… | |
… | |
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, |
… | |
… | |
84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
92 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
93 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
94 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
87 | #define ECB_CPP (__cplusplus+0) |
95 | #define ECB_CPP (__cplusplus+0) |
88 | #define ECB_CPP11 (__cplusplus >= 201103L) |
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 |
89 | |
107 | |
90 | /*****************************************************************************/ |
108 | /*****************************************************************************/ |
91 | |
109 | |
92 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
110 | /* 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 */ |
111 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
… | |
… | |
123 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
141 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
124 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
125 | #elif defined __s390__ || defined __s390x__ |
143 | #elif defined __s390__ || defined __s390x__ |
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
144 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
127 | #elif defined __mips__ |
145 | #elif defined __mips__ |
|
|
146 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
147 | /* 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") |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
129 | #elif defined __alpha__ |
149 | #elif defined __alpha__ |
130 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
131 | #elif defined __hppa__ |
151 | #elif defined __hppa__ |
132 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
152 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
153 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
134 | #elif defined __ia64__ |
154 | #elif defined __ia64__ |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
155 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
156 | #elif defined __m68k__ |
|
|
157 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
158 | #elif defined __m88k__ |
|
|
159 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
160 | #elif defined __sh__ |
|
|
161 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
136 | #endif |
162 | #endif |
137 | #endif |
163 | #endif |
138 | #endif |
164 | #endif |
139 | |
165 | |
140 | #ifndef ECB_MEMORY_FENCE |
166 | #ifndef ECB_MEMORY_FENCE |
141 | #if ECB_GCC_VERSION(4,7) |
167 | #if ECB_GCC_VERSION(4,7) |
142 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
168 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
143 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
169 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
170 | |
|
|
171 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
|
|
172 | * without risking compile time errors with other compilers. We *could* |
|
|
173 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
|
|
174 | * around this shit time and again. |
144 | #elif defined __clang && __has_feature (cxx_atomic) |
175 | * #elif defined __clang && __has_feature (cxx_atomic) |
145 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
176 | * // see comment below (stdatomic.h) about the C11 memory model. |
146 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
177 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
178 | */ |
|
|
179 | |
147 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
180 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
148 | #define ECB_MEMORY_FENCE __sync_synchronize () |
181 | #define ECB_MEMORY_FENCE __sync_synchronize () |
149 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
182 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
150 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
183 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
151 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
184 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
… | |
… | |
459 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
492 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
460 | ecb_inline void ecb_unreachable (void) { } |
493 | ecb_inline void ecb_unreachable (void) { } |
461 | #endif |
494 | #endif |
462 | |
495 | |
463 | /* try to tell the compiler that some condition is definitely true */ |
496 | /* try to tell the compiler that some condition is definitely true */ |
464 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
497 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
465 | |
498 | |
466 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
499 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
467 | ecb_inline unsigned char |
500 | ecb_inline unsigned char |
468 | ecb_byteorder_helper (void) |
501 | ecb_byteorder_helper (void) |
469 | { |
502 | { |
470 | /* the union code still generates code under pressure in gcc, */ |
503 | /* the union code still generates code under pressure in gcc, */ |
471 | /* but less than using pointers, and always seem to */ |
504 | /* but less than using pointers, and always seems to */ |
472 | /* successfully return a constant. */ |
505 | /* successfully return a constant. */ |
473 | /* the reason why we have this horrible preprocessor mess */ |
506 | /* the reason why we have this horrible preprocessor mess */ |
474 | /* is to avoid it in all cases, at least on common architectures */ |
507 | /* is to avoid it in all cases, at least on common architectures */ |
475 | /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ |
508 | /* or when using a recent enough gcc version (>= 4.6) */ |
476 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
509 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
477 | return 0x44; |
510 | return 0x44; |
478 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
511 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
479 | return 0x44; |
512 | return 0x44; |
480 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
513 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
481 | retrurn 0x11; |
514 | return 0x11; |
482 | #else |
515 | #else |
483 | union |
516 | union |
484 | { |
517 | { |
485 | uint32_t i; |
518 | uint32_t i; |
486 | uint8_t c; |
519 | uint8_t c; |
… | |
… | |
525 | } |
558 | } |
526 | #else |
559 | #else |
527 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
560 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
528 | #endif |
561 | #endif |
529 | |
562 | |
|
|
563 | /*******************************************************************************/ |
|
|
564 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
565 | |
|
|
566 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
567 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
568 | #if 0 \ |
|
|
569 | || __i386 || __i386__ \ |
|
|
570 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
571 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
572 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
573 | || defined __s390__ || defined __s390x__ \ |
|
|
574 | || defined __mips__ \ |
|
|
575 | || defined __alpha__ \ |
|
|
576 | || defined __hppa__ \ |
|
|
577 | || defined __ia64__ \ |
|
|
578 | || defined __m68k__ \ |
|
|
579 | || defined __m88k__ \ |
|
|
580 | || defined __sh__ \ |
|
|
581 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
|
|
582 | #define ECB_STDFP 1 |
|
|
583 | #include <string.h> /* for memcpy */ |
|
|
584 | #else |
|
|
585 | #define ECB_STDFP 0 |
|
|
586 | #endif |
|
|
587 | |
|
|
588 | #ifndef ECB_NO_LIBM |
|
|
589 | |
|
|
590 | #include <math.h> /* for frexp*, ldexp* */ |
|
|
591 | |
|
|
592 | /* converts an ieee half/binary16 to a float */ |
|
|
593 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
594 | ecb_function_ float |
|
|
595 | ecb_binary16_to_float (uint16_t x) |
|
|
596 | { |
|
|
597 | int e = (x >> 10) & 0x1f; |
|
|
598 | int m = x & 0x3ff; |
|
|
599 | float r; |
|
|
600 | |
|
|
601 | if (!e) |
|
|
602 | r = ldexpf (m , -24); |
|
|
603 | else if (e != 31) |
|
|
604 | r = ldexpf (m + 0x400, e - 25); |
|
|
605 | else if (m) |
|
|
606 | #ifdef NAN |
|
|
607 | r = NAN; |
|
|
608 | #endif |
|
|
609 | else |
|
|
610 | r = INFINITY; |
|
|
611 | |
|
|
612 | return x & 0x8000 ? -r : r; |
|
|
613 | } |
|
|
614 | |
|
|
615 | /* convert a float to ieee single/binary32 */ |
|
|
616 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
617 | ecb_function_ uint32_t |
|
|
618 | ecb_float_to_binary32 (float x) |
|
|
619 | { |
|
|
620 | uint32_t r; |
|
|
621 | |
|
|
622 | #if ECB_STDFP |
|
|
623 | memcpy (&r, &x, 4); |
|
|
624 | #else |
|
|
625 | /* slow emulation, works for anything but -0 */ |
|
|
626 | uint32_t m; |
|
|
627 | int e; |
|
|
628 | |
|
|
629 | if (x == 0e0f ) return 0x00000000U; |
|
|
630 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
631 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
632 | if (x != x ) return 0x7fbfffffU; |
|
|
633 | |
|
|
634 | m = frexpf (x, &e) * 0x1000000U; |
|
|
635 | |
|
|
636 | r = m & 0x80000000U; |
|
|
637 | |
|
|
638 | if (r) |
|
|
639 | m = -m; |
|
|
640 | |
|
|
641 | if (e <= -126) |
|
|
642 | { |
|
|
643 | m &= 0xffffffU; |
|
|
644 | m >>= (-125 - e); |
|
|
645 | e = -126; |
|
|
646 | } |
|
|
647 | |
|
|
648 | r |= (e + 126) << 23; |
|
|
649 | r |= m & 0x7fffffU; |
530 | #endif |
650 | #endif |
531 | |
651 | |
|
|
652 | return r; |
|
|
653 | } |
|
|
654 | |
|
|
655 | /* converts an ieee single/binary32 to a float */ |
|
|
656 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
657 | ecb_function_ float |
|
|
658 | ecb_binary32_to_float (uint32_t x) |
|
|
659 | { |
|
|
660 | float r; |
|
|
661 | |
|
|
662 | #if ECB_STDFP |
|
|
663 | memcpy (&r, &x, 4); |
|
|
664 | #else |
|
|
665 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
666 | int neg = x >> 31; |
|
|
667 | int e = (x >> 23) & 0xffU; |
|
|
668 | |
|
|
669 | x &= 0x7fffffU; |
|
|
670 | |
|
|
671 | if (e) |
|
|
672 | x |= 0x800000U; |
|
|
673 | else |
|
|
674 | e = 1; |
|
|
675 | |
|
|
676 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
677 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
678 | |
|
|
679 | r = neg ? -r : r; |
|
|
680 | #endif |
|
|
681 | |
|
|
682 | return r; |
|
|
683 | } |
|
|
684 | |
|
|
685 | /* convert a double to ieee double/binary64 */ |
|
|
686 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
687 | ecb_function_ uint64_t |
|
|
688 | ecb_double_to_binary64 (double x) |
|
|
689 | { |
|
|
690 | uint64_t r; |
|
|
691 | |
|
|
692 | #if ECB_STDFP |
|
|
693 | memcpy (&r, &x, 8); |
|
|
694 | #else |
|
|
695 | /* slow emulation, works for anything but -0 */ |
|
|
696 | uint64_t m; |
|
|
697 | int e; |
|
|
698 | |
|
|
699 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
700 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
701 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
702 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
703 | |
|
|
704 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
705 | |
|
|
706 | r = m & 0x8000000000000000;; |
|
|
707 | |
|
|
708 | if (r) |
|
|
709 | m = -m; |
|
|
710 | |
|
|
711 | if (e <= -1022) |
|
|
712 | { |
|
|
713 | m &= 0x1fffffffffffffU; |
|
|
714 | m >>= (-1021 - e); |
|
|
715 | e = -1022; |
|
|
716 | } |
|
|
717 | |
|
|
718 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
719 | r |= m & 0xfffffffffffffU; |
|
|
720 | #endif |
|
|
721 | |
|
|
722 | return r; |
|
|
723 | } |
|
|
724 | |
|
|
725 | /* converts an ieee double/binary64 to a double */ |
|
|
726 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
727 | ecb_function_ double |
|
|
728 | ecb_binary64_to_double (uint64_t x) |
|
|
729 | { |
|
|
730 | double r; |
|
|
731 | |
|
|
732 | #if ECB_STDFP |
|
|
733 | memcpy (&r, &x, 8); |
|
|
734 | #else |
|
|
735 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
736 | int neg = x >> 63; |
|
|
737 | int e = (x >> 52) & 0x7ffU; |
|
|
738 | |
|
|
739 | x &= 0xfffffffffffffU; |
|
|
740 | |
|
|
741 | if (e) |
|
|
742 | x |= 0x10000000000000U; |
|
|
743 | else |
|
|
744 | e = 1; |
|
|
745 | |
|
|
746 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
747 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
748 | |
|
|
749 | r = neg ? -r : r; |
|
|
750 | #endif |
|
|
751 | |
|
|
752 | return r; |
|
|
753 | } |
|
|
754 | |
|
|
755 | #endif |
|
|
756 | |
|
|
757 | #endif |
|
|
758 | |