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

Comparing libecb/ecb.h (file contents):
Revision 1.42 by root, Sun Jun 12 08:14:03 2011 UTC vs.
Revision 1.60 by root, Thu Aug 4 13:35:46 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 enormous 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
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 #endif
88 #endif
89#endif
90
91#ifndef ECB_MEMORY_FENCE
92 #if ECB_GCC_VERSION(4,4)
93 #define ECB_MEMORY_FENCE __sync_synchronize ()
94 #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); })
95 #define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); })
96 #elif _MSC_VER >= 1400 /* VC++ 2005 */
97 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
98 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
99 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
100 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
101 #elif defined(_WIN32)
102 #include <WinNT.h>
103 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
104 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
105 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
106 #endif
107#endif
108
109#ifndef ECB_MEMORY_FENCE
110 /*
111 * if you get undefined symbol references to pthread_mutex_lock,
112 * or failure to find pthread.h, then you should implement
113 * the ECB_MEMORY_FENCE operations for your cpu/compiler
114 * OR provide pthread.h and link against the posix thread library
115 * of your system.
116 */
117 #include <pthread.h>
118 #define ECB_NEEDS_PTHREADS 1
119 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
120
121 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
122 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
123 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
124 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
125#endif
126
127/*****************************************************************************/
128
50#define ECB_C99 (__STDC_VERSION__ >= 199901L) 129#define ECB_C99 (__STDC_VERSION__ >= 199901L)
51 130
52#if __cplusplus 131#if __cplusplus
53 #define ECB_INLINE static inline 132 #define ecb_inline static inline
54#elif ECB_GCC_VERSION(2,5) 133#elif ECB_GCC_VERSION(2,5)
55 #define ECB_INLINE static __inline__ 134 #define ecb_inline static __inline__
56#elif ECB_C99 135#elif ECB_C99
57 #define ECB_INLINE static inline 136 #define ecb_inline static inline
58#else 137#else
59 #define ECB_INLINE static 138 #define ecb_inline static
60#endif 139#endif
61 140
62#if ECB_GCC_VERSION(3,3) 141#if ECB_GCC_VERSION(3,3)
63 #define ecb_restrict __restrict__ 142 #define ecb_restrict __restrict__
64#elif ECB_C99 143#elif ECB_C99
… …
72#define ECB_CONCAT_(a, b) a ## b 151#define ECB_CONCAT_(a, b) a ## b
73#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 152#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
74#define ECB_STRINGIFY_(a) # a 153#define ECB_STRINGIFY_(a) # a
75#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 154#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
76 155
77#define ecb_function_ ECB_INLINE 156#define ecb_function_ ecb_inline
78 157
79#if ECB_GCC_VERSION(3,1) 158#if ECB_GCC_VERSION(3,1)
80 #define ecb_attribute(attrlist) __attribute__(attrlist) 159 #define ecb_attribute(attrlist) __attribute__(attrlist)
81 #define ecb_is_constant(expr) __builtin_constant_p (expr) 160 #define ecb_is_constant(expr) __builtin_constant_p (expr)
82 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 161 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
… …
120#define ecb_likely(expr) ecb_expect_true (expr) 199#define ecb_likely(expr) ecb_expect_true (expr)
121#define ecb_unlikely(expr) ecb_expect_false (expr) 200#define ecb_unlikely(expr) ecb_expect_false (expr)
122 201
123/* count trailing zero bits and count # of one bits */ 202/* count trailing zero bits and count # of one bits */
124#if ECB_GCC_VERSION(3,4) 203#if ECB_GCC_VERSION(3,4)
204 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
205 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
206 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
125 #define ecb_ctz32(x) __builtin_ctz (x) 207 #define ecb_ctz32(x) __builtin_ctz (x)
208 #define ecb_ctz64(x) __builtin_ctzll (x)
126 #define ecb_popcount32(x) __builtin_popcount (x) 209 #define ecb_popcount32(x) __builtin_popcount (x)
210 /* no popcountll */
127#else 211#else
128 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 212 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
129 ecb_function_ int 213 ecb_function_ int
130 ecb_ctz32 (uint32_t x) 214 ecb_ctz32 (uint32_t x)
131 { 215 {
132 int r = 0; 216 int r = 0;
133 217
134 x &= -x; /* this isolates the lowest bit */ 218 x &= ~x + 1; /* this isolates the lowest bit */
135 219
220#if ECB_branchless_on_i386
221 r += !!(x & 0xaaaaaaaa) << 0;
222 r += !!(x & 0xcccccccc) << 1;
223 r += !!(x & 0xf0f0f0f0) << 2;
224 r += !!(x & 0xff00ff00) << 3;
225 r += !!(x & 0xffff0000) << 4;
226#else
136 if (x & 0xaaaaaaaa) r += 1; 227 if (x & 0xaaaaaaaa) r += 1;
137 if (x & 0xcccccccc) r += 2; 228 if (x & 0xcccccccc) r += 2;
138 if (x & 0xf0f0f0f0) r += 4; 229 if (x & 0xf0f0f0f0) r += 4;
139 if (x & 0xff00ff00) r += 8; 230 if (x & 0xff00ff00) r += 8;
140 if (x & 0xffff0000) r += 16; 231 if (x & 0xffff0000) r += 16;
232#endif
141 233
142 return r; 234 return r;
235 }
236
237 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
238 ecb_function_ int
239 ecb_ctz64 (uint64_t x)
240 {
241 int shift = x & 0xffffffffU ? 0 : 32;
242 return ecb_ctz32 (x >> shift) + shift;
143 } 243 }
144 244
145 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 245 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
146 ecb_function_ int 246 ecb_function_ int
147 ecb_popcount32 (uint32_t x) 247 ecb_popcount32 (uint32_t x)
… …
151 x = ((x >> 4) + x) & 0x0f0f0f0f; 251 x = ((x >> 4) + x) & 0x0f0f0f0f;
152 x *= 0x01010101; 252 x *= 0x01010101;
153 253
154 return x >> 24; 254 return x >> 24;
155 } 255 }
256
257 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
258 ecb_function_ int ecb_ld32 (uint32_t x)
259 {
260 int r = 0;
261
262 if (x >> 16) { x >>= 16; r += 16; }
263 if (x >> 8) { x >>= 8; r += 8; }
264 if (x >> 4) { x >>= 4; r += 4; }
265 if (x >> 2) { x >>= 2; r += 2; }
266 if (x >> 1) { r += 1; }
267
268 return r;
269 }
270
271 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
272 ecb_function_ int ecb_ld64 (uint64_t x)
273 {
274 int r = 0;
275
276 if (x >> 32) { x >>= 32; r += 32; }
277
278 return r + ecb_ld32 (x);
279 }
156#endif 280#endif
281
282/* popcount64 is only available on 64 bit cpus as gcc builtin */
283/* so for this version we are lazy */
284ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
285ecb_function_ int
286ecb_popcount64 (uint64_t x)
287{
288 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
289}
290
291ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
292ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
293ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
294ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
295ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
296ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
297ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
298ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
299
300ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
301ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
302ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
303ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
304ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
305ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
306ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
307ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
157 308
158#if ECB_GCC_VERSION(4,3) 309#if ECB_GCC_VERSION(4,3)
159 #define ecb_bswap32(x) __builtin_bswap32 (x)
160 #define ecb_bswap16(x) (__builtin_bswap32(x) >> 16) 310 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
311 #define ecb_bswap32(x) __builtin_bswap32 (x)
312 #define ecb_bswap64(x) __builtin_bswap64 (x)
161#else 313#else
314 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
315 ecb_function_ uint16_t
316 ecb_bswap16 (uint16_t x)
317 {
318 return ecb_rotl16 (x, 8);
319 }
320
162 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 321 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
163 ecb_function_ uint32_t 322 ecb_function_ uint32_t
164 ecb_bswap32 (uint32_t x) 323 ecb_bswap32 (uint32_t x)
165 { 324 {
166 return (x >> 24) 325 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
167 | ((x >> 8) & 0x0000ff00)
168 | ((x << 8) & 0x00ff0000)
169 | (x << 24);
170 } 326 }
171 327
172 ecb_function_ uint32_t ecb_bswap16 (uint32_t x) ecb_const; 328 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
173 ecb_function_ uint32_t 329 ecb_function_ uint64_t
174 ecb_bswap16 (uint32_t x) 330 ecb_bswap64 (uint64_t x)
175 { 331 {
176 return ((x >> 8) & 0xff) 332 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
177 | ((x << 8) & 0x00ff0000)
178 | (x << 24);
179 } 333 }
180#endif 334#endif
181 335
182#if ECB_GCC_VERSION(4,5) 336#if ECB_GCC_VERSION(4,5)
183 #define ecb_unreachable() __builtin_unreachable () 337 #define ecb_unreachable() __builtin_unreachable ()
… …
218 } 372 }
219#else 373#else
220 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 374 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
221#endif 375#endif
222 376
223ECB_INLINE uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
224ECB_INLINE uint32_t
225ecb_rotr32 (uint32_t x, unsigned int count)
226{
227 return (x << (32 - count)) | (x >> count);
228}
229
230ECB_INLINE uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
231ECB_INLINE uint32_t
232ecb_rotl32 (uint32_t x, unsigned int count)
233{
234 return (x >> (32 - count)) | (x << count);
235}
236
237#endif 377#endif
238 378

Diff Legend

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