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

Comparing libeio/ecb.h (file contents):
Revision 1.17 by root, Fri Dec 28 07:33:41 2012 UTC vs.
Revision 1.23 by root, Wed Feb 11 19:32:36 2015 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de>
5 * Copyright (©) 2011 Emanuele Giaquinta 5 * Copyright (©) 2011 Emanuele Giaquinta
6 * All rights reserved. 6 * All rights reserved.
7 * 7 *
8 * Redistribution and use in source and binary forms, with or without modifica- 8 * Redistribution and use in source and binary forms, with or without modifica-
9 * tion, are permitted provided that the following conditions are met: 9 * tion, are permitted provided that the following conditions are met:
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- 25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED 26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27 * OF THE POSSIBILITY OF SUCH DAMAGE. 27 * OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * Alternatively, the contents of this file may be used under the terms of
30 * the GNU General Public License ("GPL") version 2 or any later version,
31 * in which case the provisions of the GPL are applicable instead of
32 * the above. If you wish to allow the use of your version of this file
33 * only under the terms of the GPL and not to allow others to use your
34 * version of this file under the BSD license, indicate your decision
35 * by deleting the provisions above and replace them with the notice
36 * and other provisions required by the GPL. If you do not delete the
37 * provisions above, a recipient may use your version of this file under
38 * either the BSD or the GPL.
28 */ 39 */
29 40
30#ifndef ECB_H 41#ifndef ECB_H
31#define ECB_H 42#define ECB_H
32 43
33/* 16 bits major, 16 bits minor */ 44/* 16 bits major, 16 bits minor */
34#define ECB_VERSION 0x00010003 45#define ECB_VERSION 0x00010004
35 46
36#ifdef _WIN32 47#ifdef _WIN32
37 typedef signed char int8_t; 48 typedef signed char int8_t;
38 typedef unsigned char uint8_t; 49 typedef unsigned char uint8_t;
39 typedef signed short int16_t; 50 typedef signed short int16_t;
64 #define ECB_PTRSIZE 4 75 #define ECB_PTRSIZE 4
65 #endif 76 #endif
66#endif 77#endif
67 78
68/* work around x32 idiocy by defining proper macros */ 79/* work around x32 idiocy by defining proper macros */
69#if __x86_64 || _M_AMD64 80#if __amd64 || __x86_64 || _M_AMD64 || _M_X64
70 #if __ILP32 81 #if _ILP32
71 #define ECB_AMD64_X32 1 82 #define ECB_AMD64_X32 1
72 #else 83 #else
73 #define ECB_AMD64 1 84 #define ECB_AMD64 1
74 #endif 85 #endif
75#endif 86#endif
79 * causing enormous grief in return for some better fake benchmark numbers. 90 * causing enormous grief in return for some better fake benchmark numbers.
80 * or so. 91 * or so.
81 * we try to detect these and simply assume they are not gcc - if they have 92 * we try to detect these and simply assume they are not gcc - if they have
82 * an issue with that they should have done it right in the first place. 93 * an issue with that they should have done it right in the first place.
83 */ 94 */
84#ifndef ECB_GCC_VERSION
85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ 95#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
86 #define ECB_GCC_VERSION(major,minor) 0 96 #define ECB_GCC_VERSION(major,minor) 0
87 #else 97#else
88 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 98 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
89 #endif 99#endif
90#endif
91 100
92#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ 101#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
93#define ECB_C99 (__STDC_VERSION__ >= 199901L) 102
94#define ECB_C11 (__STDC_VERSION__ >= 201112L) 103#if __clang__ && defined __has_builtin
104 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
105#else
106 #define ECB_CLANG_BUILTIN(x) 0
107#endif
108
109#if __clang__ && defined __has_extension
110 #define ECB_CLANG_EXTENSION(x) __has_extension (x)
111#else
112 #define ECB_CLANG_EXTENSION(x) 0
113#endif
114
95#define ECB_CPP (__cplusplus+0) 115#define ECB_CPP (__cplusplus+0)
96#define ECB_CPP11 (__cplusplus >= 201103L) 116#define ECB_CPP11 (__cplusplus >= 201103L)
117
118#if ECB_CPP
119 #define ECB_C 0
120 #define ECB_STDC_VERSION 0
121#else
122 #define ECB_C 1
123 #define ECB_STDC_VERSION __STDC_VERSION__
124#endif
125
126#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
127#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
97 128
98#if ECB_CPP 129#if ECB_CPP
99 #define ECB_EXTERN_C extern "C" 130 #define ECB_EXTERN_C extern "C"
100 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 131 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
101 #define ECB_EXTERN_C_END } 132 #define ECB_EXTERN_C_END }
134 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 165 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 166 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
136 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 167 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
137 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 168 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
138 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 169 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
170 #elif __aarch64__
171 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
139 #elif __sparc || __sparc__ 172 #elif (__sparc || __sparc__) && !__sparcv8
140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") 173 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
141 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 174 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 175 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
143 #elif defined __s390__ || defined __s390x__ 176 #elif defined __s390__ || defined __s390x__
144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 177 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
145 #elif defined __mips__ 178 #elif defined __mips__
179 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
180 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
146 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 181 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
147 #elif defined __alpha__ 182 #elif defined __alpha__
148 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") 183 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
149 #elif defined __hppa__ 184 #elif defined __hppa__
150 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") 185 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
151 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 186 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
152 #elif defined __ia64__ 187 #elif defined __ia64__
153 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") 188 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
189 #elif defined __m68k__
190 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
191 #elif defined __m88k__
192 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
193 #elif defined __sh__
194 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
154 #endif 195 #endif
155 #endif 196 #endif
156#endif 197#endif
157 198
158#ifndef ECB_MEMORY_FENCE 199#ifndef ECB_MEMORY_FENCE
159 #if ECB_GCC_VERSION(4,7) 200 #if ECB_GCC_VERSION(4,7)
160 /* see comment below (stdatomic.h) about the C11 memory model. */ 201 /* see comment below (stdatomic.h) about the C11 memory model. */
161 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 202 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
203 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
204 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
162 205
163 /* The __has_feature syntax from clang is so misdesigned that we cannot use it 206 #elif ECB_CLANG_EXTENSION(c_atomic)
164 * without risking compile time errors with other compilers. We *could*
165 * define our own ecb_clang_has_feature, but I just can't be bothered to work
166 * around this shit time and again.
167 * #elif defined __clang && __has_feature (cxx_atomic)
168 * // see comment below (stdatomic.h) about the C11 memory model. 207 /* see comment below (stdatomic.h) about the C11 memory model. */
169 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 208 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
170 */ 209 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
210 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
171 211
172 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 212 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
173 #define ECB_MEMORY_FENCE __sync_synchronize () 213 #define ECB_MEMORY_FENCE __sync_synchronize ()
214 #elif _MSC_VER >= 1500 /* VC++ 2008 */
215 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
216 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
217 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
218 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
219 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
174 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 220 #elif _MSC_VER >= 1400 /* VC++ 2005 */
175 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 221 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
176 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 222 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
177 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 223 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
178 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 224 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
198 /* any fence other than seq_cst, which isn't very efficient for us. */ 244 /* any fence other than seq_cst, which isn't very efficient for us. */
199 /* Why that is, we don't know - either the C11 memory model is quite useless */ 245 /* Why that is, we don't know - either the C11 memory model is quite useless */
200 /* for most usages, or gcc and clang have a bug */ 246 /* for most usages, or gcc and clang have a bug */
201 /* I *currently* lean towards the latter, and inefficiently implement */ 247 /* I *currently* lean towards the latter, and inefficiently implement */
202 /* all three of ecb's fences as a seq_cst fence */ 248 /* all three of ecb's fences as a seq_cst fence */
249 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
250 /* for all __atomic_thread_fence's except seq_cst */
203 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) 251 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
204 #endif 252 #endif
205#endif 253#endif
206 254
207#ifndef ECB_MEMORY_FENCE 255#ifndef ECB_MEMORY_FENCE
230 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 278 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
231#endif 279#endif
232 280
233/*****************************************************************************/ 281/*****************************************************************************/
234 282
235#if __cplusplus 283#if ECB_CPP
236 #define ecb_inline static inline 284 #define ecb_inline static inline
237#elif ECB_GCC_VERSION(2,5) 285#elif ECB_GCC_VERSION(2,5)
238 #define ecb_inline static __inline__ 286 #define ecb_inline static __inline__
239#elif ECB_C99 287#elif ECB_C99
240 #define ecb_inline static inline 288 #define ecb_inline static inline
257#define ECB_STRINGIFY_(a) # a 305#define ECB_STRINGIFY_(a) # a
258#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 306#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
259 307
260#define ecb_function_ ecb_inline 308#define ecb_function_ ecb_inline
261 309
262#if ECB_GCC_VERSION(3,1) 310#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
263 #define ecb_attribute(attrlist) __attribute__(attrlist) 311 #define ecb_attribute(attrlist) __attribute__ (attrlist)
312#else
313 #define ecb_attribute(attrlist)
314#endif
315
316#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
264 #define ecb_is_constant(expr) __builtin_constant_p (expr) 317 #define ecb_is_constant(expr) __builtin_constant_p (expr)
318#else
319 /* possible C11 impl for integral types
320 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
321 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
322
323 #define ecb_is_constant(expr) 0
324#endif
325
326#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
265 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 327 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
328#else
329 #define ecb_expect(expr,value) (expr)
330#endif
331
332#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
266 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 333 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
267#else 334#else
268 #define ecb_attribute(attrlist)
269 #define ecb_is_constant(expr) 0
270 #define ecb_expect(expr,value) (expr)
271 #define ecb_prefetch(addr,rw,locality) 335 #define ecb_prefetch(addr,rw,locality)
272#endif 336#endif
273 337
274/* no emulation for ecb_decltype */ 338/* no emulation for ecb_decltype */
275#if ECB_GCC_VERSION(4,5) 339#if ECB_CPP11
340 // older implementations might have problems with decltype(x)::type, work around it
341 template<class T> struct ecb_decltype_t { typedef T type; };
276 #define ecb_decltype(x) __decltype(x) 342 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
277#elif ECB_GCC_VERSION(3,0) 343#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
278 #define ecb_decltype(x) __typeof(x) 344 #define ecb_decltype(x) __typeof__ (x)
345#endif
346
347#if _MSC_VER >= 1300
348 #define ecb_deprecated __declspec (deprecated)
349#else
350 #define ecb_deprecated ecb_attribute ((__deprecated__))
279#endif 351#endif
280 352
281#define ecb_noinline ecb_attribute ((__noinline__)) 353#define ecb_noinline ecb_attribute ((__noinline__))
282#define ecb_unused ecb_attribute ((__unused__)) 354#define ecb_unused ecb_attribute ((__unused__))
283#define ecb_const ecb_attribute ((__const__)) 355#define ecb_const ecb_attribute ((__const__))
284#define ecb_pure ecb_attribute ((__pure__)) 356#define ecb_pure ecb_attribute ((__pure__))
285 357
286#if ECB_C11 358/* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
359#if ECB_C11 || __IBMC_NORETURN
360 /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */
287 #define ecb_noreturn _Noreturn 361 #define ecb_noreturn _Noreturn
288#else 362#else
289 #define ecb_noreturn ecb_attribute ((__noreturn__)) 363 #define ecb_noreturn ecb_attribute ((__noreturn__))
290#endif 364#endif
291 365
307/* for compatibility to the rest of the world */ 381/* for compatibility to the rest of the world */
308#define ecb_likely(expr) ecb_expect_true (expr) 382#define ecb_likely(expr) ecb_expect_true (expr)
309#define ecb_unlikely(expr) ecb_expect_false (expr) 383#define ecb_unlikely(expr) ecb_expect_false (expr)
310 384
311/* count trailing zero bits and count # of one bits */ 385/* count trailing zero bits and count # of one bits */
312#if ECB_GCC_VERSION(3,4) 386#if ECB_GCC_VERSION(3,4) \
387 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
388 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
389 && ECB_CLANG_BUILTIN(__builtin_popcount))
313 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ 390 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
314 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 391 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
315 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 392 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
316 #define ecb_ctz32(x) __builtin_ctz (x) 393 #define ecb_ctz32(x) __builtin_ctz (x)
317 #define ecb_ctz64(x) __builtin_ctzll (x) 394 #define ecb_ctz64(x) __builtin_ctzll (x)
318 #define ecb_popcount32(x) __builtin_popcount (x) 395 #define ecb_popcount32(x) __builtin_popcount (x)
319 /* no popcountll */ 396 /* no popcountll */
320#else 397#else
321 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 398 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
322 ecb_function_ int 399 ecb_function_ ecb_const int
323 ecb_ctz32 (uint32_t x) 400 ecb_ctz32 (uint32_t x)
324 { 401 {
325 int r = 0; 402 int r = 0;
326 403
327 x &= ~x + 1; /* this isolates the lowest bit */ 404 x &= ~x + 1; /* this isolates the lowest bit */
341#endif 418#endif
342 419
343 return r; 420 return r;
344 } 421 }
345 422
346 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 423 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
347 ecb_function_ int 424 ecb_function_ ecb_const int
348 ecb_ctz64 (uint64_t x) 425 ecb_ctz64 (uint64_t x)
349 { 426 {
350 int shift = x & 0xffffffffU ? 0 : 32; 427 int shift = x & 0xffffffffU ? 0 : 32;
351 return ecb_ctz32 (x >> shift) + shift; 428 return ecb_ctz32 (x >> shift) + shift;
352 } 429 }
353 430
354 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 431 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
355 ecb_function_ int 432 ecb_function_ ecb_const int
356 ecb_popcount32 (uint32_t x) 433 ecb_popcount32 (uint32_t x)
357 { 434 {
358 x -= (x >> 1) & 0x55555555; 435 x -= (x >> 1) & 0x55555555;
359 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 436 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
360 x = ((x >> 4) + x) & 0x0f0f0f0f; 437 x = ((x >> 4) + x) & 0x0f0f0f0f;
361 x *= 0x01010101; 438 x *= 0x01010101;
362 439
363 return x >> 24; 440 return x >> 24;
364 } 441 }
365 442
366 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 443 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
367 ecb_function_ int ecb_ld32 (uint32_t x) 444 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
368 { 445 {
369 int r = 0; 446 int r = 0;
370 447
371 if (x >> 16) { x >>= 16; r += 16; } 448 if (x >> 16) { x >>= 16; r += 16; }
372 if (x >> 8) { x >>= 8; r += 8; } 449 if (x >> 8) { x >>= 8; r += 8; }
375 if (x >> 1) { r += 1; } 452 if (x >> 1) { r += 1; }
376 453
377 return r; 454 return r;
378 } 455 }
379 456
380 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 457 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
381 ecb_function_ int ecb_ld64 (uint64_t x) 458 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
382 { 459 {
383 int r = 0; 460 int r = 0;
384 461
385 if (x >> 32) { x >>= 32; r += 32; } 462 if (x >> 32) { x >>= 32; r += 32; }
386 463
387 return r + ecb_ld32 (x); 464 return r + ecb_ld32 (x);
388 } 465 }
389#endif 466#endif
390 467
391ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 468ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
392ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 469ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
393ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 470ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
394ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 471ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
395 472
396ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 473ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
397ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 474ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
398{ 475{
399 return ( (x * 0x0802U & 0x22110U) 476 return ( (x * 0x0802U & 0x22110U)
400 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 477 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
401} 478}
402 479
403ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 480ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
404ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 481ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
405{ 482{
406 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 483 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
407 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 484 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
408 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 485 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
409 x = ( x >> 8 ) | ( x << 8); 486 x = ( x >> 8 ) | ( x << 8);
410 487
411 return x; 488 return x;
412} 489}
413 490
414ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 491ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
415ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 492ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
416{ 493{
417 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 494 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
418 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 495 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
419 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 496 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
420 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 497 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
423 return x; 500 return x;
424} 501}
425 502
426/* popcount64 is only available on 64 bit cpus as gcc builtin */ 503/* popcount64 is only available on 64 bit cpus as gcc builtin */
427/* so for this version we are lazy */ 504/* so for this version we are lazy */
428ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 505ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
429ecb_function_ int 506ecb_function_ ecb_const int
430ecb_popcount64 (uint64_t x) 507ecb_popcount64 (uint64_t x)
431{ 508{
432 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 509 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
433} 510}
434 511
435ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 512ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
436ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 513ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
437ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 514ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
438ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 515ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
439ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 516ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
440ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 517ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
441ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 518ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
442ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 519ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
443 520
444ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 521ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
445ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 522ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
446ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 523ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
447ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 524ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
448ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 525ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
449ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 526ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
450ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 527ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
451ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 528ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
452 529
453#if ECB_GCC_VERSION(4,3) 530#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
454 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 531 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
455 #define ecb_bswap32(x) __builtin_bswap32 (x) 532 #define ecb_bswap32(x) __builtin_bswap32 (x)
456 #define ecb_bswap64(x) __builtin_bswap64 (x) 533 #define ecb_bswap64(x) __builtin_bswap64 (x)
457#else 534#else
458 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 535 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
459 ecb_function_ uint16_t 536 ecb_function_ ecb_const uint16_t
460 ecb_bswap16 (uint16_t x) 537 ecb_bswap16 (uint16_t x)
461 { 538 {
462 return ecb_rotl16 (x, 8); 539 return ecb_rotl16 (x, 8);
463 } 540 }
464 541
465 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 542 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
466 ecb_function_ uint32_t 543 ecb_function_ ecb_const uint32_t
467 ecb_bswap32 (uint32_t x) 544 ecb_bswap32 (uint32_t x)
468 { 545 {
469 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 546 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
470 } 547 }
471 548
472 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 549 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
473 ecb_function_ uint64_t 550 ecb_function_ ecb_const uint64_t
474 ecb_bswap64 (uint64_t x) 551 ecb_bswap64 (uint64_t x)
475 { 552 {
476 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 553 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
477 } 554 }
478#endif 555#endif
479 556
480#if ECB_GCC_VERSION(4,5) 557#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
481 #define ecb_unreachable() __builtin_unreachable () 558 #define ecb_unreachable() __builtin_unreachable ()
482#else 559#else
483 /* this seems to work fine, but gcc always emits a warning for it :/ */ 560 /* this seems to work fine, but gcc always emits a warning for it :/ */
484 ecb_inline void ecb_unreachable (void) ecb_noreturn; 561 ecb_inline ecb_noreturn void ecb_unreachable (void);
485 ecb_inline void ecb_unreachable (void) { } 562 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
486#endif 563#endif
487 564
488/* try to tell the compiler that some condition is definitely true */ 565/* try to tell the compiler that some condition is definitely true */
489#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 566#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
490 567
491ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 568ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);
492ecb_inline unsigned char 569ecb_inline ecb_const unsigned char
493ecb_byteorder_helper (void) 570ecb_byteorder_helper (void)
494{ 571{
495 /* the union code still generates code under pressure in gcc, */ 572 /* the union code still generates code under pressure in gcc, */
496 /* but less than using pointers, and always seems to */ 573 /* but less than using pointers, and always seems to */
497 /* successfully return a constant. */ 574 /* successfully return a constant. */
512 } u = { 0x11223344 }; 589 } u = { 0x11223344 };
513 return u.c; 590 return u.c;
514#endif 591#endif
515} 592}
516 593
517ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 594ecb_inline ecb_const ecb_bool ecb_big_endian (void);
518ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 595ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
519ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 596ecb_inline ecb_const ecb_bool ecb_little_endian (void);
520ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 597ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
521 598
522#if ECB_GCC_VERSION(3,0) || ECB_C99 599#if ECB_GCC_VERSION(3,0) || ECB_C99
523 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 600 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
524#else 601#else
525 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 602 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
526#endif 603#endif
527 604
528#if __cplusplus 605#if ECB_CPP
529 template<typename T> 606 template<typename T>
530 static inline T ecb_div_rd (T val, T div) 607 static inline T ecb_div_rd (T val, T div)
531 { 608 {
532 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 609 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
533 } 610 }
559/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 636/* the only noteworthy exception is ancient armle, which uses order 43218765 */
560#if 0 \ 637#if 0 \
561 || __i386 || __i386__ \ 638 || __i386 || __i386__ \
562 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 639 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
563 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 640 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
564 || defined __arm__ && defined __ARM_EABI__ \
565 || defined __s390__ || defined __s390x__ \ 641 || defined __s390__ || defined __s390x__ \
566 || defined __mips__ \ 642 || defined __mips__ \
567 || defined __alpha__ \ 643 || defined __alpha__ \
568 || defined __hppa__ \ 644 || defined __hppa__ \
569 || defined __ia64__ \ 645 || defined __ia64__ \
646 || defined __m68k__ \
647 || defined __m88k__ \
648 || defined __sh__ \
570 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 649 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \
650 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
651 || defined __aarch64__
571 #define ECB_STDFP 1 652 #define ECB_STDFP 1
572 #include <string.h> /* for memcpy */ 653 #include <string.h> /* for memcpy */
573#else 654#else
574 #define ECB_STDFP 0 655 #define ECB_STDFP 0
575 #include <math.h> /* for frexp*, ldexp* */
576#endif 656#endif
577 657
578#ifndef ECB_NO_LIBM 658#ifndef ECB_NO_LIBM
579 659
660 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
661
662 /* only the oldest of old doesn't have this one. solaris. */
663 #ifdef INFINITY
664 #define ECB_INFINITY INFINITY
665 #else
666 #define ECB_INFINITY HUGE_VAL
667 #endif
668
669 #ifdef NAN
670 #define ECB_NAN NAN
671 #else
672 #define ECB_NAN ECB_INFINITY
673 #endif
674
675 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
676 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
677 #else
678 #define ecb_ldexpf(x,e) (float) ldexp ((x), (e))
679 #endif
680
681 /* converts an ieee half/binary16 to a float */
682 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
683 ecb_function_ ecb_const float
684 ecb_binary16_to_float (uint16_t x)
685 {
686 int e = (x >> 10) & 0x1f;
687 int m = x & 0x3ff;
688 float r;
689
690 if (!e ) r = ecb_ldexpf (m , -24);
691 else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
692 else if (m ) r = ECB_NAN;
693 else r = ECB_INFINITY;
694
695 return x & 0x8000 ? -r : r;
696 }
697
580 /* convert a float to ieee single/binary32 */ 698 /* convert a float to ieee single/binary32 */
581 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 699 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
582 ecb_function_ uint32_t 700 ecb_function_ ecb_const uint32_t
583 ecb_float_to_binary32 (float x) 701 ecb_float_to_binary32 (float x)
584 { 702 {
585 uint32_t r; 703 uint32_t r;
586 704
587 #if ECB_STDFP 705 #if ECB_STDFP
616 734
617 return r; 735 return r;
618 } 736 }
619 737
620 /* converts an ieee single/binary32 to a float */ 738 /* converts an ieee single/binary32 to a float */
621 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 739 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
622 ecb_function_ float 740 ecb_function_ ecb_const float
623 ecb_binary32_to_float (uint32_t x) 741 ecb_binary32_to_float (uint32_t x)
624 { 742 {
625 float r; 743 float r;
626 744
627 #if ECB_STDFP 745 #if ECB_STDFP
637 x |= 0x800000U; 755 x |= 0x800000U;
638 else 756 else
639 e = 1; 757 e = 1;
640 758
641 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ 759 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
642 r = ldexpf (x * (0.5f / 0x800000U), e - 126); 760 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
643 761
644 r = neg ? -r : r; 762 r = neg ? -r : r;
645 #endif 763 #endif
646 764
647 return r; 765 return r;
648 } 766 }
649 767
650 /* convert a double to ieee double/binary64 */ 768 /* convert a double to ieee double/binary64 */
651 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 769 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
652 ecb_function_ uint64_t 770 ecb_function_ ecb_const uint64_t
653 ecb_double_to_binary64 (double x) 771 ecb_double_to_binary64 (double x)
654 { 772 {
655 uint64_t r; 773 uint64_t r;
656 774
657 #if ECB_STDFP 775 #if ECB_STDFP
686 804
687 return r; 805 return r;
688 } 806 }
689 807
690 /* converts an ieee double/binary64 to a double */ 808 /* converts an ieee double/binary64 to a double */
691 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 809 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
692 ecb_function_ double 810 ecb_function_ ecb_const double
693 ecb_binary64_to_double (uint64_t x) 811 ecb_binary64_to_double (uint64_t x)
694 { 812 {
695 double r; 813 double r;
696 814
697 #if ECB_STDFP 815 #if ECB_STDFP

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines