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

Comparing libecb/ecb.h (file contents):
Revision 1.100 by root, Sun Jun 10 15:46:00 2012 UTC vs.
Revision 1.129 by root, Thu Mar 20 22:04:29 2014 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2014 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:
… …
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 __amd64 || __x86_64 || _M_AMD64 || _M_X64
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
84#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
85#define ECB_C99 (__STDC_VERSION__ >= 199901L)
86#define ECB_C11 (__STDC_VERSION__ >= 201112L)
87#define ECB_CPP (__cplusplus+0) 92#define ECB_CPP (__cplusplus+0)
88#define ECB_CPP11 (__cplusplus >= 201103L) 93#define ECB_CPP11 (__cplusplus >= 201103L)
94
95#if ECB_CPP
96 #define ECB_C 0
97 #define ECB_STDC_VERSION 0
98#else
99 #define ECB_C 1
100 #define ECB_STDC_VERSION __STDC_VERSION__
101#endif
102
103#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
104#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
105
106#if ECB_CPP
107 #define ECB_EXTERN_C extern "C"
108 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
109 #define ECB_EXTERN_C_END }
110#else
111 #define ECB_EXTERN_C extern
112 #define ECB_EXTERN_C_BEG
113 #define ECB_EXTERN_C_END
114#endif
89 115
90/*****************************************************************************/ 116/*****************************************************************************/
91 117
92/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 118/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
93/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ 119/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
… …
116 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 142 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
117 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 143 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
118 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 144 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
119 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 145 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
120 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 146 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
147 #elif __aarch64__
148 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
121 #elif __sparc || __sparc__ 149 #elif (__sparc || __sparc__) && !__sparcv8
122 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") 150 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
123 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 151 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
124 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 152 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
125 #elif defined __s390__ || defined __s390x__ 153 #elif defined __s390__ || defined __s390x__
126 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 154 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
127 #elif defined __mips__ 155 #elif defined __mips__
156 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
157 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
128 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 158 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
129 #elif defined __alpha__ 159 #elif defined __alpha__
130 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") 160 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
131 #elif defined __hppa__ 161 #elif defined __hppa__
132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") 162 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
133 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 163 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
134 #elif defined __ia64__ 164 #elif defined __ia64__
135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") 165 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
166 #elif defined __m68k__
167 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
168 #elif defined __m88k__
169 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
170 #elif defined __sh__
171 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
136 #endif 172 #endif
137 #endif 173 #endif
138#endif 174#endif
139 175
140#ifndef ECB_MEMORY_FENCE 176#ifndef ECB_MEMORY_FENCE
141 #if ECB_GCC_VERSION(4,7) 177 #if ECB_GCC_VERSION(4,7)
142 /* see comment below (stdatomic.h) about the C11 memory model. */ 178 /* see comment below (stdatomic.h) about the C11 memory model. */
143 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 179 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
180 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
181 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
182
183 /* The __has_feature syntax from clang is so misdesigned that we cannot use it
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.
144 #elif defined __clang && __has_feature (cxx_atomic) 187 * #elif defined __clang && __has_feature (cxx_atomic)
145 /* see comment below (stdatomic.h) about the C11 memory model. */ 188 * // see comment below (stdatomic.h) about the C11 memory model.
146 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 189 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
190 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
191 * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
192 */
193
147 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 194 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
148 #define ECB_MEMORY_FENCE __sync_synchronize () 195 #define ECB_MEMORY_FENCE __sync_synchronize ()
196 #elif _MSC_VER >= 1500 /* VC++ 2008 */
197 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
198 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
199 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
200 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
201 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
149 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 202 #elif _MSC_VER >= 1400 /* VC++ 2005 */
150 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 203 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
151 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 204 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
152 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 205 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
153 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 206 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
… …
173 /* any fence other than seq_cst, which isn't very efficient for us. */ 226 /* any fence other than seq_cst, which isn't very efficient for us. */
174 /* Why that is, we don't know - either the C11 memory model is quite useless */ 227 /* Why that is, we don't know - either the C11 memory model is quite useless */
175 /* for most usages, or gcc and clang have a bug */ 228 /* for most usages, or gcc and clang have a bug */
176 /* I *currently* lean towards the latter, and inefficiently implement */ 229 /* I *currently* lean towards the latter, and inefficiently implement */
177 /* all three of ecb's fences as a seq_cst fence */ 230 /* all three of ecb's fences as a seq_cst fence */
231 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
232 /* for all __atomic_thread_fence's except seq_cst */
178 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) 233 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
179 #endif 234 #endif
180#endif 235#endif
181 236
182#ifndef ECB_MEMORY_FENCE 237#ifndef ECB_MEMORY_FENCE
… …
239 #define ecb_is_constant(expr) __builtin_constant_p (expr) 294 #define ecb_is_constant(expr) __builtin_constant_p (expr)
240 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 295 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
241 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 296 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
242#else 297#else
243 #define ecb_attribute(attrlist) 298 #define ecb_attribute(attrlist)
299
300 /* possible C11 impl for integral types
301 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)) */
303
244 #define ecb_is_constant(expr) 0 304 #define ecb_is_constant(expr) 0
245 #define ecb_expect(expr,value) (expr) 305 #define ecb_expect(expr,value) (expr)
246 #define ecb_prefetch(addr,rw,locality) 306 #define ecb_prefetch(addr,rw,locality)
247#endif 307#endif
248 308
… …
466ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 526ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
467ecb_inline unsigned char 527ecb_inline unsigned char
468ecb_byteorder_helper (void) 528ecb_byteorder_helper (void)
469{ 529{
470 /* the union code still generates code under pressure in gcc, */ 530 /* the union code still generates code under pressure in gcc, */
471 /* but less than using pointers, and always seem to */ 531 /* but less than using pointers, and always seems to */
472 /* successfully return a constant. */ 532 /* successfully return a constant. */
473 /* the reason why we have this horrible preprocessor mess */ 533 /* the reason why we have this horrible preprocessor mess */
474 /* is to avoid it in all cases, at least on common architectures */ 534 /* is to avoid it in all cases, at least on common architectures */
475 /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ 535 /* or when using a recent enough gcc version (>= 4.6) */
476#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 536#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
477 return 0x44; 537 return 0x44;
478#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 538#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
479 return 0x44; 539 return 0x44;
480#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 540#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
481 retrurn 0x11; 541 return 0x11;
482#else 542#else
483 union 543 union
484 { 544 {
485 uint32_t i; 545 uint32_t i;
486 uint8_t c; 546 uint8_t c;
… …
525 } 585 }
526#else 586#else
527 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 587 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
528#endif 588#endif
529 589
590/*******************************************************************************/
591/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
592
593/* basically, everything uses "ieee pure-endian" floating point numbers */
594/* the only noteworthy exception is ancient armle, which uses order 43218765 */
595#if 0 \
596 || __i386 || __i386__ \
597 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
598 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
599 || defined __arm__ && defined __ARM_EABI__ \
600 || defined __s390__ || defined __s390x__ \
601 || defined __mips__ \
602 || defined __alpha__ \
603 || defined __hppa__ \
604 || defined __ia64__ \
605 || defined __m68k__ \
606 || defined __m88k__ \
607 || defined __sh__ \
608 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
609 #define ECB_STDFP 1
610 #include <string.h> /* for memcpy */
611#else
612 #define ECB_STDFP 0
613#endif
614
615#ifndef ECB_NO_LIBM
616
617 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
618
619 /* only the oldest of old doesn't have this one. solaris. */
620 #ifdef INFINITY
621 #define ECB_INFINITY INFINITY
622 #else
623 #define ECB_INFINITY HUGE_VAL
530#endif 624 #endif
531 625
626 #ifdef NAN
627 #define ECB_NAN NAN
628 #else
629 #define ECB_NAN ECB_INFINITY
630 #endif
631
632 /* converts an ieee half/binary16 to a float */
633 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
634 ecb_function_ float
635 ecb_binary16_to_float (uint16_t x)
636 {
637 int e = (x >> 10) & 0x1f;
638 int m = x & 0x3ff;
639 float r;
640
641 if (!e ) r = ldexpf (m , -24);
642 else if (e != 31) r = ldexpf (m + 0x400, e - 25);
643 else if (m ) r = ECB_NAN;
644 else r = ECB_INFINITY;
645
646 return x & 0x8000 ? -r : r;
647 }
648
649 /* convert a float to ieee single/binary32 */
650 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
651 ecb_function_ uint32_t
652 ecb_float_to_binary32 (float x)
653 {
654 uint32_t r;
655
656 #if ECB_STDFP
657 memcpy (&r, &x, 4);
658 #else
659 /* slow emulation, works for anything but -0 */
660 uint32_t m;
661 int e;
662
663 if (x == 0e0f ) return 0x00000000U;
664 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
665 if (x < -3.40282346638528860e+38f) return 0xff800000U;
666 if (x != x ) return 0x7fbfffffU;
667
668 m = frexpf (x, &e) * 0x1000000U;
669
670 r = m & 0x80000000U;
671
672 if (r)
673 m = -m;
674
675 if (e <= -126)
676 {
677 m &= 0xffffffU;
678 m >>= (-125 - e);
679 e = -126;
680 }
681
682 r |= (e + 126) << 23;
683 r |= m & 0x7fffffU;
684 #endif
685
686 return r;
687 }
688
689 /* converts an ieee single/binary32 to a float */
690 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
691 ecb_function_ float
692 ecb_binary32_to_float (uint32_t x)
693 {
694 float r;
695
696 #if ECB_STDFP
697 memcpy (&r, &x, 4);
698 #else
699 /* emulation, only works for normals and subnormals and +0 */
700 int neg = x >> 31;
701 int e = (x >> 23) & 0xffU;
702
703 x &= 0x7fffffU;
704
705 if (e)
706 x |= 0x800000U;
707 else
708 e = 1;
709
710 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
711 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
712
713 r = neg ? -r : r;
714 #endif
715
716 return r;
717 }
718
719 /* convert a double to ieee double/binary64 */
720 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
721 ecb_function_ uint64_t
722 ecb_double_to_binary64 (double x)
723 {
724 uint64_t r;
725
726 #if ECB_STDFP
727 memcpy (&r, &x, 8);
728 #else
729 /* slow emulation, works for anything but -0 */
730 uint64_t m;
731 int e;
732
733 if (x == 0e0 ) return 0x0000000000000000U;
734 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
735 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
736 if (x != x ) return 0X7ff7ffffffffffffU;
737
738 m = frexp (x, &e) * 0x20000000000000U;
739
740 r = m & 0x8000000000000000;;
741
742 if (r)
743 m = -m;
744
745 if (e <= -1022)
746 {
747 m &= 0x1fffffffffffffU;
748 m >>= (-1021 - e);
749 e = -1022;
750 }
751
752 r |= ((uint64_t)(e + 1022)) << 52;
753 r |= m & 0xfffffffffffffU;
754 #endif
755
756 return r;
757 }
758
759 /* converts an ieee double/binary64 to a double */
760 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
761 ecb_function_ double
762 ecb_binary64_to_double (uint64_t x)
763 {
764 double r;
765
766 #if ECB_STDFP
767 memcpy (&r, &x, 8);
768 #else
769 /* emulation, only works for normals and subnormals and +0 */
770 int neg = x >> 63;
771 int e = (x >> 52) & 0x7ffU;
772
773 x &= 0xfffffffffffffU;
774
775 if (e)
776 x |= 0x10000000000000U;
777 else
778 e = 1;
779
780 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
781 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
782
783 r = neg ? -r : r;
784 #endif
785
786 return r;
787 }
788
789#endif
790
791#endif
792

Diff Legend

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