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 1.53 by root, Mon Jul 18 01:25:46 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 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
66/*****************************************************************************/
67
68#ifndef ECB_MEMORY_FENCE
69 #if ECB_GCC_VERSION(2,5)
70 #if __x86
71 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; or.b $0, -1(%%esp)" : : : "memory")
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
79 #endif
80#endif
81
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
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) && defined(MemoryBarrier)
92 #define ECB_MEMORY_FENCE MemoryBarrier ()
93 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
94 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
95 #endif
96#endif
97
98#ifndef ECB_MEMORY_FENCE
99 #include <pthread.h>
100
101 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
102 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
103 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
104 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
105#endif
106
107/*****************************************************************************/
108
50#define ECB_C99 (__STDC_VERSION__ >= 199901L) 109#define ECB_C99 (__STDC_VERSION__ >= 199901L)
51 110
52#if __cplusplus 111#if __cplusplus
53 #define ECB_INLINE static inline 112 #define ecb_inline static inline
54#elif ECB_GCC_VERSION(2,5) 113#elif ECB_GCC_VERSION(2,5)
55 #define ECB_INLINE static __inline__ 114 #define ecb_inline static __inline__
56#elif ECB_C99 115#elif ECB_C99
57 #define ECB_INLINE static inline 116 #define ecb_inline static inline
58#else 117#else
59 #define ECB_INLINE static 118 #define ecb_inline static
60#endif 119#endif
61 120
62#if ECB_GCC_VERSION(3,3) 121#if ECB_GCC_VERSION(3,3)
63 #define ecb_restrict __restrict__ 122 #define ecb_restrict __restrict__
64#elif ECB_C99 123#elif ECB_C99
72#define ECB_CONCAT_(a, b) a ## b 131#define ECB_CONCAT_(a, b) a ## b
73#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 132#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
74#define ECB_STRINGIFY_(a) # a 133#define ECB_STRINGIFY_(a) # a
75#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 134#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
76 135
77#define ecb_function_ ECB_INLINE 136#define ecb_function_ ecb_inline
78 137
79#if ECB_GCC_VERSION(3,1) 138#if ECB_GCC_VERSION(3,1)
80 #define ecb_attribute(attrlist) __attribute__(attrlist) 139 #define ecb_attribute(attrlist) __attribute__(attrlist)
81 #define ecb_is_constant(expr) __builtin_constant_p (expr) 140 #define ecb_is_constant(expr) __builtin_constant_p (expr)
82 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 141 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
120#define ecb_likely(expr) ecb_expect_true (expr) 179#define ecb_likely(expr) ecb_expect_true (expr)
121#define ecb_unlikely(expr) ecb_expect_false (expr) 180#define ecb_unlikely(expr) ecb_expect_false (expr)
122 181
123/* count trailing zero bits and count # of one bits */ 182/* count trailing zero bits and count # of one bits */
124#if ECB_GCC_VERSION(3,4) 183#if ECB_GCC_VERSION(3,4)
184 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
185 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
186 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
125 #define ecb_ctz32(x) __builtin_ctz (x) 187 #define ecb_ctz32(x) __builtin_ctz (x)
188 #define ecb_ctz64(x) __builtin_ctzll (x)
126 #define ecb_popcount32(x) __builtin_popcount (x) 189 #define ecb_popcount32(x) __builtin_popcount (x)
190 /* no popcountll */
127#else 191#else
128 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 192 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
129 ecb_function_ int 193 ecb_function_ int
130 ecb_ctz32 (uint32_t x) 194 ecb_ctz32 (uint32_t x)
131 { 195 {
132 int r = 0; 196 int r = 0;
133 197
134 x &= -x; /* this isolates the lowest bit */ 198 x &= ~x + 1; /* this isolates the lowest bit */
135 199
200#if ECB_branchless_on_i386
201 r += !!(x & 0xaaaaaaaa) << 0;
202 r += !!(x & 0xcccccccc) << 1;
203 r += !!(x & 0xf0f0f0f0) << 2;
204 r += !!(x & 0xff00ff00) << 3;
205 r += !!(x & 0xffff0000) << 4;
206#else
136 if (x & 0xaaaaaaaa) r += 1; 207 if (x & 0xaaaaaaaa) r += 1;
137 if (x & 0xcccccccc) r += 2; 208 if (x & 0xcccccccc) r += 2;
138 if (x & 0xf0f0f0f0) r += 4; 209 if (x & 0xf0f0f0f0) r += 4;
139 if (x & 0xff00ff00) r += 8; 210 if (x & 0xff00ff00) r += 8;
140 if (x & 0xffff0000) r += 16; 211 if (x & 0xffff0000) r += 16;
212#endif
141 213
142 return r; 214 return r;
215 }
216
217 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
218 ecb_function_ int
219 ecb_ctz64 (uint64_t x)
220 {
221 int shift = x & 0xffffffffU ? 0 : 32;
222 return ecb_ctz32 (x >> shift) + shift;
143 } 223 }
144 224
145 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 225 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
146 ecb_function_ int 226 ecb_function_ int
147 ecb_popcount32 (uint32_t x) 227 ecb_popcount32 (uint32_t x)
151 x = ((x >> 4) + x) & 0x0f0f0f0f; 231 x = ((x >> 4) + x) & 0x0f0f0f0f;
152 x *= 0x01010101; 232 x *= 0x01010101;
153 233
154 return x >> 24; 234 return x >> 24;
155 } 235 }
236
237 /* you have the choice beetween something with a table lookup, */
238 /* something using lots of bit arithmetic and a simple loop */
239 /* we went for the loop */
240 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
241 ecb_function_ int ecb_ld32 (uint32_t x)
242 {
243 int r = 0;
244
245 if (x >> 16) { x >>= 16; r += 16; }
246 if (x >> 8) { x >>= 8; r += 8; }
247 if (x >> 4) { x >>= 4; r += 4; }
248 if (x >> 2) { x >>= 2; r += 2; }
249 if (x >> 1) { r += 1; }
250
251 return r;
252 }
253
254 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
255 ecb_function_ int ecb_ld64 (uint64_t x)
256 {
257 int r = 0;
258
259 if (x >> 32) { x >>= 32; r += 32; }
260
261 return r + ecb_ld32 (x);
262 }
156#endif 263#endif
264
265/* popcount64 is only available on 64 bit cpus as gcc builtin */
266/* so for this version we are lazy */
267ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
268ecb_function_ int
269ecb_popcount64 (uint64_t x)
270{
271 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
272}
273
274ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
275ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
276ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
277ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
278ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
279ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
280ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
281ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
282
283ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
284ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
285ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
286ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
287ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
288ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
289ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
290ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
157 291
158#if ECB_GCC_VERSION(4,3) 292#if ECB_GCC_VERSION(4,3)
159 #define ecb_bswap32(x) __builtin_bswap32 (x)
160 #define ecb_bswap16(x) (__builtin_bswap32(x) >> 16) 293 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
294 #define ecb_bswap32(x) __builtin_bswap32 (x)
295 #define ecb_bswap64(x) __builtin_bswap64 (x)
161#else 296#else
297 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
298 ecb_function_ uint16_t
299 ecb_bswap16 (uint16_t x)
300 {
301 return ecb_rotl16 (x, 8);
302 }
303
162 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 304 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
163 ecb_function_ uint32_t 305 ecb_function_ uint32_t
164 ecb_bswap32 (uint32_t x) 306 ecb_bswap32 (uint32_t x)
165 { 307 {
166 return (x >> 24) 308 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
167 | ((x >> 8) & 0x0000ff00)
168 | ((x << 8) & 0x00ff0000)
169 | (x << 24);
170 } 309 }
171 310
172 ecb_function_ uint32_t ecb_bswap16 (uint32_t x) ecb_const; 311 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
173 ecb_function_ uint32_t 312 ecb_function_ uint64_t
174 ecb_bswap16 (uint32_t x) 313 ecb_bswap64 (uint64_t x)
175 { 314 {
176 return ((x >> 8) & 0xff) 315 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
177 | ((x << 8) & 0x00ff0000)
178 | (x << 24);
179 } 316 }
180#endif 317#endif
181 318
182#if ECB_GCC_VERSION(4,5) 319#if ECB_GCC_VERSION(4,5)
183 #define ecb_unreachable() __builtin_unreachable () 320 #define ecb_unreachable() __builtin_unreachable ()
218 } 355 }
219#else 356#else
220 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 357 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
221#endif 358#endif
222 359
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 360#endif
238 361

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines