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

Comparing libecb/ecb.h (file contents):
Revision 1.91 by root, Tue May 29 14:30:44 2012 UTC vs.
Revision 1.105 by root, Thu Jun 28 19:42:30 2012 UTC

… …
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
… …
83 82
84#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ 83#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
85#define ECB_C99 (__STDC_VERSION__ >= 199901L) 84#define ECB_C99 (__STDC_VERSION__ >= 199901L)
86#define ECB_C11 (__STDC_VERSION__ >= 201112L) 85#define ECB_C11 (__STDC_VERSION__ >= 201112L)
87#define ECB_CPP (__cplusplus+0) 86#define ECB_CPP (__cplusplus+0)
88#define ECB_CPP98 (__cplusplus >= 199711L)
89#define ECB_CPP11 (__cplusplus >= 201103L) 87#define ECB_CPP11 (__cplusplus >= 201103L)
88
89#if ECB_CPP
90 #define ECB_EXTERN_C extern "C"
91 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
92 #define ECB_EXTERN_C_END }
93#else
94 #define ECB_EXTERN_C extern
95 #define ECB_EXTERN_C_BEG
96 #define ECB_EXTERN_C_END
97#endif
90 98
91/*****************************************************************************/ 99/*****************************************************************************/
92 100
93/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 101/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
94/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ 102/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
95 103
96#if ECB_NO_THREADS 104#if ECB_NO_THREADS
97# define ECB_NO_SMP 1 105 #define ECB_NO_SMP 1
98#endif 106#endif
99 107
100#if ECB_NO_THREADS || ECB_NO_SMP 108#if ECB_NO_SMP
101 #define ECB_MEMORY_FENCE do { } while (0) 109 #define ECB_MEMORY_FENCE do { } while (0)
102#endif
103
104#ifndef ECB_MEMORY_FENCE
105 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
106 /* we assume that these memory fences work on all variables/all memory accesses, */
107 /* not just C11 atomics and atomic accesses */
108 #include <stdatomic.h>
109 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_acq_rel)
110 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
111 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
112 #endif
113#endif 110#endif
114 111
115#ifndef ECB_MEMORY_FENCE 112#ifndef ECB_MEMORY_FENCE
116 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 113 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
117 #if __i386 || __i386__ 114 #if __i386 || __i386__
118 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 115 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
119 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ 116 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
120 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ 117 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
121 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 118 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
122 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 119 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
123 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 120 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
124 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ 121 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
125 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 122 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
126 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 123 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
127 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 124 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
128 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 125 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
129 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 126 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
130 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 127 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
131 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 128 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 129 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
133 #elif __sparc || __sparc__ 130 #elif __sparc || __sparc__
134 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") 131 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
135 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 132 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
136 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 133 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
137 #elif defined __s390__ || defined __s390x__ 134 #elif defined __s390__ || defined __s390x__
138 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
139 #elif defined __mips__ 136 #elif defined __mips__
140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 137 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
141 #elif defined __alpha__ 138 #elif defined __alpha__
142 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") 139 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
140 #elif defined __hppa__
141 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
143 #elif defined __ia64__
144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
143 #endif 145 #endif
144 #endif 146 #endif
145#endif 147#endif
146 148
147#ifndef ECB_MEMORY_FENCE 149#ifndef ECB_MEMORY_FENCE
150 #if ECB_GCC_VERSION(4,7)
151 /* see comment below (stdatomic.h) about the C11 memory model. */
152 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
153 /*#elif defined __clang && __has_feature (cxx_atomic)*/
154 /* see comment below (stdatomic.h) about the C11 memory model. */
155 /*#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)*/
148 #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 156 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
149 #define ECB_MEMORY_FENCE __sync_synchronize () 157 #define ECB_MEMORY_FENCE __sync_synchronize ()
150 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */
151 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */
152 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 158 #elif _MSC_VER >= 1400 /* VC++ 2005 */
153 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 159 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
154 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 160 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
155 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 161 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
156 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 162 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
… …
166 #define ECB_MEMORY_FENCE __sync () 172 #define ECB_MEMORY_FENCE __sync ()
167 #endif 173 #endif
168#endif 174#endif
169 175
170#ifndef ECB_MEMORY_FENCE 176#ifndef ECB_MEMORY_FENCE
177 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
178 /* we assume that these memory fences work on all variables/all memory accesses, */
179 /* not just C11 atomics and atomic accesses */
180 #include <stdatomic.h>
181 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
182 /* any fence other than seq_cst, which isn't very efficient for us. */
183 /* Why that is, we don't know - either the C11 memory model is quite useless */
184 /* for most usages, or gcc and clang have a bug */
185 /* I *currently* lean towards the latter, and inefficiently implement */
186 /* all three of ecb's fences as a seq_cst fence */
187 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
188 #endif
189#endif
190
191#ifndef ECB_MEMORY_FENCE
171 #if !ECB_AVOID_PTHREADS 192 #if !ECB_AVOID_PTHREADS
172 /* 193 /*
173 * if you get undefined symbol references to pthread_mutex_lock, 194 * if you get undefined symbol references to pthread_mutex_lock,
174 * or failure to find pthread.h, then you should implement 195 * or failure to find pthread.h, then you should implement
175 * the ECB_MEMORY_FENCE operations for your cpu/compiler 196 * the ECB_MEMORY_FENCE operations for your cpu/compiler
… …
447 ecb_inline void ecb_unreachable (void) ecb_noreturn; 468 ecb_inline void ecb_unreachable (void) ecb_noreturn;
448 ecb_inline void ecb_unreachable (void) { } 469 ecb_inline void ecb_unreachable (void) { }
449#endif 470#endif
450 471
451/* try to tell the compiler that some condition is definitely true */ 472/* try to tell the compiler that some condition is definitely true */
452#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 473#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
453 474
454ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 475ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
455ecb_inline unsigned char 476ecb_inline unsigned char
456ecb_byteorder_helper (void) 477ecb_byteorder_helper (void)
457{ 478{
458 const uint32_t u = 0x11223344; 479 /* the union code still generates code under pressure in gcc, */
459 return *(unsigned char *)&u; 480 /* but less than using pointers, and always seem to */
481 /* successfully return a constant. */
482 /* the reason why we have this horrible preprocessor mess */
483 /* is to avoid it in all cases, at least on common architectures */
484 /* and yes, gcc defines __BYTE_ORDER__, g++ does not */
485#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
486 return 0x44;
487#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
488 return 0x44;
489#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
490 retrurn 0x11;
491#else
492 union
493 {
494 uint32_t i;
495 uint8_t c;
496 } u = { 0x11223344 };
497 return u.c;
498#endif
460} 499}
461 500
462ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 501ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
463ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 502ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
464ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 503ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
… …
495 } 534 }
496#else 535#else
497 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 536 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
498#endif 537#endif
499 538
539/*******************************************************************************/
540/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
541
542/* basically, everything uses "ieee pure-endian" floating point numbers */
543/* the only noteworthy exception is ancient armle, which uses order 43218765 */
544#if 0 \
545 || __i386 || __i386__ \
546 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
547 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
548 || defined __arm__ && defined __ARM_EABI__ \
549 || defined __s390__ || defined __s390x__ \
550 || defined __mips__ \
551 || defined __alpha__ \
552 || defined __hppa__ \
553 || defined __ia64__ \
554 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
555 #define ECB_STDFP 1
556 #include <string.h> /* for memcpy */
557#else
558 #define ECB_STDFP 0
559#endif
560
561#ifndef ECB_NO_LIBM
562
563 #if __STDC_IEC_559__ || ECB_STDFP
564 /* we assume this is defined for most C and many C++ compilers */
565 ecb_inline ecb_bool ecb_float_ieee (void) ecb_const;
566 ecb_inline ecb_bool ecb_float_ieee (void) { return 1; }
567 ecb_inline ecb_bool ecb_double_ieee (void) ecb_const;
568 ecb_inline ecb_bool ecb_double_ieee (void) { return 1; }
569 #elif ECB_CPP
570 #include <limits>
571 ecb_inline ecb_bool ecb_float_ieee (void) ecb_const;
572 ecb_inline ecb_bool ecb_float_ieee (void) { return std::numeric_limits<float >::is_iec559; }
573 ecb_inline ecb_bool ecb_double_ieee (void) ecb_const;
574 ecb_inline ecb_bool ecb_double_ieee (void) { return std::numeric_limits<double>::is_iec559; }
575 #else
576 ecb_inline ecb_bool ecb_float_ieee (void) ecb_const;
577 ecb_inline ecb_bool ecb_float_ieee (void) { return 0; }
578 ecb_inline ecb_bool ecb_double_ieee (void) ecb_const;
579 ecb_inline ecb_bool ecb_double_ieee (void) { return 0; }
500#endif 580 #endif
501 581
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 ECB_EXTERN_C float frexpf (float v, int *e);
594 uint32_t m;
595 int e;
596
597 if (x == 0e0f ) return 0;
598 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
599 if (x < -3.40282346638528860e+38f) return 0xff800000U;
600 if (x != x ) return 0x7fbfffffU;
601
602 m = frexpf (x, &e) * 0x1000000U;
603
604 r = m & 0x80000000U;
605
606 if (r)
607 m = -m;
608
609 if (e < -125)
610 {
611 m &= 0xffffffU;
612 m >>= (-125 - e);
613 e = -126;
614 }
615
616 r |= (e + 126) << 23;
617 r |= m & 0x7fffffU;
618 #endif
619
620 return r;
621 }
622
623 /* converts an ieee single/binary32 to a float */
624 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
625 ecb_function_ float
626 ecb_binary32_to_float (uint32_t x)
627 {
628 float r;
629
630 #if ECB_STDFP
631 memcpy (&r, &x, 4);
632 #else
633 /* emulation, only works for normals and subnormals and +0 */
634 ECB_EXTERN_C float ldexpf (float x, int e);
635
636 int neg = x >> 31;
637 int e = (x >> 23) & 0xffU;
638
639 x &= 0x7fffffU;
640
641 if (e)
642 x |= 0x800000U;
643 else
644 e = 1;
645
646 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
647 r = ldexpf (x * (1.f / 0x1000000U), e - 126);
648
649 r = neg ? -r : r;
650 #endif
651
652 return r;
653 }
654
655 /* convert a double to ieee double/binary64 */
656 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
657 ecb_function_ uint64_t
658 ecb_double_to_binary64 (double x)
659 {
660 uint64_t r;
661
662 #if ECB_STDFP
663 memcpy (&r, &x, 8);
664 #else
665 /* slow emulation, works for anything but -0 */
666 ECB_EXTERN_C double frexp (double v, int *e);
667 uint64_t m;
668 int e;
669
670 if (x == 0e0 ) return 0;
671 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
672 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
673 if (x != x ) return 0X7ff7ffffffffffffU;
674
675 m = frexpf (x, &e) * 0x1000000U;
676 m = frexp (x, &e) * 0x20000000000000U;
677
678 r = m & 0x8000000000000000;;
679
680 if (r)
681 m = -m;
682
683 if (e < -1021)
684 {
685 m &= 0x1fffffffffffffU;
686 m >>= (-1021 - e);
687 e = -1022;
688 }
689
690 r |= ((uint64_t)(e + 1022)) << 52;
691 r |= m & 0xfffffffffffffU;
692 #endif
693
694 return r;
695 }
696
697 /* converts an ieee double/binary64 to a double */
698 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
699 ecb_function_ double
700 ecb_binary64_to_double (uint64_t x)
701 {
702 double r;
703
704 #if ECB_STDFP
705 memcpy (&r, &x, 8);
706 #else
707 /* emulation, only works for normals and subnormals and +0 */
708 ECB_EXTERN_C double ldexp (double x, int e);
709
710 int neg = x >> 63;
711 int e = (x >> 52) & 0x7ffU;
712
713 x &= 0xfffffffffffffU;
714
715 if (e)
716 x |= 0x10000000000000U;
717 else
718 e = 1;
719
720 /* we distrust ldexpf a bit and do the 2**-53 scaling by an extra multiply */
721 r = ldexp (x * (1. / 0x20000000000000U), e - 1022);
722
723 r = neg ? -r : r;
724 #endif
725
726 return r;
727 }
728
729#endif
730
731#endif
732

Diff Legend

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