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

Comparing libecb/ecb.h (file contents):
Revision 1.2 by root, Thu May 26 16:46:48 2011 UTC vs.
Revision 1.62 by root, Thu Aug 4 14:34:00 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 * All rights reserved. 6 * All rights reserved.
6 * 7 *
7 * 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-
8 * tion, are permitted provided that the following conditions are met: 9 * tion, are permitted provided that the following conditions are met:
9 * 10 *
… …
27 */ 28 */
28 29
29#ifndef ECB_H 30#ifndef ECB_H
30#define ECB_H 31#define ECB_H
31 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
48 #include <inttypes.h>
49#endif
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
32#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
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 defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \
88 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) \
89 || defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \
90 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ )
91 #define ECB_MEMORY_FENCE \
92 do { \
93 int null = 0; \
94 __asm__ __volatile__ ("mcr p15,0,%0,c6,c10,5", : "=&r" (null) : : "memory"); \
95 while (0)
96 #endif
97 #endif
98#endif
99
100#ifndef ECB_MEMORY_FENCE
101 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER)
102 #define ECB_MEMORY_FENCE __sync_synchronize ()
103 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
104 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
105 #elif _MSC_VER >= 1400 /* VC++ 2005 */
106 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
107 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
108 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
109 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
110 #elif defined(_WIN32)
111 #include <WinNT.h>
112 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
113 #endif
114#endif
115
116#ifndef ECB_MEMORY_FENCE
117 #if !ECB_AVOID_PTHREADS
118 /*
119 * if you get undefined symbol references to pthread_mutex_lock,
120 * or failure to find pthread.h, then you should implement
121 * the ECB_MEMORY_FENCE operations for your cpu/compiler
122 * OR provide pthread.h and link against the posix thread library
123 * of your system.
124 */
125 #include <pthread.h>
126 #define ECB_NEEDS_PTHREADS 1
127 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
128
129 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
130 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
131 #endif
132#endif
133
134#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE)
135 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
136#endif
137
138#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE)
139 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
140#endif
141
142/*****************************************************************************/
143
144#define ECB_C99 (__STDC_VERSION__ >= 199901L)
145
146#if __cplusplus
147 #define ecb_inline static inline
148#elif ECB_GCC_VERSION(2,5)
149 #define ecb_inline static __inline__
150#elif ECB_C99
151 #define ecb_inline static inline
152#else
153 #define ecb_inline static
154#endif
155
156#if ECB_GCC_VERSION(3,3)
157 #define ecb_restrict __restrict__
158#elif ECB_C99
159 #define ecb_restrict restrict
160#else
161 #define ecb_restrict
162#endif
163
164typedef int ecb_bool;
165
166#define ECB_CONCAT_(a, b) a ## b
167#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
168#define ECB_STRINGIFY_(a) # a
169#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
170
171#define ecb_function_ ecb_inline
33 172
34#if ECB_GCC_VERSION(3,1) 173#if ECB_GCC_VERSION(3,1)
35# define ecb_attribute(attrlist) __attribute__(attrlist) 174 #define ecb_attribute(attrlist) __attribute__(attrlist)
36# define ecb_is_constant(c) __builtin_constant_p (c) 175 #define ecb_is_constant(expr) __builtin_constant_p (expr)
37# define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 176 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
38# define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 177 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
39#else 178#else
40# define ecb_attribute(attrlist) 179 #define ecb_attribute(attrlist)
41# define ecb_is_constant(c) 0 180 #define ecb_is_constant(expr) 0
42# define ecb_expect(expr,value) (expr) 181 #define ecb_expect(expr,value) (expr)
43# define ecb_prefetch(addr,rw,locality) 182 #define ecb_prefetch(addr,rw,locality)
44#endif 183#endif
45 184
46/* no emulation for ecb_decltype */ 185/* no emulation for ecb_decltype */
47#if ECB_GCC_VERSION(4,5) 186#if ECB_GCC_VERSION(4,5)
48# define ecb_decltype(x) __decltype(x) 187 #define ecb_decltype(x) __decltype(x)
49#elif ECB_GCC_VERSION(3,0) 188#elif ECB_GCC_VERSION(3,0)
50# define ecb_decltype(x) typeof(x) 189 #define ecb_decltype(x) __typeof(x)
51#endif 190#endif
52 191
53#define ecb_noinline ecb_attribute ((noinline)) 192#define ecb_noinline ecb_attribute ((__noinline__))
54#define ecb_noreturn ecb_attribute ((noreturn)) 193#define ecb_noreturn ecb_attribute ((__noreturn__))
55#define ecb_unused ecb_attribute ((unused)) 194#define ecb_unused ecb_attribute ((__unused__))
56#define ecb_const ecb_attribute ((const)) 195#define ecb_const ecb_attribute ((__const__))
57#define ecb_pure ecb_attribute ((pure)) 196#define ecb_pure ecb_attribute ((__pure__))
197
198#if ECB_GCC_VERSION(4,3)
199 #define ecb_artificial ecb_attribute ((__artificial__))
58#define ecb_hot ecb_attribute ((hot)) // 4.3 200 #define ecb_hot ecb_attribute ((__hot__))
59#define ecb_cold ecb_attribute ((cold)) // 4.3 201 #define ecb_cold ecb_attribute ((__cold__))
202#else
203 #define ecb_artificial
204 #define ecb_hot
205 #define ecb_cold
206#endif
207
208/* put around conditional expressions if you are very sure that the */
209/* expression is mostly true or mostly false. note that these return */
210/* booleans, not the expression. */
211#define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
212#define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
213/* for compatibility to the rest of the world */
214#define ecb_likely(expr) ecb_expect_true (expr)
215#define ecb_unlikely(expr) ecb_expect_false (expr)
216
217/* count trailing zero bits and count # of one bits */
218#if ECB_GCC_VERSION(3,4)
219 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
220 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
221 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
222 #define ecb_ctz32(x) __builtin_ctz (x)
223 #define ecb_ctz64(x) __builtin_ctzll (x)
224 #define ecb_popcount32(x) __builtin_popcount (x)
225 /* no popcountll */
226#else
227 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
228 ecb_function_ int
229 ecb_ctz32 (uint32_t x)
230 {
231 int r = 0;
232
233 x &= ~x + 1; /* this isolates the lowest bit */
234
235#if ECB_branchless_on_i386
236 r += !!(x & 0xaaaaaaaa) << 0;
237 r += !!(x & 0xcccccccc) << 1;
238 r += !!(x & 0xf0f0f0f0) << 2;
239 r += !!(x & 0xff00ff00) << 3;
240 r += !!(x & 0xffff0000) << 4;
241#else
242 if (x & 0xaaaaaaaa) r += 1;
243 if (x & 0xcccccccc) r += 2;
244 if (x & 0xf0f0f0f0) r += 4;
245 if (x & 0xff00ff00) r += 8;
246 if (x & 0xffff0000) r += 16;
247#endif
248
249 return r;
250 }
251
252 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
253 ecb_function_ int
254 ecb_ctz64 (uint64_t x)
255 {
256 int shift = x & 0xffffffffU ? 0 : 32;
257 return ecb_ctz32 (x >> shift) + shift;
258 }
259
260 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
261 ecb_function_ int
262 ecb_popcount32 (uint32_t x)
263 {
264 x -= (x >> 1) & 0x55555555;
265 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
266 x = ((x >> 4) + x) & 0x0f0f0f0f;
267 x *= 0x01010101;
268
269 return x >> 24;
270 }
271
272 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
273 ecb_function_ int ecb_ld32 (uint32_t x)
274 {
275 int r = 0;
276
277 if (x >> 16) { x >>= 16; r += 16; }
278 if (x >> 8) { x >>= 8; r += 8; }
279 if (x >> 4) { x >>= 4; r += 4; }
280 if (x >> 2) { x >>= 2; r += 2; }
281 if (x >> 1) { r += 1; }
282
283 return r;
284 }
285
286 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
287 ecb_function_ int ecb_ld64 (uint64_t x)
288 {
289 int r = 0;
290
291 if (x >> 32) { x >>= 32; r += 32; }
292
293 return r + ecb_ld32 (x);
294 }
295#endif
296
297/* popcount64 is only available on 64 bit cpus as gcc builtin */
298/* so for this version we are lazy */
299ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
300ecb_function_ int
301ecb_popcount64 (uint64_t x)
302{
303 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
304}
305
306ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
307ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
308ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
309ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
310ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
311ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
312ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
313ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
314
315ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
316ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
317ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
318ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
319ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
320ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
321ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
322ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
323
324#if ECB_GCC_VERSION(4,3)
325 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
326 #define ecb_bswap32(x) __builtin_bswap32 (x)
327 #define ecb_bswap64(x) __builtin_bswap64 (x)
328#else
329 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
330 ecb_function_ uint16_t
331 ecb_bswap16 (uint16_t x)
332 {
333 return ecb_rotl16 (x, 8);
334 }
335
336 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
337 ecb_function_ uint32_t
338 ecb_bswap32 (uint32_t x)
339 {
340 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
341 }
342
343 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
344 ecb_function_ uint64_t
345 ecb_bswap64 (uint64_t x)
346 {
347 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
348 }
349#endif
60 350
61#if ECB_GCC_VERSION(4,5) 351#if ECB_GCC_VERSION(4,5)
62# define ecb_unreachable() __builtin_unreachable () 352 #define ecb_unreachable() __builtin_unreachable ()
63#else 353#else
64 // this seems to work fine, but gcc always emits a warning for it :/ 354 /* this seems to work fine, but gcc always emits a warning for it :/ */
65 static inline void ecb_unreachable () ecb_attribute ((noreturn)); 355 ecb_function_ void ecb_unreachable (void) ecb_noreturn;
66 static inline void ecb_unreachable () { } 356 ecb_function_ void ecb_unreachable (void) { }
67#endif 357#endif
68 358
69// put into ifs if you are very sure that the expression
70// is mostly true or mosty false. note that these return
71// booleans, not the expression.
72#define ecb_expect_false(expr) ecb_expect ((expr) ? 1 : 0, 0)
73#define ecb_expect_true(expr) ecb_expect ((expr) ? 1 : 0, 1)
74
75// try to tell the compiler that some condition is definitely true 359/* try to tell the compiler that some condition is definitely true */
76#define ecb_assume(cond) do { if (!(cond)) unreachable (); } while (0) 360#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
77 361
78// count trailing zero bits and count # of one bits 362ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const;
79#if ECB_GCC_VERSION(3,4) 363ecb_function_ unsigned char
80static inline int ecb_ctz (unsigned int x) { return __builtin_ctz (x); } 364ecb_byteorder_helper (void)
81static inline int ecb_popcount (unsigned int x) { return __builtin_popcount (x); }
82#else
83static int ecb_ctz (unsigned int x) ecb_const
84{ 365{
85 int r = 0; 366 const uint32_t u = 0x11223344;
86 367 return *(unsigned char *)&u;
87 x &= -x; // this isolates the lowest bit
88
89 if (x & 0xaaaaaaaa) r += 1;
90 if (x & 0xcccccccc) r += 2;
91 if (x & 0xf0f0f0f0) r += 4;
92 if (x & 0xff00ff00) r += 8;
93 if (x & 0xffff0000) r += 16;
94
95 return r;
96} 368}
97 369
98static int ecb_popcount (unsigned int x) ecb_const 370ecb_function_ ecb_bool ecb_big_endian (void) ecb_const;
99{ 371ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
100 x -= (x >> 1) & 0x55555555; 372ecb_function_ ecb_bool ecb_little_endian (void) ecb_const;
101 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 373ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
102 x = ((x >> 4) + x) & 0x0f0f0f0f;
103 x *= 0x01010101;
104 374
105 return x >> 24; 375#if ECB_GCC_VERSION(3,0) || ECB_C99
106} 376 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
377#else
378 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
107#endif 379#endif
108 380
381#if ecb_cplusplus_does_not_suck
382 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
383 template<typename T, int N>
384 static inline int ecb_array_length (const T (&arr)[N])
385 {
386 return N;
387 }
388#else
389 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
109#endif 390#endif
110 391
392#endif
393

Diff Legend

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