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

Comparing libecb/ecb.h (file contents):
Revision 1.40 by root, Sat Jun 11 17:34:57 2011 UTC vs.
Revision 1.70 by root, Sat Dec 10 11:58:38 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#endif
74
75#ifndef ECB_MEMORY_FENCE
76 #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__)
77 #if __i386__
78 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
79 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */
80 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */
81 #elif __amd64
82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
83 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")
84 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */
85 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
86 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
87 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \
88 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__)
89 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
90 #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \
91 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ )
92 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
93 #endif
94 #endif
95#endif
96
97#ifndef ECB_MEMORY_FENCE
98 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__)
99 #define ECB_MEMORY_FENCE __sync_synchronize ()
100 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
101 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
102 #elif _MSC_VER >= 1400 /* VC++ 2005 */
103 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
104 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
105 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
106 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
107 #elif defined(_WIN32)
108 #include <WinNT.h>
109 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
110 #endif
111#endif
112
113#ifndef ECB_MEMORY_FENCE
114 #if !ECB_AVOID_PTHREADS
115 /*
116 * if you get undefined symbol references to pthread_mutex_lock,
117 * or failure to find pthread.h, then you should implement
118 * the ECB_MEMORY_FENCE operations for your cpu/compiler
119 * OR provide pthread.h and link against the posix thread library
120 * of your system.
121 */
122 #include <pthread.h>
123 #define ECB_NEEDS_PTHREADS 1
124 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
125
126 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
127 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
128 #endif
129#endif
130
131#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE)
132 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
133#endif
134
135#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE)
136 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
137#endif
138
139/*****************************************************************************/
140
50#define ECB_C99 (__STDC_VERSION__ >= 199901L) 141#define ECB_C99 (__STDC_VERSION__ >= 199901L)
51 142
52#if __cplusplus 143#if __cplusplus
53 #define ECB_INLINE static inline 144 #define ecb_inline static inline
54#elif ECB_GCC_VERSION(2,5) 145#elif ECB_GCC_VERSION(2,5)
55 #define ECB_INLINE static __inline__ 146 #define ecb_inline static __inline__
56#elif ECB_C99 147#elif ECB_C99
57 #define ECB_INLINE static inline 148 #define ecb_inline static inline
58#else 149#else
59 #define ECB_INLINE static 150 #define ecb_inline static
60#endif 151#endif
61 152
62#if ECB_GCC_VERSION(3,3) 153#if ECB_GCC_VERSION(3,3)
63 #define ecb_restrict __restrict__ 154 #define ecb_restrict __restrict__
64#elif ECB_C99 155#elif ECB_C99
72#define ECB_CONCAT_(a, b) a ## b 163#define ECB_CONCAT_(a, b) a ## b
73#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 164#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
74#define ECB_STRINGIFY_(a) # a 165#define ECB_STRINGIFY_(a) # a
75#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 166#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
76 167
77#define ecb_function_ ECB_INLINE 168#define ecb_function_ ecb_inline
78 169
79#if ECB_GCC_VERSION(3,1) 170#if ECB_GCC_VERSION(3,1)
80 #define ecb_attribute(attrlist) __attribute__(attrlist) 171 #define ecb_attribute(attrlist) __attribute__(attrlist)
81 #define ecb_is_constant(expr) __builtin_constant_p (expr) 172 #define ecb_is_constant(expr) __builtin_constant_p (expr)
82 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 173 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
118#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) 209#define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
119/* for compatibility to the rest of the world */ 210/* for compatibility to the rest of the world */
120#define ecb_likely(expr) ecb_expect_true (expr) 211#define ecb_likely(expr) ecb_expect_true (expr)
121#define ecb_unlikely(expr) ecb_expect_false (expr) 212#define ecb_unlikely(expr) ecb_expect_false (expr)
122 213
123/* try to tell the compiler that some condition is definitely true */
124#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
125
126/* count trailing zero bits and count # of one bits */ 214/* count trailing zero bits and count # of one bits */
127#if ECB_GCC_VERSION(3,4) 215#if ECB_GCC_VERSION(3,4)
216 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
217 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
218 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
128 #define ecb_ctz32(x) __builtin_ctz (x) 219 #define ecb_ctz32(x) __builtin_ctz (x)
220 #define ecb_ctz64(x) __builtin_ctzll (x)
129 #define ecb_popcount32(x) __builtin_popcount (x) 221 #define ecb_popcount32(x) __builtin_popcount (x)
222 /* no popcountll */
130#else 223#else
131 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 224 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
132 ecb_function_ int 225 ecb_function_ int
133 ecb_ctz32 (uint32_t x) 226 ecb_ctz32 (uint32_t x)
134 { 227 {
135 int r = 0; 228 int r = 0;
136 229
137 x &= -x; /* this isolates the lowest bit */ 230 x &= ~x + 1; /* this isolates the lowest bit */
138 231
232#if ECB_branchless_on_i386
233 r += !!(x & 0xaaaaaaaa) << 0;
234 r += !!(x & 0xcccccccc) << 1;
235 r += !!(x & 0xf0f0f0f0) << 2;
236 r += !!(x & 0xff00ff00) << 3;
237 r += !!(x & 0xffff0000) << 4;
238#else
139 if (x & 0xaaaaaaaa) r += 1; 239 if (x & 0xaaaaaaaa) r += 1;
140 if (x & 0xcccccccc) r += 2; 240 if (x & 0xcccccccc) r += 2;
141 if (x & 0xf0f0f0f0) r += 4; 241 if (x & 0xf0f0f0f0) r += 4;
142 if (x & 0xff00ff00) r += 8; 242 if (x & 0xff00ff00) r += 8;
143 if (x & 0xffff0000) r += 16; 243 if (x & 0xffff0000) r += 16;
244#endif
144 245
145 return r; 246 return r;
247 }
248
249 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
250 ecb_function_ int
251 ecb_ctz64 (uint64_t x)
252 {
253 int shift = x & 0xffffffffU ? 0 : 32;
254 return ecb_ctz32 (x >> shift) + shift;
146 } 255 }
147 256
148 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 257 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
149 ecb_function_ int 258 ecb_function_ int
150 ecb_popcount32 (uint32_t x) 259 ecb_popcount32 (uint32_t x)
154 x = ((x >> 4) + x) & 0x0f0f0f0f; 263 x = ((x >> 4) + x) & 0x0f0f0f0f;
155 x *= 0x01010101; 264 x *= 0x01010101;
156 265
157 return x >> 24; 266 return x >> 24;
158 } 267 }
268
269 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
270 ecb_function_ int ecb_ld32 (uint32_t x)
271 {
272 int r = 0;
273
274 if (x >> 16) { x >>= 16; r += 16; }
275 if (x >> 8) { x >>= 8; r += 8; }
276 if (x >> 4) { x >>= 4; r += 4; }
277 if (x >> 2) { x >>= 2; r += 2; }
278 if (x >> 1) { r += 1; }
279
280 return r;
281 }
282
283 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
284 ecb_function_ int ecb_ld64 (uint64_t x)
285 {
286 int r = 0;
287
288 if (x >> 32) { x >>= 32; r += 32; }
289
290 return r + ecb_ld32 (x);
291 }
159#endif 292#endif
293
294ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
295{
296 return ( (x * 0x0802U & 0x22110U)
297 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
298}
299
300ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
301{
302 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
303 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
304 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
305 x = ( x >> 8 ) | ( x << 8);
306
307 return x;
308}
309
310ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
311{
312 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
313 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
314 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
315 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
316 x = ( x >> 16 ) | ( x << 16);
317
318 return x;
319}
320
321/* popcount64 is only available on 64 bit cpus as gcc builtin */
322/* so for this version we are lazy */
323ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
324ecb_function_ int
325ecb_popcount64 (uint64_t x)
326{
327 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
328}
329
330ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
331ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
332ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
333ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
334ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
335ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
336ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
337ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
338
339ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
340ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
341ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
342ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
343ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
344ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
345ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
346ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
160 347
161#if ECB_GCC_VERSION(4,3) 348#if ECB_GCC_VERSION(4,3)
162 #define ecb_bswap32(x) __builtin_bswap32 (x)
163 #define ecb_bswap16(x) (__builtin_bswap32(x) >> 16) 349 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
350 #define ecb_bswap32(x) __builtin_bswap32 (x)
351 #define ecb_bswap64(x) __builtin_bswap64 (x)
164#else 352#else
353 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
354 ecb_function_ uint16_t
355 ecb_bswap16 (uint16_t x)
356 {
357 return ecb_rotl16 (x, 8);
358 }
359
165 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 360 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
166 ecb_function_ uint32_t 361 ecb_function_ uint32_t
167 ecb_bswap32 (uint32_t x) 362 ecb_bswap32 (uint32_t x)
168 { 363 {
169 return (x >> 24) 364 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
170 | ((x >> 8) & 0x0000ff00)
171 | ((x << 8) & 0x00ff0000)
172 | (x << 24);
173 } 365 }
174 366
175 ecb_function_ uint32_t ecb_bswap16 (uint32_t x) ecb_const; 367 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
176 ecb_function_ uint32_t 368 ecb_function_ uint64_t
177 ecb_bswap16 (uint32_t x) 369 ecb_bswap64 (uint64_t x)
178 { 370 {
179 return ((x >> 8) & 0xff) 371 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
180 | ((x << 8) & 0x00ff0000)
181 | (x << 24);
182 } 372 }
183#endif 373#endif
184 374
185#if ECB_GCC_VERSION(4,5) 375#if ECB_GCC_VERSION(4,5)
186 #define ecb_unreachable() __builtin_unreachable () 376 #define ecb_unreachable() __builtin_unreachable ()
187#else 377#else
188 /* this seems to work fine, but gcc always emits a warning for it :/ */ 378 /* this seems to work fine, but gcc always emits a warning for it :/ */
189 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 379 ecb_function_ void ecb_unreachable (void) ecb_noreturn;
190 ecb_function_ void ecb_unreachable (void) { } 380 ecb_function_ void ecb_unreachable (void) { }
191#endif 381#endif
382
383/* try to tell the compiler that some condition is definitely true */
384#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)
192 385
193ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 386ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const;
194ecb_function_ unsigned char 387ecb_function_ unsigned char
195ecb_byteorder_helper (void) 388ecb_byteorder_helper (void)
196{ 389{
197 const uint32_t u = 0x11223344; 390 const uint32_t u = 0x11223344;
198 return *(unsigned char *)&u; 391 return *(unsigned char *)&u;
199} 392}
200 393
201ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 394ecb_function_ ecb_bool ecb_big_endian (void) ecb_const;
202ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }; 395ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
203ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 396ecb_function_ ecb_bool ecb_little_endian (void) ecb_const;
204ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }; 397ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
205 398
206#if ECB_GCC_VERSION(3,0) || ECB_C99 399#if ECB_GCC_VERSION(3,0) || ECB_C99
207 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 400 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
208#else 401#else
209 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 402 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
403#endif
404
405#if __cplusplus
406 template<typename T>
407 static inline T ecb_div_rd (T val, T div)
408 {
409 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
410 }
411 template<typename T>
412 static inline T ecb_div_ru (T val, T div)
413 {
414 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
415 }
416#else
417 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
418 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
210#endif 419#endif
211 420
212#if ecb_cplusplus_does_not_suck 421#if ecb_cplusplus_does_not_suck
213 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 422 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
214 template<typename T, int N> 423 template<typename T, int N>
218 } 427 }
219#else 428#else
220 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 429 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
221#endif 430#endif
222 431
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 432#endif
238 433

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines