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

Comparing libeio/ecb.h (file contents):
Revision 1.20 by root, Fri Apr 11 05:19:41 2014 UTC vs.
Revision 1.25 by root, Thu Jun 25 15:59:57 2015 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2014 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;
… …
63 #else 74 #else
64 #define ECB_PTRSIZE 4 75 #define ECB_PTRSIZE 4
65 #endif 76 #endif
66#endif 77#endif
67 78
79#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
80#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
81
68/* work around x32 idiocy by defining proper macros */ 82/* work around x32 idiocy by defining proper macros */
69#if __amd64 || __x86_64 || _M_AMD64 || _M_X64 83#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
70 #if _ILP32 84 #if _ILP32
71 #define ECB_AMD64_X32 1 85 #define ECB_AMD64_X32 1
72 #else 86 #else
73 #define ECB_AMD64 1 87 #define ECB_AMD64 1
74 #endif 88 #endif
… …
79 * causing enormous grief in return for some better fake benchmark numbers. 93 * causing enormous grief in return for some better fake benchmark numbers.
80 * or so. 94 * or so.
81 * we try to detect these and simply assume they are not gcc - if they have 95 * 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. 96 * an issue with that they should have done it right in the first place.
83 */ 97 */
84#ifndef ECB_GCC_VERSION
85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ 98#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 99 #define ECB_GCC_VERSION(major,minor) 0
87 #else 100#else
88 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 101 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
89 #endif 102#endif
103
104#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
105
106#if __clang__ && defined __has_builtin
107 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
108#else
109 #define ECB_CLANG_BUILTIN(x) 0
110#endif
111
112#if __clang__ && defined __has_extension
113 #define ECB_CLANG_EXTENSION(x) __has_extension (x)
114#else
115 #define ECB_CLANG_EXTENSION(x) 0
90#endif 116#endif
91 117
92#define ECB_CPP (__cplusplus+0) 118#define ECB_CPP (__cplusplus+0)
93#define ECB_CPP11 (__cplusplus >= 201103L) 119#define ECB_CPP11 (__cplusplus >= 201103L)
94 120
… …
122 #define ECB_NO_SMP 1 148 #define ECB_NO_SMP 1
123#endif 149#endif
124 150
125#if ECB_NO_SMP 151#if ECB_NO_SMP
126 #define ECB_MEMORY_FENCE do { } while (0) 152 #define ECB_MEMORY_FENCE do { } while (0)
153#endif
154
155/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
156#if __xlC__ && ECB_CPP
157 #include <builtins.h>
127#endif 158#endif
128 159
129#ifndef ECB_MEMORY_FENCE 160#ifndef ECB_MEMORY_FENCE
130 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 161 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
131 #if __i386 || __i386__ 162 #if __i386 || __i386__
132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 163 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
133 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 164 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
134 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 165 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
135 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 166 #elif ECB_GCC_AMD64
136 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 167 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
137 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 168 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
138 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 169 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
139 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 170 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 171 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
… …
178 /* see comment below (stdatomic.h) about the C11 memory model. */ 209 /* see comment below (stdatomic.h) about the C11 memory model. */
179 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 210 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
180 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) 211 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
181 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) 212 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
182 213
183 /* The __has_feature syntax from clang is so misdesigned that we cannot use it 214 #elif ECB_CLANG_EXTENSION(c_atomic)
184 * without risking compile time errors with other compilers. We *could*
185 * define our own ecb_clang_has_feature, but I just can't be bothered to work
186 * around this shit time and again.
187 * #elif defined __clang && __has_feature (cxx_atomic)
188 * // see comment below (stdatomic.h) about the C11 memory model. 215 /* see comment below (stdatomic.h) about the C11 memory model. */
189 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 216 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
190 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) 217 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
191 * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) 218 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
192 */
193 219
194 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 220 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
195 #define ECB_MEMORY_FENCE __sync_synchronize () 221 #define ECB_MEMORY_FENCE __sync_synchronize ()
196 #elif _MSC_VER >= 1500 /* VC++ 2008 */ 222 #elif _MSC_VER >= 1500 /* VC++ 2008 */
197 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ 223 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
… …
260 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 286 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
261#endif 287#endif
262 288
263/*****************************************************************************/ 289/*****************************************************************************/
264 290
265#if __cplusplus 291#if ECB_CPP
266 #define ecb_inline static inline 292 #define ecb_inline static inline
267#elif ECB_GCC_VERSION(2,5) 293#elif ECB_GCC_VERSION(2,5)
268 #define ecb_inline static __inline__ 294 #define ecb_inline static __inline__
269#elif ECB_C99 295#elif ECB_C99
270 #define ecb_inline static inline 296 #define ecb_inline static inline
… …
284 310
285#define ECB_CONCAT_(a, b) a ## b 311#define ECB_CONCAT_(a, b) a ## b
286#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 312#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
287#define ECB_STRINGIFY_(a) # a 313#define ECB_STRINGIFY_(a) # a
288#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 314#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
315#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
289 316
290#define ecb_function_ ecb_inline 317#define ecb_function_ ecb_inline
291 318
292#if ECB_GCC_VERSION(3,1) 319#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
293 #define ecb_attribute(attrlist) __attribute__(attrlist) 320 #define ecb_attribute(attrlist) __attribute__ (attrlist)
321#else
322 #define ecb_attribute(attrlist)
323#endif
324
325#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
294 #define ecb_is_constant(expr) __builtin_constant_p (expr) 326 #define ecb_is_constant(expr) __builtin_constant_p (expr)
295 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
296 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
297#else 327#else
298 #define ecb_attribute(attrlist)
299
300 /* possible C11 impl for integral types 328 /* possible C11 impl for integral types
301 typedef struct ecb_is_constant_struct ecb_is_constant_struct; 329 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
302 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ 330 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
303 331
304 #define ecb_is_constant(expr) 0 332 #define ecb_is_constant(expr) 0
333#endif
334
335#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
336 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
337#else
305 #define ecb_expect(expr,value) (expr) 338 #define ecb_expect(expr,value) (expr)
339#endif
340
341#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
342 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
343#else
306 #define ecb_prefetch(addr,rw,locality) 344 #define ecb_prefetch(addr,rw,locality)
307#endif 345#endif
308 346
309/* no emulation for ecb_decltype */ 347/* no emulation for ecb_decltype */
310#if ECB_GCC_VERSION(4,5) 348#if ECB_CPP11
349 // older implementations might have problems with decltype(x)::type, work around it
350 template<class T> struct ecb_decltype_t { typedef T type; };
311 #define ecb_decltype(x) __decltype(x) 351 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
312#elif ECB_GCC_VERSION(3,0) 352#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
313 #define ecb_decltype(x) __typeof(x) 353 #define ecb_decltype(x) __typeof__ (x)
314#endif 354#endif
315 355
356#if _MSC_VER >= 1300
357 #define ecb_deprecated __declspec (deprecated)
358#else
359 #define ecb_deprecated ecb_attribute ((__deprecated__))
360#endif
361
362#if _MSC_VER >= 1500
363 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
364#elif ECB_GCC_VERSION(4,5)
365 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
366#else
367 #define ecb_deprecated_message(msg) ecb_deprecated
368#endif
369
370#if _MSC_VER >= 1400
371 #define ecb_noinline __declspec (noinline)
372#else
316#define ecb_noinline ecb_attribute ((__noinline__)) 373 #define ecb_noinline ecb_attribute ((__noinline__))
374#endif
375
317#define ecb_unused ecb_attribute ((__unused__)) 376#define ecb_unused ecb_attribute ((__unused__))
318#define ecb_const ecb_attribute ((__const__)) 377#define ecb_const ecb_attribute ((__const__))
319#define ecb_pure ecb_attribute ((__pure__)) 378#define ecb_pure ecb_attribute ((__pure__))
320 379
321#if ECB_C11 380#if ECB_C11 || __IBMC_NORETURN
381 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
322 #define ecb_noreturn _Noreturn 382 #define ecb_noreturn _Noreturn
383#elif ECB_CPP11
384 #define ecb_noreturn [[noreturn]]
385#elif _MSC_VER >= 1200
386 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
387 #define ecb_noreturn __declspec (noreturn)
323#else 388#else
324 #define ecb_noreturn ecb_attribute ((__noreturn__)) 389 #define ecb_noreturn ecb_attribute ((__noreturn__))
325#endif 390#endif
326 391
327#if ECB_GCC_VERSION(4,3) 392#if ECB_GCC_VERSION(4,3)
… …
342/* for compatibility to the rest of the world */ 407/* for compatibility to the rest of the world */
343#define ecb_likely(expr) ecb_expect_true (expr) 408#define ecb_likely(expr) ecb_expect_true (expr)
344#define ecb_unlikely(expr) ecb_expect_false (expr) 409#define ecb_unlikely(expr) ecb_expect_false (expr)
345 410
346/* count trailing zero bits and count # of one bits */ 411/* count trailing zero bits and count # of one bits */
347#if ECB_GCC_VERSION(3,4) 412#if ECB_GCC_VERSION(3,4) \
413 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
414 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
415 && ECB_CLANG_BUILTIN(__builtin_popcount))
348 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ 416 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
349 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 417 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
350 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 418 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
351 #define ecb_ctz32(x) __builtin_ctz (x) 419 #define ecb_ctz32(x) __builtin_ctz (x)
352 #define ecb_ctz64(x) __builtin_ctzll (x) 420 #define ecb_ctz64(x) __builtin_ctzll (x)
353 #define ecb_popcount32(x) __builtin_popcount (x) 421 #define ecb_popcount32(x) __builtin_popcount (x)
354 /* no popcountll */ 422 /* no popcountll */
355#else 423#else
356 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 424 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
357 ecb_function_ int 425 ecb_function_ ecb_const int
358 ecb_ctz32 (uint32_t x) 426 ecb_ctz32 (uint32_t x)
359 { 427 {
360 int r = 0; 428 int r = 0;
361 429
362 x &= ~x + 1; /* this isolates the lowest bit */ 430 x &= ~x + 1; /* this isolates the lowest bit */
… …
376#endif 444#endif
377 445
378 return r; 446 return r;
379 } 447 }
380 448
381 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 449 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
382 ecb_function_ int 450 ecb_function_ ecb_const int
383 ecb_ctz64 (uint64_t x) 451 ecb_ctz64 (uint64_t x)
384 { 452 {
385 int shift = x & 0xffffffffU ? 0 : 32; 453 int shift = x & 0xffffffffU ? 0 : 32;
386 return ecb_ctz32 (x >> shift) + shift; 454 return ecb_ctz32 (x >> shift) + shift;
387 } 455 }
388 456
389 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 457 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
390 ecb_function_ int 458 ecb_function_ ecb_const int
391 ecb_popcount32 (uint32_t x) 459 ecb_popcount32 (uint32_t x)
392 { 460 {
393 x -= (x >> 1) & 0x55555555; 461 x -= (x >> 1) & 0x55555555;
394 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 462 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
395 x = ((x >> 4) + x) & 0x0f0f0f0f; 463 x = ((x >> 4) + x) & 0x0f0f0f0f;
396 x *= 0x01010101; 464 x *= 0x01010101;
397 465
398 return x >> 24; 466 return x >> 24;
399 } 467 }
400 468
401 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 469 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
402 ecb_function_ int ecb_ld32 (uint32_t x) 470 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
403 { 471 {
404 int r = 0; 472 int r = 0;
405 473
406 if (x >> 16) { x >>= 16; r += 16; } 474 if (x >> 16) { x >>= 16; r += 16; }
407 if (x >> 8) { x >>= 8; r += 8; } 475 if (x >> 8) { x >>= 8; r += 8; }
… …
410 if (x >> 1) { r += 1; } 478 if (x >> 1) { r += 1; }
411 479
412 return r; 480 return r;
413 } 481 }
414 482
415 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 483 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
416 ecb_function_ int ecb_ld64 (uint64_t x) 484 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
417 { 485 {
418 int r = 0; 486 int r = 0;
419 487
420 if (x >> 32) { x >>= 32; r += 32; } 488 if (x >> 32) { x >>= 32; r += 32; }
421 489
422 return r + ecb_ld32 (x); 490 return r + ecb_ld32 (x);
423 } 491 }
424#endif 492#endif
425 493
426ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 494ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
427ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 495ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
428ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 496ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
429ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 497ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
430 498
431ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 499ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
432ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 500ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
433{ 501{
434 return ( (x * 0x0802U & 0x22110U) 502 return ( (x * 0x0802U & 0x22110U)
435 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 503 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
436} 504}
437 505
438ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 506ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
439ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 507ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
440{ 508{
441 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 509 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
442 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 510 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
443 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 511 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
444 x = ( x >> 8 ) | ( x << 8); 512 x = ( x >> 8 ) | ( x << 8);
445 513
446 return x; 514 return x;
447} 515}
448 516
449ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 517ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
450ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 518ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
451{ 519{
452 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 520 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
453 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 521 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
454 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 522 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
455 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 523 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
… …
458 return x; 526 return x;
459} 527}
460 528
461/* popcount64 is only available on 64 bit cpus as gcc builtin */ 529/* popcount64 is only available on 64 bit cpus as gcc builtin */
462/* so for this version we are lazy */ 530/* so for this version we are lazy */
463ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 531ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
464ecb_function_ int 532ecb_function_ ecb_const int
465ecb_popcount64 (uint64_t x) 533ecb_popcount64 (uint64_t x)
466{ 534{
467 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 535 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
468} 536}
469 537
470ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 538ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
471ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 539ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
472ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 540ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
473ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 541ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
474ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 542ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
475ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 543ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
476ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 544ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
477ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 545ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
478 546
479ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 547ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
480ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 548ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
481ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 549ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
482ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 550ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
483ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 551ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
484ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 552ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
485ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 553ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
486ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 554ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
487 555
488#if ECB_GCC_VERSION(4,3) 556#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
557 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
558 #define ecb_bswap16(x) __builtin_bswap16 (x)
559 #else
489 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 560 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
561 #endif
490 #define ecb_bswap32(x) __builtin_bswap32 (x) 562 #define ecb_bswap32(x) __builtin_bswap32 (x)
491 #define ecb_bswap64(x) __builtin_bswap64 (x) 563 #define ecb_bswap64(x) __builtin_bswap64 (x)
564#elif _MSC_VER
565 #include <stdlib.h>
566 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
567 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
568 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
492#else 569#else
493 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 570 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
494 ecb_function_ uint16_t 571 ecb_function_ ecb_const uint16_t
495 ecb_bswap16 (uint16_t x) 572 ecb_bswap16 (uint16_t x)
496 { 573 {
497 return ecb_rotl16 (x, 8); 574 return ecb_rotl16 (x, 8);
498 } 575 }
499 576
500 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 577 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
501 ecb_function_ uint32_t 578 ecb_function_ ecb_const uint32_t
502 ecb_bswap32 (uint32_t x) 579 ecb_bswap32 (uint32_t x)
503 { 580 {
504 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 581 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
505 } 582 }
506 583
507 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 584 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
508 ecb_function_ uint64_t 585 ecb_function_ ecb_const uint64_t
509 ecb_bswap64 (uint64_t x) 586 ecb_bswap64 (uint64_t x)
510 { 587 {
511 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 588 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
512 } 589 }
513#endif 590#endif
514 591
515#if ECB_GCC_VERSION(4,5) 592#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
516 #define ecb_unreachable() __builtin_unreachable () 593 #define ecb_unreachable() __builtin_unreachable ()
517#else 594#else
518 /* this seems to work fine, but gcc always emits a warning for it :/ */ 595 /* this seems to work fine, but gcc always emits a warning for it :/ */
519 ecb_inline void ecb_unreachable (void) ecb_noreturn; 596 ecb_inline ecb_noreturn void ecb_unreachable (void);
520 ecb_inline void ecb_unreachable (void) { } 597 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
521#endif 598#endif
522 599
523/* try to tell the compiler that some condition is definitely true */ 600/* try to tell the compiler that some condition is definitely true */
524#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 601#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
525 602
526ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 603ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);
527ecb_inline unsigned char 604ecb_inline ecb_const unsigned char
528ecb_byteorder_helper (void) 605ecb_byteorder_helper (void)
529{ 606{
530 /* the union code still generates code under pressure in gcc, */ 607 /* the union code still generates code under pressure in gcc, */
531 /* but less than using pointers, and always seems to */ 608 /* but less than using pointers, and always seems to */
532 /* successfully return a constant. */ 609 /* successfully return a constant. */
533 /* the reason why we have this horrible preprocessor mess */ 610 /* the reason why we have this horrible preprocessor mess */
534 /* is to avoid it in all cases, at least on common architectures */ 611 /* is to avoid it in all cases, at least on common architectures */
535 /* or when using a recent enough gcc version (>= 4.6) */ 612 /* or when using a recent enough gcc version (>= 4.6) */
536#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 613#if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64
537 return 0x44; 614 return 0x44;
538#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 615#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
539 return 0x44; 616 return 0x44;
540#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 617#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
541 return 0x11; 618 return 0x11;
… …
547 } u = { 0x11223344 }; 624 } u = { 0x11223344 };
548 return u.c; 625 return u.c;
549#endif 626#endif
550} 627}
551 628
552ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 629ecb_inline ecb_const ecb_bool ecb_big_endian (void);
553ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 630ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
554ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 631ecb_inline ecb_const ecb_bool ecb_little_endian (void);
555ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 632ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
556 633
557#if ECB_GCC_VERSION(3,0) || ECB_C99 634#if ECB_GCC_VERSION(3,0) || ECB_C99
558 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 635 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
559#else 636#else
560 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 637 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
561#endif 638#endif
562 639
563#if __cplusplus 640#if ECB_CPP
564 template<typename T> 641 template<typename T>
565 static inline T ecb_div_rd (T val, T div) 642 static inline T ecb_div_rd (T val, T div)
566 { 643 {
567 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 644 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
568 } 645 }
… …
592 669
593/* basically, everything uses "ieee pure-endian" floating point numbers */ 670/* basically, everything uses "ieee pure-endian" floating point numbers */
594/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 671/* the only noteworthy exception is ancient armle, which uses order 43218765 */
595#if 0 \ 672#if 0 \
596 || __i386 || __i386__ \ 673 || __i386 || __i386__ \
597 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 674 || ECB_GCC_AMD64 \
598 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 675 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
599 || defined __s390__ || defined __s390x__ \ 676 || defined __s390__ || defined __s390x__ \
600 || defined __mips__ \ 677 || defined __mips__ \
601 || defined __alpha__ \ 678 || defined __alpha__ \
602 || defined __hppa__ \ 679 || defined __hppa__ \
603 || defined __ia64__ \ 680 || defined __ia64__ \
604 || defined __m68k__ \ 681 || defined __m68k__ \
605 || defined __m88k__ \ 682 || defined __m88k__ \
606 || defined __sh__ \ 683 || defined __sh__ \
607 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ 684 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
608 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ 685 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
609 || defined __aarch64__ 686 || defined __aarch64__
610 #define ECB_STDFP 1 687 #define ECB_STDFP 1
611 #include <string.h> /* for memcpy */ 688 #include <string.h> /* for memcpy */
612#else 689#else
… …
628 #define ECB_NAN NAN 705 #define ECB_NAN NAN
629 #else 706 #else
630 #define ECB_NAN ECB_INFINITY 707 #define ECB_NAN ECB_INFINITY
631 #endif 708 #endif
632 709
710 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
711 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
712 #define ecb_frexpf(x,e) frexpf ((x), (e))
713 #else
714 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
715 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
716 #endif
717
633 /* converts an ieee half/binary16 to a float */ 718 /* converts an ieee half/binary16 to a float */
634 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; 719 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
635 ecb_function_ float 720 ecb_function_ ecb_const float
636 ecb_binary16_to_float (uint16_t x) 721 ecb_binary16_to_float (uint16_t x)
637 { 722 {
638 int e = (x >> 10) & 0x1f; 723 int e = (x >> 10) & 0x1f;
639 int m = x & 0x3ff; 724 int m = x & 0x3ff;
640 float r; 725 float r;
641 726
642 if (!e ) r = ldexpf (m , -24); 727 if (!e ) r = ecb_ldexpf (m , -24);
643 else if (e != 31) r = ldexpf (m + 0x400, e - 25); 728 else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
644 else if (m ) r = ECB_NAN; 729 else if (m ) r = ECB_NAN;
645 else r = ECB_INFINITY; 730 else r = ECB_INFINITY;
646 731
647 return x & 0x8000 ? -r : r; 732 return x & 0x8000 ? -r : r;
648 } 733 }
649 734
650 /* convert a float to ieee single/binary32 */ 735 /* convert a float to ieee single/binary32 */
651 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 736 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
652 ecb_function_ uint32_t 737 ecb_function_ ecb_const uint32_t
653 ecb_float_to_binary32 (float x) 738 ecb_float_to_binary32 (float x)
654 { 739 {
655 uint32_t r; 740 uint32_t r;
656 741
657 #if ECB_STDFP 742 #if ECB_STDFP
… …
664 if (x == 0e0f ) return 0x00000000U; 749 if (x == 0e0f ) return 0x00000000U;
665 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 750 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
666 if (x < -3.40282346638528860e+38f) return 0xff800000U; 751 if (x < -3.40282346638528860e+38f) return 0xff800000U;
667 if (x != x ) return 0x7fbfffffU; 752 if (x != x ) return 0x7fbfffffU;
668 753
669 m = frexpf (x, &e) * 0x1000000U; 754 m = ecb_frexpf (x, &e) * 0x1000000U;
670 755
671 r = m & 0x80000000U; 756 r = m & 0x80000000U;
672 757
673 if (r) 758 if (r)
674 m = -m; 759 m = -m;
… …
686 771
687 return r; 772 return r;
688 } 773 }
689 774
690 /* converts an ieee single/binary32 to a float */ 775 /* converts an ieee single/binary32 to a float */
691 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 776 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
692 ecb_function_ float 777 ecb_function_ ecb_const float
693 ecb_binary32_to_float (uint32_t x) 778 ecb_binary32_to_float (uint32_t x)
694 { 779 {
695 float r; 780 float r;
696 781
697 #if ECB_STDFP 782 #if ECB_STDFP
… …
707 x |= 0x800000U; 792 x |= 0x800000U;
708 else 793 else
709 e = 1; 794 e = 1;
710 795
711 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ 796 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
712 r = ldexpf (x * (0.5f / 0x800000U), e - 126); 797 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
713 798
714 r = neg ? -r : r; 799 r = neg ? -r : r;
715 #endif 800 #endif
716 801
717 return r; 802 return r;
718 } 803 }
719 804
720 /* convert a double to ieee double/binary64 */ 805 /* convert a double to ieee double/binary64 */
721 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 806 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
722 ecb_function_ uint64_t 807 ecb_function_ ecb_const uint64_t
723 ecb_double_to_binary64 (double x) 808 ecb_double_to_binary64 (double x)
724 { 809 {
725 uint64_t r; 810 uint64_t r;
726 811
727 #if ECB_STDFP 812 #if ECB_STDFP
… …
756 841
757 return r; 842 return r;
758 } 843 }
759 844
760 /* converts an ieee double/binary64 to a double */ 845 /* converts an ieee double/binary64 to a double */
761 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 846 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
762 ecb_function_ double 847 ecb_function_ ecb_const double
763 ecb_binary64_to_double (uint64_t x) 848 ecb_binary64_to_double (uint64_t x)
764 { 849 {
765 double r; 850 double r;
766 851
767 #if ECB_STDFP 852 #if ECB_STDFP

Diff Legend

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