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

Comparing libecb/ecb.h (file contents):
Revision 1.37 by root, Sat Jun 11 13:47:47 2011 UTC vs.
Revision 1.56 by sf-exg, Thu Jul 28 13:45:44 2011 UTC

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
35/* many compilers define _GNUC_ to some versions but then only implement 51/* many compilers define _GNUC_ to some versions but then only implement
36 * what their idiot authors think are the "more important" extensions, 52 * what their idiot authors think are the "more important" extensions,
37 * causing enourmous grief in return for some better fake benchmark numbers. 53 * causing enourmous grief in return for some better fake benchmark numbers.
38 * or so. 54 * or so.
39 * we try to detect these and simply assume they are not gcc - if they have 55 * 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. 56 * an issue with that they should have done it right in the first place.
41 */ 57 */
42#ifndef ECB_GCC_VERSION 58#ifndef ECB_GCC_VERSION
43 #if defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__llvm__) 59 #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 60 #define ECB_GCC_VERSION(major,minor) 0
45 #else 61 #else
46 #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)))
47 #endif 63 #endif
48#endif 64#endif
49 65
50#ifndef __cplusplus 66/*****************************************************************************/
51 #if __STDC_VERSION__ >= 199901L 67
52 #define ECB_INLINE static inline 68#ifndef ECB_MEMORY_FENCE
53 typedef _Bool ecb_bool; 69 #if ECB_GCC_VERSION(2,5)
54 #else 70 #if __x86
55 #define ECB_INLINE static inline /* we assume the extension is ubiquituous, please tell us when we are wrong, or upgrade to the GCC */ 71 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
56 typedef int ecb_bool; 72 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
73 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE /* better be safe than sorry */
74 #elif __amd64
75 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
76 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")
77 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence")
78 #endif
57 #endif 79 #endif
58#else 80#endif
59 #define ECB_INLINE static inline 81
60 typedef bool ecb_bool; 82#ifndef ECB_MEMORY_FENCE
83 #if ECB_GCC_VERSION(4,4)
84 #define ECB_MEMORY_FENCE __sync_synchronize ()
85 #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); })
86 #define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); })
87 #elif _MSC_VER >= 1400 && 0 /* TODO: only true when using volatiles */
88 #define ECB_MEMORY_FENCE do { } while (0)
89 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
90 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
91 #elif defined(_WIN32)
92 #include <WinNT.h>
93 #define ECB_MEMORY_FENCE MemoryBarrier ()
94 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
95 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
61#endif 96 #endif
97#endif
98
99#ifndef ECB_MEMORY_FENCE
100 #include <pthread.h>
101
102 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
103 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
104 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
105 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
106#endif
107
108/*****************************************************************************/
109
110#define ECB_C99 (__STDC_VERSION__ >= 199901L)
111
112#if __cplusplus
113 #define ecb_inline static inline
114#elif ECB_GCC_VERSION(2,5)
115 #define ecb_inline static __inline__
116#elif ECB_C99
117 #define ecb_inline static inline
118#else
119 #define ecb_inline static
120#endif
121
122#if ECB_GCC_VERSION(3,3)
123 #define ecb_restrict __restrict__
124#elif ECB_C99
125 #define ecb_restrict restrict
126#else
127 #define ecb_restrict
128#endif
129
130typedef int ecb_bool;
62 131
63#define ECB_CONCAT_(a, b) a ## b 132#define ECB_CONCAT_(a, b) a ## b
64#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 133#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
65#define ECB_STRINGIFY_(a) # a 134#define ECB_STRINGIFY_(a) # a
66#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 135#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
67 136
68#define ECB_HEADER_INLINE ECB_INLINE 137#define ecb_function_ ecb_inline
69 138
70#if ECB_GCC_VERSION(3,1) 139#if ECB_GCC_VERSION(3,1)
71 #define ecb_attribute(attrlist) __attribute__(attrlist) 140 #define ecb_attribute(attrlist) __attribute__(attrlist)
72 #define ecb_is_constant(expr) __builtin_constant_p (expr) 141 #define ecb_is_constant(expr) __builtin_constant_p (expr)
73 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 142 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
81 150
82/* no emulation for ecb_decltype */ 151/* no emulation for ecb_decltype */
83#if ECB_GCC_VERSION(4,5) 152#if ECB_GCC_VERSION(4,5)
84 #define ecb_decltype(x) __decltype(x) 153 #define ecb_decltype(x) __decltype(x)
85#elif ECB_GCC_VERSION(3,0) 154#elif ECB_GCC_VERSION(3,0)
86 #define ecb_decltype(x) typeof(x) 155 #define ecb_decltype(x) __typeof(x)
87#endif 156#endif
88 157
89#define ecb_noinline ecb_attribute ((__noinline__)) 158#define ecb_noinline ecb_attribute ((__noinline__))
90#define ecb_noreturn ecb_attribute ((__noreturn__)) 159#define ecb_noreturn ecb_attribute ((__noreturn__))
91#define ecb_unused ecb_attribute ((__unused__)) 160#define ecb_unused ecb_attribute ((__unused__))
92#define ecb_const ecb_attribute ((__const__)) 161#define ecb_const ecb_attribute ((__const__))
93#define ecb_pure ecb_attribute ((__pure__)) 162#define ecb_pure ecb_attribute ((__pure__))
94 163
95#if ECB_GCC_VERSION(4,3) 164#if ECB_GCC_VERSION(4,3)
96 #define ecb_artificial ecb_attribute ((__artificial__)) /* 4.3 */ 165 #define ecb_artificial ecb_attribute ((__artificial__))
97 #define ecb_hot ecb_attribute ((__hot__)) /* 4.3 */ 166 #define ecb_hot ecb_attribute ((__hot__))
98 #define ecb_cold ecb_attribute ((__cold__)) /* 4.3 */ 167 #define ecb_cold ecb_attribute ((__cold__))
99#else 168#else
100 #define ecb_artificial 169 #define ecb_artificial
101 #define ecb_hot 170 #define ecb_hot
102 #define ecb_cold 171 #define ecb_cold
103#endif 172#endif
104 173
105/* put into if's if you are very sure that the expression */ 174/* put around conditional expressions if you are very sure that the */
106/* is mostly true or mostly false. note that these return */ 175/* expression is mostly true or mostly false. note that these return */
107/* booleans, not the expression. */ 176/* booleans, not the expression. */
108#define ecb_expect_false(expr) ecb_expect (!!(expr), 0) 177#define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
109#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) 178#define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
110/* for compatibility to the rest of the world */ 179/* for compatibility to the rest of the world */
111#define ecb_likely(expr) ecb_expect_true (expr) 180#define ecb_likely(expr) ecb_expect_true (expr)
112#define ecb_unlikely(expr) ecb_expect_false (expr) 181#define ecb_unlikely(expr) ecb_expect_false (expr)
113 182
114/* try to tell the compiler that some condition is definitely true */
115#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
116
117/* count trailing zero bits and count # of one bits */ 183/* count trailing zero bits and count # of one bits */
118#if ECB_GCC_VERSION(3,4) 184#if ECB_GCC_VERSION(3,4)
185 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
186 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
187 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
119 #define ecb_ctz32(x) __builtin_ctz (x) 188 #define ecb_ctz32(x) __builtin_ctz (x)
189 #define ecb_ctz64(x) __builtin_ctzll (x)
120 #define ecb_popcount32(x) __builtin_popcount (x) 190 #define ecb_popcount32(x) __builtin_popcount (x)
191 /* no popcountll */
121#else 192#else
122 ECB_HEADER_INLINE int ecb_ctz32 (uint32_t x) ecb_const; 193 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
123 ECB_HEADER_INLINE int 194 ecb_function_ int
124 ecb_ctz32 (uint32_t x) 195 ecb_ctz32 (uint32_t x)
125 { 196 {
126 int r = 0; 197 int r = 0;
127 198
128 x &= -x; /* this isolates the lowest bit */ 199 x &= ~x + 1; /* this isolates the lowest bit */
129 200
201#if ECB_branchless_on_i386
202 r += !!(x & 0xaaaaaaaa) << 0;
203 r += !!(x & 0xcccccccc) << 1;
204 r += !!(x & 0xf0f0f0f0) << 2;
205 r += !!(x & 0xff00ff00) << 3;
206 r += !!(x & 0xffff0000) << 4;
207#else
130 if (x & 0xaaaaaaaa) r += 1; 208 if (x & 0xaaaaaaaa) r += 1;
131 if (x & 0xcccccccc) r += 2; 209 if (x & 0xcccccccc) r += 2;
132 if (x & 0xf0f0f0f0) r += 4; 210 if (x & 0xf0f0f0f0) r += 4;
133 if (x & 0xff00ff00) r += 8; 211 if (x & 0xff00ff00) r += 8;
134 if (x & 0xffff0000) r += 16; 212 if (x & 0xffff0000) r += 16;
213#endif
135 214
136 return r; 215 return r;
137 } 216 }
138 217
218 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
219 ecb_function_ int
220 ecb_ctz64 (uint64_t x)
221 {
222 int shift = x & 0xffffffffU ? 0 : 32;
223 return ecb_ctz32 (x >> shift) + shift;
224 }
225
139 ECB_HEADER_INLINE int ecb_popcount32 (uint32_t x) ecb_const; 226 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
140 ECB_HEADER_INLINE int 227 ecb_function_ int
141 ecb_popcount32 (uint32_t x) 228 ecb_popcount32 (uint32_t x)
142 { 229 {
143 x -= (x >> 1) & 0x55555555; 230 x -= (x >> 1) & 0x55555555;
144 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 231 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
145 x = ((x >> 4) + x) & 0x0f0f0f0f; 232 x = ((x >> 4) + x) & 0x0f0f0f0f;
146 x *= 0x01010101; 233 x *= 0x01010101;
147 234
148 return x >> 24; 235 return x >> 24;
149 } 236 }
237
238 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
239 ecb_function_ int ecb_ld32 (uint32_t x)
240 {
241 int r = 0;
242
243 if (x >> 16) { x >>= 16; r += 16; }
244 if (x >> 8) { x >>= 8; r += 8; }
245 if (x >> 4) { x >>= 4; r += 4; }
246 if (x >> 2) { x >>= 2; r += 2; }
247 if (x >> 1) { r += 1; }
248
249 return r;
250 }
251
252 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
253 ecb_function_ int ecb_ld64 (uint64_t x)
254 {
255 int r = 0;
256
257 if (x >> 32) { x >>= 32; r += 32; }
258
259 return r + ecb_ld32 (x);
260 }
150#endif 261#endif
262
263/* popcount64 is only available on 64 bit cpus as gcc builtin */
264/* so for this version we are lazy */
265ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
266ecb_function_ int
267ecb_popcount64 (uint64_t x)
268{
269 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
270}
271
272ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
273ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
274ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
275ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
276ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
277ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
278ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
279ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
280
281ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
282ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
283ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
284ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
285ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
286ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
287ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
288ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
151 289
152#if ECB_GCC_VERSION(4,3) 290#if ECB_GCC_VERSION(4,3)
153 #define ecb_bswap32(x) __builtin_bswap32 (x)
154 #define ecb_bswap16(x) (__builtin_bswap32(x) >> 16) 291 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
292 #define ecb_bswap32(x) __builtin_bswap32 (x)
293 #define ecb_bswap64(x) __builtin_bswap64 (x)
155#else 294#else
295 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
296 ecb_function_ uint16_t
297 ecb_bswap16 (uint16_t x)
298 {
299 return ecb_rotl16 (x, 8);
300 }
301
156 ECB_HEADER_INLINE uint32_t ecb_bswap32 (uint32_t x) ecb_const; 302 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
157 ECB_HEADER_INLINE uint32_t 303 ecb_function_ uint32_t
158 ecb_bswap32 (uint32_t x) 304 ecb_bswap32 (uint32_t x)
159 { 305 {
160 return (x >> 24) 306 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
161 | ((x >> 8) & 0x0000ff00)
162 | ((x << 8) & 0x00ff0000)
163 | (x << 24);
164 } 307 }
165 308
166 ECB_HEADER_INLINE uint32_t ecb_bswap16 (uint32_t x) ecb_const; 309 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
167 ECB_HEADER_INLINE uint32_t 310 ecb_function_ uint64_t
168 ecb_bswap16 (uint32_t x) 311 ecb_bswap64 (uint64_t x)
169 { 312 {
170 return ((x >> 8) & 0xff) 313 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
171 | ((x << 8) & 0x00ff0000)
172 | (x << 24);
173 } 314 }
174#endif 315#endif
175 316
176#if ECB_GCC_VERSION(4,5) 317#if ECB_GCC_VERSION(4,5)
177 #define ecb_unreachable() __builtin_unreachable () 318 #define ecb_unreachable() __builtin_unreachable ()
178#else 319#else
179 /* this seems to work fine, but gcc always emits a warning for it :/ */ 320 /* this seems to work fine, but gcc always emits a warning for it :/ */
180 ECB_HEADER_INLINE void ecb_unreachable (void) ecb_noreturn; 321 ecb_function_ void ecb_unreachable (void) ecb_noreturn;
181 ECB_HEADER_INLINE void ecb_unreachable (void) { } 322 ecb_function_ void ecb_unreachable (void) { }
182#endif 323#endif
183 324
325/* try to tell the compiler that some condition is definitely true */
326#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
327
184ECB_HEADER_INLINE unsigned char ecb_byteorder_helper (void) ecb_const; 328ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const;
185ECB_HEADER_INLINE unsigned char 329ecb_function_ unsigned char
186ecb_byteorder_helper (void) 330ecb_byteorder_helper (void)
187{ 331{
188 const uint32_t u = 0x11223344; 332 const uint32_t u = 0x11223344;
189 return *(unsigned char *)&u; 333 return *(unsigned char *)&u;
190} 334}
191 335
192ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) ecb_const; 336ecb_function_ ecb_bool ecb_big_endian (void) ecb_const;
193ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }; 337ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
194ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) ecb_const; 338ecb_function_ ecb_bool ecb_little_endian (void) ecb_const;
195ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }; 339ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
196 340
197#if ECB_GCC_VERSION(3,0) 341#if ECB_GCC_VERSION(3,0) || ECB_C99
198 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 342 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
199#else 343#else
200 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 344 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
201#endif 345#endif
202 346
203#if ecb_cplusplus_does_not_suck 347#if ecb_cplusplus_does_not_suck
204 // does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) 348 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
205 template<typename T, int N> 349 template<typename T, int N>
206 static inline int ecb_array_length (const T (&arr)[N]) 350 static inline int ecb_array_length (const T (&arr)[N])
207 { 351 {
208 return N; 352 return N;
209 } 353 }
210#else 354#else
211 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 355 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
212#endif 356#endif
213 357
214ECB_INLINE uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
215ECB_INLINE uint32_t
216ecb_rotr32 (uint32_t x, unsigned int count)
217{
218 return (x << (32 - count)) | (x >> count);
219}
220
221ECB_INLINE uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
222ECB_INLINE uint32_t
223ecb_rotl32 (uint32_t x, unsigned int count)
224{
225 return (x >> (32 - count)) | (x << count);
226}
227
228#endif 358#endif
229 359

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines