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

Comparing libeio/ecb.h (file contents):
Revision 1.5 by root, Sun Jul 24 03:32:54 2011 UTC vs.
Revision 1.17 by root, Fri Dec 28 07:33:41 2012 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2011 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2012 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:
… …
28 */ 28 */
29 29
30#ifndef ECB_H 30#ifndef ECB_H
31#define ECB_H 31#define ECB_H
32 32
33/* 16 bits major, 16 bits minor */
34#define ECB_VERSION 0x00010003
35
33#ifdef _WIN32 36#ifdef _WIN32
34 typedef signed char int8_t; 37 typedef signed char int8_t;
35 typedef unsigned char uint8_t; 38 typedef unsigned char uint8_t;
36 typedef signed short int16_t; 39 typedef signed short int16_t;
37 typedef unsigned short uint16_t; 40 typedef unsigned short uint16_t;
… …
42 typedef unsigned long long uint64_t; 45 typedef unsigned long long uint64_t;
43 #else /* _MSC_VER || __BORLANDC__ */ 46 #else /* _MSC_VER || __BORLANDC__ */
44 typedef signed __int64 int64_t; 47 typedef signed __int64 int64_t;
45 typedef unsigned __int64 uint64_t; 48 typedef unsigned __int64 uint64_t;
46 #endif 49 #endif
50 #ifdef _WIN64
51 #define ECB_PTRSIZE 8
52 typedef uint64_t uintptr_t;
53 typedef int64_t intptr_t;
54 #else
55 #define ECB_PTRSIZE 4
56 typedef uint32_t uintptr_t;
57 typedef int32_t intptr_t;
58 #endif
47#else 59#else
48 #include <inttypes.h> 60 #include <inttypes.h>
61 #if UINTMAX_MAX > 0xffffffffU
62 #define ECB_PTRSIZE 8
63 #else
64 #define ECB_PTRSIZE 4
65 #endif
66#endif
67
68/* work around x32 idiocy by defining proper macros */
69#if __x86_64 || _M_AMD64
70 #if __ILP32
71 #define ECB_AMD64_X32 1
72 #else
73 #define ECB_AMD64 1
74 #endif
49#endif 75#endif
50 76
51/* many compilers define _GNUC_ to some versions but then only implement 77/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions, 78 * what their idiot authors think are the "more important" extensions,
53 * causing enourmous grief in return for some better fake benchmark numbers. 79 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so. 80 * or so.
55 * we try to detect these and simply assume they are not gcc - if they have 81 * we try to detect these and simply assume they are not gcc - if they have
56 * an issue with that they should have done it right in the first place. 82 * an issue with that they should have done it right in the first place.
57 */ 83 */
58#ifndef ECB_GCC_VERSION 84#ifndef ECB_GCC_VERSION
59 #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) 85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
60 #define ECB_GCC_VERSION(major,minor) 0 86 #define ECB_GCC_VERSION(major,minor) 0
61 #else 87 #else
62 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 88 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
63 #endif 89 #endif
64#endif 90#endif
65 91
92#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
93#define ECB_C99 (__STDC_VERSION__ >= 199901L)
94#define ECB_C11 (__STDC_VERSION__ >= 201112L)
95#define ECB_CPP (__cplusplus+0)
96#define ECB_CPP11 (__cplusplus >= 201103L)
97
98#if ECB_CPP
99 #define ECB_EXTERN_C extern "C"
100 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
101 #define ECB_EXTERN_C_END }
102#else
103 #define ECB_EXTERN_C extern
104 #define ECB_EXTERN_C_BEG
105 #define ECB_EXTERN_C_END
106#endif
107
66/*****************************************************************************/ 108/*****************************************************************************/
67 109
110/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
111/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
112
113#if ECB_NO_THREADS
114 #define ECB_NO_SMP 1
115#endif
116
117#if ECB_NO_SMP
118 #define ECB_MEMORY_FENCE do { } while (0)
119#endif
120
68#ifndef ECB_MEMORY_FENCE 121#ifndef ECB_MEMORY_FENCE
69 #if ECB_GCC_VERSION(2,5) 122 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
70 #if __x86 123 #if __i386 || __i386__
71 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 124 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
72 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 125 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
73 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE /* better be safe than sorry */ 126 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
74 #elif __amd64 127 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
75 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 128 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
76 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 129 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
77 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") 130 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
131 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
133 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
134 || 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")
136 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
137 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
138 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
139 #elif __sparc || __sparc__
140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
141 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
143 #elif defined __s390__ || defined __s390x__
144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
145 #elif defined __mips__
146 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
147 #elif defined __alpha__
148 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
149 #elif defined __hppa__
150 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
151 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
152 #elif defined __ia64__
153 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
78 #endif 154 #endif
79 #endif 155 #endif
80#endif 156#endif
81 157
82#ifndef ECB_MEMORY_FENCE 158#ifndef ECB_MEMORY_FENCE
83 #if ECB_GCC_VERSION(4,4) 159 #if ECB_GCC_VERSION(4,7)
160 /* see comment below (stdatomic.h) about the C11 memory model. */
161 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
162
163 /* The __has_feature syntax from clang is so misdesigned that we cannot use it
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.
169 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
170 */
171
172 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
84 #define ECB_MEMORY_FENCE __sync_synchronize () 173 #define ECB_MEMORY_FENCE __sync_synchronize ()
85 #define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) 174 #elif _MSC_VER >= 1400 /* VC++ 2005 */
86 #define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) 175 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
87 #elif _MSC_VER >= 1400 && 0 /* TODO: only true when using volatiles */
88 #define ECB_MEMORY_FENCE do { } while (0)
89 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
90 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
91 #elif defined(_WIN32) && defined(MemoryBarrier)
92 #define ECB_MEMORY_FENCE MemoryBarrier () 176 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
93 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 177 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
94 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 178 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
179 #elif defined _WIN32
180 #include <WinNT.h>
181 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
182 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
183 #include <mbarrier.h>
184 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
185 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
186 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
187 #elif __xlC__
188 #define ECB_MEMORY_FENCE __sync ()
95 #endif 189 #endif
96#endif 190#endif
97 191
98#ifndef ECB_MEMORY_FENCE 192#ifndef ECB_MEMORY_FENCE
193 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
194 /* we assume that these memory fences work on all variables/all memory accesses, */
195 /* not just C11 atomics and atomic accesses */
196 #include <stdatomic.h>
197 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
198 /* 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 */
200 /* for most usages, or gcc and clang have a bug */
201 /* I *currently* lean towards the latter, and inefficiently implement */
202 /* all three of ecb's fences as a seq_cst fence */
203 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
204 #endif
205#endif
206
207#ifndef ECB_MEMORY_FENCE
208 #if !ECB_AVOID_PTHREADS
209 /*
210 * if you get undefined symbol references to pthread_mutex_lock,
211 * or failure to find pthread.h, then you should implement
212 * the ECB_MEMORY_FENCE operations for your cpu/compiler
213 * OR provide pthread.h and link against the posix thread library
214 * of your system.
215 */
99 #include <pthread.h> 216 #include <pthread.h>
217 #define ECB_NEEDS_PTHREADS 1
218 #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
100 219
101 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 220 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
102 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) 221 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
222 #endif
223#endif
224
225#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
103 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 226 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
227#endif
228
229#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
104 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 230 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
105#endif 231#endif
106 232
107/*****************************************************************************/ 233/*****************************************************************************/
108
109#define ECB_C99 (__STDC_VERSION__ >= 199901L)
110 234
111#if __cplusplus 235#if __cplusplus
112 #define ecb_inline static inline 236 #define ecb_inline static inline
113#elif ECB_GCC_VERSION(2,5) 237#elif ECB_GCC_VERSION(2,5)
114 #define ecb_inline static __inline__ 238 #define ecb_inline static __inline__
… …
153#elif ECB_GCC_VERSION(3,0) 277#elif ECB_GCC_VERSION(3,0)
154 #define ecb_decltype(x) __typeof(x) 278 #define ecb_decltype(x) __typeof(x)
155#endif 279#endif
156 280
157#define ecb_noinline ecb_attribute ((__noinline__)) 281#define ecb_noinline ecb_attribute ((__noinline__))
158#define ecb_noreturn ecb_attribute ((__noreturn__))
159#define ecb_unused ecb_attribute ((__unused__)) 282#define ecb_unused ecb_attribute ((__unused__))
160#define ecb_const ecb_attribute ((__const__)) 283#define ecb_const ecb_attribute ((__const__))
161#define ecb_pure ecb_attribute ((__pure__)) 284#define ecb_pure ecb_attribute ((__pure__))
285
286#if ECB_C11
287 #define ecb_noreturn _Noreturn
288#else
289 #define ecb_noreturn ecb_attribute ((__noreturn__))
290#endif
162 291
163#if ECB_GCC_VERSION(4,3) 292#if ECB_GCC_VERSION(4,3)
164 #define ecb_artificial ecb_attribute ((__artificial__)) 293 #define ecb_artificial ecb_attribute ((__artificial__))
165 #define ecb_hot ecb_attribute ((__hot__)) 294 #define ecb_hot ecb_attribute ((__hot__))
166 #define ecb_cold ecb_attribute ((__cold__)) 295 #define ecb_cold ecb_attribute ((__cold__))
… …
232 x *= 0x01010101; 361 x *= 0x01010101;
233 362
234 return x >> 24; 363 return x >> 24;
235 } 364 }
236 365
237 /* you have the choice beetween something with a table lookup, */
238 /* something using lots of bit arithmetic and a simple loop */
239 /* we went for the loop */
240 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 366 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
241 ecb_function_ int ecb_ld32 (uint32_t x) 367 ecb_function_ int ecb_ld32 (uint32_t x)
242 { 368 {
243 int r = 0; 369 int r = 0;
244 370
… …
259 if (x >> 32) { x >>= 32; r += 32; } 385 if (x >> 32) { x >>= 32; r += 32; }
260 386
261 return r + ecb_ld32 (x); 387 return r + ecb_ld32 (x);
262 } 388 }
263#endif 389#endif
390
391ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
392ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
393ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
394ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
395
396ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
397ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
398{
399 return ( (x * 0x0802U & 0x22110U)
400 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
401}
402
403ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
404ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
405{
406 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
407 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
408 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
409 x = ( x >> 8 ) | ( x << 8);
410
411 return x;
412}
413
414ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
415ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
416{
417 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
418 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
419 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
420 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
421 x = ( x >> 16 ) | ( x << 16);
422
423 return x;
424}
264 425
265/* popcount64 is only available on 64 bit cpus as gcc builtin */ 426/* popcount64 is only available on 64 bit cpus as gcc builtin */
266/* so for this version we are lazy */ 427/* so for this version we are lazy */
267ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 428ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
268ecb_function_ int 429ecb_function_ int
… …
318 479
319#if ECB_GCC_VERSION(4,5) 480#if ECB_GCC_VERSION(4,5)
320 #define ecb_unreachable() __builtin_unreachable () 481 #define ecb_unreachable() __builtin_unreachable ()
321#else 482#else
322 /* this seems to work fine, but gcc always emits a warning for it :/ */ 483 /* this seems to work fine, but gcc always emits a warning for it :/ */
323 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 484 ecb_inline void ecb_unreachable (void) ecb_noreturn;
324 ecb_function_ void ecb_unreachable (void) { } 485 ecb_inline void ecb_unreachable (void) { }
325#endif 486#endif
326 487
327/* try to tell the compiler that some condition is definitely true */ 488/* try to tell the compiler that some condition is definitely true */
328#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 489#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
329 490
330ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 491ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
331ecb_function_ unsigned char 492ecb_inline unsigned char
332ecb_byteorder_helper (void) 493ecb_byteorder_helper (void)
333{ 494{
334 const uint32_t u = 0x11223344; 495 /* the union code still generates code under pressure in gcc, */
335 return *(unsigned char *)&u; 496 /* but less than using pointers, and always seems to */
497 /* successfully return a constant. */
498 /* the reason why we have this horrible preprocessor mess */
499 /* is to avoid it in all cases, at least on common architectures */
500 /* or when using a recent enough gcc version (>= 4.6) */
501#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
502 return 0x44;
503#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
504 return 0x44;
505#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
506 return 0x11;
507#else
508 union
509 {
510 uint32_t i;
511 uint8_t c;
512 } u = { 0x11223344 };
513 return u.c;
514#endif
336} 515}
337 516
338ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 517ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
339ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 518ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
340ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 519ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
341ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 520ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
342 521
343#if ECB_GCC_VERSION(3,0) || ECB_C99 522#if ECB_GCC_VERSION(3,0) || ECB_C99
344 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 523 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
345#else 524#else
346 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 525 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
526#endif
527
528#if __cplusplus
529 template<typename T>
530 static inline T ecb_div_rd (T val, T div)
531 {
532 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
533 }
534 template<typename T>
535 static inline T ecb_div_ru (T val, T div)
536 {
537 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
538 }
539#else
540 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
541 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
347#endif 542#endif
348 543
349#if ecb_cplusplus_does_not_suck 544#if ecb_cplusplus_does_not_suck
350 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 545 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
351 template<typename T, int N> 546 template<typename T, int N>
… …
355 } 550 }
356#else 551#else
357 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 552 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
358#endif 553#endif
359 554
555/*******************************************************************************/
556/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
557
558/* basically, everything uses "ieee pure-endian" floating point numbers */
559/* the only noteworthy exception is ancient armle, which uses order 43218765 */
560#if 0 \
561 || __i386 || __i386__ \
562 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
563 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
564 || defined __arm__ && defined __ARM_EABI__ \
565 || defined __s390__ || defined __s390x__ \
566 || defined __mips__ \
567 || defined __alpha__ \
568 || defined __hppa__ \
569 || defined __ia64__ \
570 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
571 #define ECB_STDFP 1
572 #include <string.h> /* for memcpy */
573#else
574 #define ECB_STDFP 0
575 #include <math.h> /* for frexp*, ldexp* */
576#endif
577
578#ifndef ECB_NO_LIBM
579
580 /* convert a float to ieee single/binary32 */
581 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
582 ecb_function_ uint32_t
583 ecb_float_to_binary32 (float x)
584 {
585 uint32_t r;
586
587 #if ECB_STDFP
588 memcpy (&r, &x, 4);
589 #else
590 /* slow emulation, works for anything but -0 */
591 uint32_t m;
592 int e;
593
594 if (x == 0e0f ) return 0x00000000U;
595 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
596 if (x < -3.40282346638528860e+38f) return 0xff800000U;
597 if (x != x ) return 0x7fbfffffU;
598
599 m = frexpf (x, &e) * 0x1000000U;
600
601 r = m & 0x80000000U;
602
603 if (r)
604 m = -m;
605
606 if (e <= -126)
607 {
608 m &= 0xffffffU;
609 m >>= (-125 - e);
610 e = -126;
611 }
612
613 r |= (e + 126) << 23;
614 r |= m & 0x7fffffU;
360#endif 615 #endif
361 616
617 return r;
618 }
619
620 /* converts an ieee single/binary32 to a float */
621 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
622 ecb_function_ float
623 ecb_binary32_to_float (uint32_t x)
624 {
625 float r;
626
627 #if ECB_STDFP
628 memcpy (&r, &x, 4);
629 #else
630 /* emulation, only works for normals and subnormals and +0 */
631 int neg = x >> 31;
632 int e = (x >> 23) & 0xffU;
633
634 x &= 0x7fffffU;
635
636 if (e)
637 x |= 0x800000U;
638 else
639 e = 1;
640
641 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
642 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
643
644 r = neg ? -r : r;
645 #endif
646
647 return r;
648 }
649
650 /* convert a double to ieee double/binary64 */
651 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
652 ecb_function_ uint64_t
653 ecb_double_to_binary64 (double x)
654 {
655 uint64_t r;
656
657 #if ECB_STDFP
658 memcpy (&r, &x, 8);
659 #else
660 /* slow emulation, works for anything but -0 */
661 uint64_t m;
662 int e;
663
664 if (x == 0e0 ) return 0x0000000000000000U;
665 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
666 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
667 if (x != x ) return 0X7ff7ffffffffffffU;
668
669 m = frexp (x, &e) * 0x20000000000000U;
670
671 r = m & 0x8000000000000000;;
672
673 if (r)
674 m = -m;
675
676 if (e <= -1022)
677 {
678 m &= 0x1fffffffffffffU;
679 m >>= (-1021 - e);
680 e = -1022;
681 }
682
683 r |= ((uint64_t)(e + 1022)) << 52;
684 r |= m & 0xfffffffffffffU;
685 #endif
686
687 return r;
688 }
689
690 /* converts an ieee double/binary64 to a double */
691 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
692 ecb_function_ double
693 ecb_binary64_to_double (uint64_t x)
694 {
695 double r;
696
697 #if ECB_STDFP
698 memcpy (&r, &x, 8);
699 #else
700 /* emulation, only works for normals and subnormals and +0 */
701 int neg = x >> 63;
702 int e = (x >> 52) & 0x7ffU;
703
704 x &= 0xfffffffffffffU;
705
706 if (e)
707 x |= 0x10000000000000U;
708 else
709 e = 1;
710
711 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
712 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
713
714 r = neg ? -r : r;
715 #endif
716
717 return r;
718 }
719
720#endif
721
722#endif
723

Diff Legend

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