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

Comparing libeio/ecb.h (file contents):
Revision 1.12 by root, Mon May 28 17:00:19 2012 UTC vs.
Revision 1.18 by root, Sat Sep 7 23:18:23 2013 UTC

… …
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 */ 33/* 16 bits major, 16 bits minor */
34#define ECB_VERSION 0x00010001 34#define ECB_VERSION 0x00010003
35 35
36#ifdef _WIN32 36#ifdef _WIN32
37 typedef signed char int8_t; 37 typedef signed char int8_t;
38 typedef unsigned char uint8_t; 38 typedef unsigned char uint8_t;
39 typedef signed short int16_t; 39 typedef signed short int16_t;
… …
54 #else 54 #else
55 #define ECB_PTRSIZE 4 55 #define ECB_PTRSIZE 4
56 typedef uint32_t uintptr_t; 56 typedef uint32_t uintptr_t;
57 typedef int32_t intptr_t; 57 typedef int32_t intptr_t;
58 #endif 58 #endif
59 typedef intptr_t ptrdiff_t;
60#else 59#else
61 #include <inttypes.h> 60 #include <inttypes.h>
62 #if UINTMAX_MAX > 0xffffffffU 61 #if UINTMAX_MAX > 0xffffffffU
63 #define ECB_PTRSIZE 8 62 #define ECB_PTRSIZE 8
64 #else 63 #else
65 #define ECB_PTRSIZE 4 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
66 #endif 74 #endif
67#endif 75#endif
68 76
69/* many compilers define _GNUC_ to some versions but then only implement 77/* many compilers define _GNUC_ to some versions but then only implement
70 * what their idiot authors think are the "more important" extensions, 78 * what their idiot authors think are the "more important" extensions,
… …
79 #else 87 #else
80 #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)))
81 #endif 89 #endif
82#endif 90#endif
83 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
84/*****************************************************************************/ 108/*****************************************************************************/
85 109
86/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 110/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
87/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ 111/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
88 112
89#if ECB_NO_THREADS 113#if ECB_NO_THREADS
90# define ECB_NO_SMP 1 114 #define ECB_NO_SMP 1
91#endif 115#endif
92 116
93#if ECB_NO_THREADS || ECB_NO_SMP 117#if ECB_NO_SMP
94 #define ECB_MEMORY_FENCE do { } while (0) 118 #define ECB_MEMORY_FENCE do { } while (0)
95#endif 119#endif
96 120
97#ifndef ECB_MEMORY_FENCE 121#ifndef ECB_MEMORY_FENCE
98 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 122 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
99 #if __i386 || __i386__ 123 #if __i386 || __i386__
100 #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")
101 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ 125 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
102 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ 126 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
103 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 127 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
104 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 128 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
105 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 129 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
106 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ 130 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
107 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 131 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
108 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
109 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 133 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
110 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 134 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
111 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
112 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 136 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
113 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 137 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
114 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 138 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
115 #elif __sparc || __sparc__ 139 #elif __sparc || __sparc__
116 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") 140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
117 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 141 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
118 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
119 #elif defined __s390__ || defined __s390x__ 143 #elif defined __s390__ || defined __s390x__
120 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
121 #elif defined __mips__ 145 #elif defined __mips__
146 /* GNU/Linux emulates sync on mips1 architectures, so we force it's use */
147 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
122 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 148 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
123 #elif defined __alpha__ 149 #elif defined __alpha__
124 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") 150 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
151 #elif defined __hppa__
152 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
153 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
154 #elif defined __ia64__
155 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
125 #endif 156 #endif
126 #endif 157 #endif
127#endif 158#endif
128 159
129#ifndef ECB_MEMORY_FENCE 160#ifndef ECB_MEMORY_FENCE
161 #if ECB_GCC_VERSION(4,7)
162 /* see comment below (stdatomic.h) about the C11 memory model. */
163 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
164
165 /* The __has_feature syntax from clang is so misdesigned that we cannot use it
166 * without risking compile time errors with other compilers. We *could*
167 * define our own ecb_clang_has_feature, but I just can't be bothered to work
168 * around this shit time and again.
169 * #elif defined __clang && __has_feature (cxx_atomic)
170 * // see comment below (stdatomic.h) about the C11 memory model.
171 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
172 */
173
130 #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 174 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
131 #define ECB_MEMORY_FENCE __sync_synchronize () 175 #define ECB_MEMORY_FENCE __sync_synchronize ()
132 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
133 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
134 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 176 #elif _MSC_VER >= 1400 /* VC++ 2005 */
135 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 177 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
136 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 178 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
137 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 179 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
138 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 180 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
… …
148 #define ECB_MEMORY_FENCE __sync () 190 #define ECB_MEMORY_FENCE __sync ()
149 #endif 191 #endif
150#endif 192#endif
151 193
152#ifndef ECB_MEMORY_FENCE 194#ifndef ECB_MEMORY_FENCE
195 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
196 /* we assume that these memory fences work on all variables/all memory accesses, */
197 /* not just C11 atomics and atomic accesses */
198 #include <stdatomic.h>
199 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
200 /* any fence other than seq_cst, which isn't very efficient for us. */
201 /* Why that is, we don't know - either the C11 memory model is quite useless */
202 /* for most usages, or gcc and clang have a bug */
203 /* I *currently* lean towards the latter, and inefficiently implement */
204 /* all three of ecb's fences as a seq_cst fence */
205 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
206 #endif
207#endif
208
209#ifndef ECB_MEMORY_FENCE
153 #if !ECB_AVOID_PTHREADS 210 #if !ECB_AVOID_PTHREADS
154 /* 211 /*
155 * if you get undefined symbol references to pthread_mutex_lock, 212 * if you get undefined symbol references to pthread_mutex_lock,
156 * or failure to find pthread.h, then you should implement 213 * or failure to find pthread.h, then you should implement
157 * the ECB_MEMORY_FENCE operations for your cpu/compiler 214 * the ECB_MEMORY_FENCE operations for your cpu/compiler
… …
175 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 232 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
176#endif 233#endif
177 234
178/*****************************************************************************/ 235/*****************************************************************************/
179 236
180#define ECB_C99 (__STDC_VERSION__ >= 199901L)
181
182#if __cplusplus 237#if __cplusplus
183 #define ecb_inline static inline 238 #define ecb_inline static inline
184#elif ECB_GCC_VERSION(2,5) 239#elif ECB_GCC_VERSION(2,5)
185 #define ecb_inline static __inline__ 240 #define ecb_inline static __inline__
186#elif ECB_C99 241#elif ECB_C99
… …
224#elif ECB_GCC_VERSION(3,0) 279#elif ECB_GCC_VERSION(3,0)
225 #define ecb_decltype(x) __typeof(x) 280 #define ecb_decltype(x) __typeof(x)
226#endif 281#endif
227 282
228#define ecb_noinline ecb_attribute ((__noinline__)) 283#define ecb_noinline ecb_attribute ((__noinline__))
229#define ecb_noreturn ecb_attribute ((__noreturn__))
230#define ecb_unused ecb_attribute ((__unused__)) 284#define ecb_unused ecb_attribute ((__unused__))
231#define ecb_const ecb_attribute ((__const__)) 285#define ecb_const ecb_attribute ((__const__))
232#define ecb_pure ecb_attribute ((__pure__)) 286#define ecb_pure ecb_attribute ((__pure__))
287
288#if ECB_C11
289 #define ecb_noreturn _Noreturn
290#else
291 #define ecb_noreturn ecb_attribute ((__noreturn__))
292#endif
233 293
234#if ECB_GCC_VERSION(4,3) 294#if ECB_GCC_VERSION(4,3)
235 #define ecb_artificial ecb_attribute ((__artificial__)) 295 #define ecb_artificial ecb_attribute ((__artificial__))
236 #define ecb_hot ecb_attribute ((__hot__)) 296 #define ecb_hot ecb_attribute ((__hot__))
237 #define ecb_cold ecb_attribute ((__cold__)) 297 #define ecb_cold ecb_attribute ((__cold__))
… …
426 ecb_inline void ecb_unreachable (void) ecb_noreturn; 486 ecb_inline void ecb_unreachable (void) ecb_noreturn;
427 ecb_inline void ecb_unreachable (void) { } 487 ecb_inline void ecb_unreachable (void) { }
428#endif 488#endif
429 489
430/* try to tell the compiler that some condition is definitely true */ 490/* try to tell the compiler that some condition is definitely true */
431#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 491#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
432 492
433ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 493ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
434ecb_inline unsigned char 494ecb_inline unsigned char
435ecb_byteorder_helper (void) 495ecb_byteorder_helper (void)
436{ 496{
437 const uint32_t u = 0x11223344; 497 /* the union code still generates code under pressure in gcc, */
438 return *(unsigned char *)&u; 498 /* but less than using pointers, and always seems to */
499 /* successfully return a constant. */
500 /* the reason why we have this horrible preprocessor mess */
501 /* is to avoid it in all cases, at least on common architectures */
502 /* or when using a recent enough gcc version (>= 4.6) */
503#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
504 return 0x44;
505#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
506 return 0x44;
507#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
508 return 0x11;
509#else
510 union
511 {
512 uint32_t i;
513 uint8_t c;
514 } u = { 0x11223344 };
515 return u.c;
516#endif
439} 517}
440 518
441ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 519ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
442ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 520ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
443ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 521ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
… …
474 } 552 }
475#else 553#else
476 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 554 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
477#endif 555#endif
478 556
557/*******************************************************************************/
558/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
559
560/* basically, everything uses "ieee pure-endian" floating point numbers */
561/* the only noteworthy exception is ancient armle, which uses order 43218765 */
562#if 0 \
563 || __i386 || __i386__ \
564 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
565 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
566 || defined __arm__ && defined __ARM_EABI__ \
567 || defined __s390__ || defined __s390x__ \
568 || defined __mips__ \
569 || defined __alpha__ \
570 || defined __hppa__ \
571 || defined __ia64__ \
572 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
573 #define ECB_STDFP 1
574 #include <string.h> /* for memcpy */
575#else
576 #define ECB_STDFP 0
577 #include <math.h> /* for frexp*, ldexp* */
578#endif
579
580#ifndef ECB_NO_LIBM
581
582 /* convert a float to ieee single/binary32 */
583 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
584 ecb_function_ uint32_t
585 ecb_float_to_binary32 (float x)
586 {
587 uint32_t r;
588
589 #if ECB_STDFP
590 memcpy (&r, &x, 4);
591 #else
592 /* slow emulation, works for anything but -0 */
593 uint32_t m;
594 int e;
595
596 if (x == 0e0f ) return 0x00000000U;
597 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
598 if (x < -3.40282346638528860e+38f) return 0xff800000U;
599 if (x != x ) return 0x7fbfffffU;
600
601 m = frexpf (x, &e) * 0x1000000U;
602
603 r = m & 0x80000000U;
604
605 if (r)
606 m = -m;
607
608 if (e <= -126)
609 {
610 m &= 0xffffffU;
611 m >>= (-125 - e);
612 e = -126;
613 }
614
615 r |= (e + 126) << 23;
616 r |= m & 0x7fffffU;
479#endif 617 #endif
480 618
619 return r;
620 }
621
622 /* converts an ieee single/binary32 to a float */
623 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
624 ecb_function_ float
625 ecb_binary32_to_float (uint32_t x)
626 {
627 float r;
628
629 #if ECB_STDFP
630 memcpy (&r, &x, 4);
631 #else
632 /* emulation, only works for normals and subnormals and +0 */
633 int neg = x >> 31;
634 int e = (x >> 23) & 0xffU;
635
636 x &= 0x7fffffU;
637
638 if (e)
639 x |= 0x800000U;
640 else
641 e = 1;
642
643 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
644 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
645
646 r = neg ? -r : r;
647 #endif
648
649 return r;
650 }
651
652 /* convert a double to ieee double/binary64 */
653 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
654 ecb_function_ uint64_t
655 ecb_double_to_binary64 (double x)
656 {
657 uint64_t r;
658
659 #if ECB_STDFP
660 memcpy (&r, &x, 8);
661 #else
662 /* slow emulation, works for anything but -0 */
663 uint64_t m;
664 int e;
665
666 if (x == 0e0 ) return 0x0000000000000000U;
667 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
668 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
669 if (x != x ) return 0X7ff7ffffffffffffU;
670
671 m = frexp (x, &e) * 0x20000000000000U;
672
673 r = m & 0x8000000000000000;;
674
675 if (r)
676 m = -m;
677
678 if (e <= -1022)
679 {
680 m &= 0x1fffffffffffffU;
681 m >>= (-1021 - e);
682 e = -1022;
683 }
684
685 r |= ((uint64_t)(e + 1022)) << 52;
686 r |= m & 0xfffffffffffffU;
687 #endif
688
689 return r;
690 }
691
692 /* converts an ieee double/binary64 to a double */
693 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
694 ecb_function_ double
695 ecb_binary64_to_double (uint64_t x)
696 {
697 double r;
698
699 #if ECB_STDFP
700 memcpy (&r, &x, 8);
701 #else
702 /* emulation, only works for normals and subnormals and +0 */
703 int neg = x >> 63;
704 int e = (x >> 52) & 0x7ffU;
705
706 x &= 0xfffffffffffffU;
707
708 if (e)
709 x |= 0x10000000000000U;
710 else
711 e = 1;
712
713 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
714 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
715
716 r = neg ? -r : r;
717 #endif
718
719 return r;
720 }
721
722#endif
723
724#endif
725

Diff Legend

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