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

Comparing libecb/ecb.h (file contents):
Revision 1.49 by root, Fri Jun 17 18:46:19 2011 UTC vs.
Revision 1.60 by root, Thu Aug 4 13:35:46 2011 UTC

… …
38 typedef signed int int32_t; 38 typedef signed int int32_t;
39 typedef unsigned int uint32_t; 39 typedef unsigned int uint32_t;
40 #if __GNUC__ 40 #if __GNUC__
41 typedef signed long long int64_t; 41 typedef signed long long int64_t;
42 typedef unsigned long long uint64_t; 42 typedef unsigned long long uint64_t;
43 #else 43 #else /* _MSC_VER || __BORLANDC__ */
44 typedef signed __int64 int64_t; 44 typedef signed __int64 int64_t;
45 typedef unsigned __int64 uint64_t; 45 typedef unsigned __int64 uint64_t;
46 #endif 46 #endif
47#else 47#else
48 #include <inttypes.h> 48 #include <inttypes.h>
49#endif 49#endif
50 50
51/* many compilers define _GNUC_ to some versions but then only implement 51/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions, 52 * what their idiot authors think are the "more important" extensions,
53 * causing enourmous grief in return for some better fake benchmark numbers. 53 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so. 54 * or so.
55 * 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
56 * 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.
57 */ 57 */
58#ifndef ECB_GCC_VERSION 58#ifndef ECB_GCC_VERSION
… …
61 #else 61 #else
62 #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 63 #endif
64#endif 64#endif
65 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
66#define ECB_C99 (__STDC_VERSION__ >= 199901L) 129#define ECB_C99 (__STDC_VERSION__ >= 199901L)
67 130
68#if __cplusplus 131#if __cplusplus
69 #define ecb_inline static inline 132 #define ecb_inline static inline
70#elif ECB_GCC_VERSION(2,5) 133#elif ECB_GCC_VERSION(2,5)
… …
152 { 215 {
153 int r = 0; 216 int r = 0;
154 217
155 x &= ~x + 1; /* this isolates the lowest bit */ 218 x &= ~x + 1; /* this isolates the lowest bit */
156 219
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
157 if (x & 0xaaaaaaaa) r += 1; 227 if (x & 0xaaaaaaaa) r += 1;
158 if (x & 0xcccccccc) r += 2; 228 if (x & 0xcccccccc) r += 2;
159 if (x & 0xf0f0f0f0) r += 4; 229 if (x & 0xf0f0f0f0) r += 4;
160 if (x & 0xff00ff00) r += 8; 230 if (x & 0xff00ff00) r += 8;
161 if (x & 0xffff0000) r += 16; 231 if (x & 0xffff0000) r += 16;
232#endif
162 233
163 return r; 234 return r;
164 } 235 }
165 236
166 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 237 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
167 ecb_function_ int 238 ecb_function_ int
168 ecb_ctz64 (uint64_t x) 239 ecb_ctz64 (uint64_t x)
169 { 240 {
170 int shift = x & 0xffffffffU ? 0 : 32; 241 int shift = x & 0xffffffffU ? 0 : 32;
171 return ecb_ctz (x >> shift) + shift; 242 return ecb_ctz32 (x >> shift) + shift;
172 } 243 }
173 244
174 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 245 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
175 ecb_function_ int 246 ecb_function_ int
176 ecb_popcount32 (uint32_t x) 247 ecb_popcount32 (uint32_t x)
… …
181 x *= 0x01010101; 252 x *= 0x01010101;
182 253
183 return x >> 24; 254 return x >> 24;
184 } 255 }
185 256
186 /* you have the choice beetween something with a table lookup, */
187 /* something using lots of bit arithmetic and a simple loop */
188 /* we went for the loop */
189 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 257 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
190 ecb_function_ int ecb_ld32 (uint32_t x) 258 ecb_function_ int ecb_ld32 (uint32_t x)
191 { 259 {
192 int r = -1; 260 int r = 0;
193 261
194 do 262 if (x >> 16) { x >>= 16; r += 16; }
195 { 263 if (x >> 8) { x >>= 8; r += 8; }
196 x >>= 1; 264 if (x >> 4) { x >>= 4; r += 4; }
197 ++r; 265 if (x >> 2) { x >>= 2; r += 2; }
198 } 266 if (x >> 1) { r += 1; }
199 while (x);
200 267
201 return r; 268 return r;
202 } 269 }
203 270
204 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 271 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
205 ecb_function_ int ecb_ld64 (uint64_t x) 272 ecb_function_ int ecb_ld64 (uint64_t x)
206 { 273 {
207 int r = -1; 274 int r = 0;
208 275
209 do 276 if (x >> 32) { x >>= 32; r += 32; }
210 {
211 x >>= 1;
212 ++r;
213 }
214 while (x);
215 277
216 return r; 278 return r + ecb_ld32 (x);
217 } 279 }
218#endif 280#endif
219 281
220/* popcount64 is only available on 64 bit cpus as gcc builtin */ 282/* popcount64 is only available on 64 bit cpus as gcc builtin */
221/* so for this version we are lazy */ 283/* so for this version we are lazy */
… …
224ecb_popcount64 (uint64_t x) 286ecb_popcount64 (uint64_t x)
225{ 287{
226 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 288 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
227} 289}
228 290
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); }
308
229#if ECB_GCC_VERSION(4,3) 309#if ECB_GCC_VERSION(4,3)
230 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 310 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
231 #define ecb_bswap32(x) __builtin_bswap32 (x) 311 #define ecb_bswap32(x) __builtin_bswap32 (x)
232 #define ecb_bswap64(x) __builtin_bswap64 (x) 312 #define ecb_bswap64(x) __builtin_bswap64 (x)
233#else 313#else
234 ecb_function_ uint32_t ecb_bswap16 (uint32_t x) ecb_const; 314 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
235 ecb_function_ uint32_t 315 ecb_function_ uint16_t
236 ecb_bswap16 (uint32_t x) 316 ecb_bswap16 (uint16_t x)
237 { 317 {
238 return ((x >> 8) & 0xff) 318 return ecb_rotl16 (x, 8);
239 | ((x << 8) & 0x00ff0000)
240 | (x << 24);
241 } 319 }
242 320
243 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 321 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
244 ecb_function_ uint32_t 322 ecb_function_ uint32_t
245 ecb_bswap32 (uint32_t x) 323 ecb_bswap32 (uint32_t x)
246 { 324 {
247 return (x >> 24) 325 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
248 | ((x >> 8) & 0x0000ff00)
249 | ((x << 8) & 0x00ff0000)
250 | (x << 24);
251 } 326 }
252 327
253 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 328 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
254 ecb_function_ uint64_t 329 ecb_function_ uint64_t
255 ecb_bswap64 (uint64_t x) 330 ecb_bswap64 (uint64_t x)
256 { 331 {
257 return (((uint64_t)ecb_bswap32 (x)) << 32) 332 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
258 | ecb_bswap32 (x >> 32);
259 } 333 }
260#endif 334#endif
261 335
262#if ECB_GCC_VERSION(4,5) 336#if ECB_GCC_VERSION(4,5)
263 #define ecb_unreachable() __builtin_unreachable () 337 #define ecb_unreachable() __builtin_unreachable ()
… …
298 } 372 }
299#else 373#else
300 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 374 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
301#endif 375#endif
302 376
303ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
304ecb_inline uint32_t
305ecb_rotr32 (uint32_t x, unsigned int count)
306{
307 return (x << (32 - count)) | (x >> count);
308}
309
310ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
311ecb_inline uint32_t
312ecb_rotl32 (uint32_t x, unsigned int count)
313{
314 return (x >> (32 - count)) | (x << count);
315}
316
317ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
318ecb_inline uint64_t
319ecb_rotr64 (uint64_t x, unsigned int count)
320{
321 return (x << (64 - count)) | (x >> count);
322}
323
324ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
325ecb_inline uint64_t
326ecb_rotl64 (uint64_t x, unsigned int count)
327{
328 return (x >> (64 - count)) | (x << count);
329}
330
331#endif 377#endif
332 378

Diff Legend

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