;
44# endif
45#endif
46 167
47#define ECB_CONCAT_(a, b) a ## b 168#define ECB_CONCAT_(a, b) a ## b
48#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 169#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
49#define ECB_STRINGIFY_(a) # a 170#define ECB_STRINGIFY_(a) # a
50#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 171#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
51 172
52#define ECB_HEADER_INLINE static ECB_INLINE 173#define ecb_function_ ecb_inline
53 174
54#if ECB_GCC_VERSION(3,1) 175#if ECB_GCC_VERSION(3,1)
55# define ecb_attribute(attrlist) __attribute__(attrlist) 176 #define ecb_attribute(attrlist) __attribute__(attrlist)
56# define ecb_is_constant(c) __builtin_constant_p (c) 177 #define ecb_is_constant(expr) __builtin_constant_p (expr)
57# define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 178 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
58# define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 179 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
59#else 180#else
60# define ecb_attribute(attrlist) 181 #define ecb_attribute(attrlist)
61# define ecb_is_constant(c) 0 182 #define ecb_is_constant(expr) 0
62# define ecb_expect(expr,value) (expr) 183 #define ecb_expect(expr,value) (expr)
63# define ecb_prefetch(addr,rw,locality) 184 #define ecb_prefetch(addr,rw,locality)
64#endif 185#endif
65 186
66/* no emulation for ecb_decltype */ 187/* no emulation for ecb_decltype */
67#if ECB_GCC_VERSION(4,5) 188#if ECB_GCC_VERSION(4,5)
68# define ecb_decltype(x) __decltype(x) 189 #define ecb_decltype(x) __decltype(x)
69#elif ECB_GCC_VERSION(3,0) 190#elif ECB_GCC_VERSION(3,0)
70# define ecb_decltype(x) typeof(x) 191 #define ecb_decltype(x) __typeof(x)
71#endif 192#endif
72 193
73#define ecb_noinline ecb_attribute ((noinline)) 194#define ecb_noinline ecb_attribute ((__noinline__))
74#define ecb_noreturn ecb_attribute ((noreturn)) 195#define ecb_noreturn ecb_attribute ((__noreturn__))
75#define ecb_unused ecb_attribute ((unused)) 196#define ecb_unused ecb_attribute ((__unused__))
76#define ecb_const ecb_attribute ((const)) 197#define ecb_const ecb_attribute ((__const__))
77#define ecb_pure ecb_attribute ((pure)) 198#define ecb_pure ecb_attribute ((__pure__))
78#define ecb_hot ecb_attribute ((hot)) /* 4.3 */
79#define ecb_cold ecb_attribute ((cold)) /* 4.3 */
80 199
81#if ECB_GCC_VERSION(4,5) 200#if ECB_GCC_VERSION(4,3)
82# define ecb_unreachable() __builtin_unreachable () 201 #define ecb_artificial ecb_attribute ((__artificial__))
202 #define ecb_hot ecb_attribute ((__hot__))
203 #define ecb_cold ecb_attribute ((__cold__))
83#else 204#else
84/* this seems to work fine, but gcc always emits a warning for it :/ */ 205 #define ecb_artificial
85ECB_HEADER_INLINE void ecb_unreachable () ecb_attribute ((noreturn)); 206 #define ecb_hot
86ECB_HEADER_INLINE void ecb_unreachable () { } 207 #define ecb_cold
87#endif 208#endif
88 209
89/* put into ifs if you are very sure that the expression */ 210/* put around conditional expressions if you are very sure that the */
90/* is mostly true or mosty false. note that these return */ 211/* expression is mostly true or mostly false. note that these return */
91/* booleans, not the expression. */ 212/* booleans, not the expression. */
92#define ecb_unlikely(expr) ecb_expect (!!(expr), 0) 213#define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
214#define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
215/* for compatibility to the rest of the world */
93#define ecb_likely(expr) ecb_expect (!!(expr), 1) 216#define ecb_likely(expr) ecb_expect_true (expr)
94 217#define ecb_unlikely(expr) ecb_expect_false (expr)
95/* try to tell the compiler that some condition is definitely true */
96#define ecb_assume(cond) do { if (!(cond)) unreachable (); } while (0)
97 218
98/* count trailing zero bits and count # of one bits */ 219/* count trailing zero bits and count # of one bits */
99#if ECB_GCC_VERSION(3,4) 220#if ECB_GCC_VERSION(3,4)
100ECB_HEADER_INLINE int ecb_ctz32 (uint32_t x) { return __builtin_ctz (x); } 221 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
101ECB_HEADER_INLINE int ecb_popcount32 (uint32_t x) { return __builtin_popcount (x); } 222 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
223 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
224 #define ecb_ctz32(x) __builtin_ctz (x)
225 #define ecb_ctz64(x) __builtin_ctzll (x)
226 #define ecb_popcount32(x) __builtin_popcount (x)
227 /* no popcountll */
102#else 228#else
103ECB_HEADER_INLINE 229 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
230 ecb_function_ int
104ecb_ctz32 (uint32_t x) ecb_const 231 ecb_ctz32 (uint32_t x)
232 {
233 int r = 0;
234
235 x &= ~x + 1; /* this isolates the lowest bit */
236
237#if ECB_branchless_on_i386
238 r += !!(x & 0xaaaaaaaa) << 0;
239 r += !!(x & 0xcccccccc) << 1;
240 r += !!(x & 0xf0f0f0f0) << 2;
241 r += !!(x & 0xff00ff00) << 3;
242 r += !!(x & 0xffff0000) << 4;
243#else
244 if (x & 0xaaaaaaaa) r += 1;
245 if (x & 0xcccccccc) r += 2;
246 if (x & 0xf0f0f0f0) r += 4;
247 if (x & 0xff00ff00) r += 8;
248 if (x & 0xffff0000) r += 16;
249#endif
250
251 return r;
252 }
253
254 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
255 ecb_function_ int
256 ecb_ctz64 (uint64_t x)
257 {
258 int shift = x & 0xffffffffU ? 0 : 32;
259 return ecb_ctz32 (x >> shift) + shift;
260 }
261
262 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
263 ecb_function_ int
264 ecb_popcount32 (uint32_t x)
265 {
266 x -= (x >> 1) & 0x55555555;
267 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
268 x = ((x >> 4) + x) & 0x0f0f0f0f;
269 x *= 0x01010101;
270
271 return x >> 24;
272 }
273
274 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
275 ecb_function_ int ecb_ld32 (uint32_t x)
276 {
277 int r = 0;
278
279 if (x >> 16) { x >>= 16; r += 16; }
280 if (x >> 8) { x >>= 8; r += 8; }
281 if (x >> 4) { x >>= 4; r += 4; }
282 if (x >> 2) { x >>= 2; r += 2; }
283 if (x >> 1) { r += 1; }
284
285 return r;
286 }
287
288 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
289 ecb_function_ int ecb_ld64 (uint64_t x)
290 {
291 int r = 0;
292
293 if (x >> 32) { x >>= 32; r += 32; }
294
295 return r + ecb_ld32 (x);
296 }
297#endif
298
299/* popcount64 is only available on 64 bit cpus as gcc builtin */
300/* so for this version we are lazy */
301ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
302ecb_function_ int
303ecb_popcount64 (uint64_t x)
105{ 304{
106 int r = 0; 305 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
107
108 x &= -x; /* this isolates the lowest bit */
109
110 if (x & 0xaaaaaaaa) r += 1;
111 if (x & 0xcccccccc) r += 2;
112 if (x & 0xf0f0f0f0) r += 4;
113 if (x & 0xff00ff00) r += 8;
114 if (x & 0xffff0000) r += 16;
115
116 return r;
117} 306}
118 307
119ECB_HEADER_INLINE 308ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
120ecb_popcount32 (uint32_t x) ecb_const 309ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
121{ 310ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
122 x -= (x >> 1) & 0x55555555; 311ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
123 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 312ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
124 x = ((x >> 4) + x) & 0x0f0f0f0f; 313ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
125 x *= 0x01010101; 314ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
315ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
126 316
127 return x >> 24; 317ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
128} 318ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
129#endif 319ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
320ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
321ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
322ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
323ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
324ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
130 325
131ECB_HEADER_INLINE unsigned char 326#if ECB_GCC_VERSION(4,3)
327 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
328 #define ecb_bswap32(x) __builtin_bswap32 (x)
329 #define ecb_bswap64(x) __builtin_bswap64 (x)
330#else
331 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
332 ecb_function_ uint16_t
333 ecb_bswap16 (uint16_t x)
334 {
335 return ecb_rotl16 (x, 8);
336 }
337
338 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
339 ecb_function_ uint32_t
340 ecb_bswap32 (uint32_t x)
341 {
342 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
343 }
344
345 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
346 ecb_function_ uint64_t
347 ecb_bswap64 (uint64_t x)
348 {
349 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
350 }
351#endif
352
353#if ECB_GCC_VERSION(4,5)
354 #define ecb_unreachable() __builtin_unreachable ()
355#else
356 /* this seems to work fine, but gcc always emits a warning for it :/ */
357 ecb_function_ void ecb_unreachable (void) ecb_noreturn;
358 ecb_function_ void ecb_unreachable (void) { }
359#endif
360
361/* try to tell the compiler that some condition is definitely true */
362#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
363
364ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const;
365ecb_function_ unsigned char
132ecb_byteorder_helper () ecb_const 366ecb_byteorder_helper (void)
133{ 367{
134 const uint32_t u = 0x11223344; 368 const uint32_t u = 0x11223344;
135 return *(unsigned char *)&u; 369 return *(unsigned char *)&u;
136} 370}
137 371
372ecb_function_ ecb_bool ecb_big_endian (void) ecb_const;
138ECB_HEADER_INLINE ecb_bool ecb_big_endian () ecb_const { return ecb_byteorder_helper () == 0x11; }; 373ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
374ecb_function_ ecb_bool ecb_little_endian (void) ecb_const;
139ECB_HEADER_INLINE ecb_bool ecb_little_endian () ecb_const { return ecb_byteorder_helper () == 0x44; }; 375ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
376
377#if ECB_GCC_VERSION(3,0) || ECB_C99
378 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
379#else
380 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
381#endif
140 382
141#if ecb_cplusplus_does_not_suck 383#if ecb_cplusplus_does_not_suck
142// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) 384 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
143template<typename T, int N> 385 template<typename T, int N>
144static inline int ecb_array_length (const T (&arr)[N]) 386 static inline int ecb_array_length (const T (&arr)[N])
145{ 387 {
146 return N; 388 return N;
147} 389 }
148#else 390#else
149#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 391 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
150#endif 392#endif
151 393
152ECB_INLINE uint32_t
153ecb_rotr32 (uint32_t x, unsigned int count)
154{
155 return (x << (32 - count)) | (x >> count);
156}
157
158ECB_INLINE uint32_t
159ecb_rotl32 (uint32_t x, unsigned int count)
160{
161 return (x >> (32 - count)) | (x << count);
162}
163
164#if ECB_GCC_VERSION(4,3)
165# define ecb_bswap32(x) __builtin_bswap32 (x)
166#else
167ECB_GCC_VERSION uint32_t
168ecb_bswap32 (uint32_t x)
169{
170 return (x >> 24)
171 | ((x >> 8) & 0x0000ff00)
172 | ((x << 8) & 0x00ff0000)
173 | (x << 24);
174}
175#endif 394#endif
176 395
177#endif
178

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines