… | |
… | |
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; |
… | |
… | |
60 | #include <inttypes.h> |
60 | #include <inttypes.h> |
61 | #if UINTMAX_MAX > 0xffffffffU |
61 | #if UINTMAX_MAX > 0xffffffffU |
62 | #define ECB_PTRSIZE 8 |
62 | #define ECB_PTRSIZE 8 |
63 | #else |
63 | #else |
64 | #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 |
65 | #endif |
74 | #endif |
66 | #endif |
75 | #endif |
67 | |
76 | |
68 | /* many compilers define _GNUC_ to some versions but then only implement |
77 | /* many compilers define _GNUC_ to some versions but then only implement |
69 | * what their idiot authors think are the "more important" extensions, |
78 | * what their idiot authors think are the "more important" extensions, |
… | |
… | |
132 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
141 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
134 | #elif defined __s390__ || defined __s390x__ |
143 | #elif defined __s390__ || defined __s390x__ |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
144 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
136 | #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. */ |
137 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
138 | #elif defined __alpha__ |
149 | #elif defined __alpha__ |
139 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
140 | #elif defined __hppa__ |
151 | #elif defined __hppa__ |
141 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
152 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
153 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
143 | #elif defined __ia64__ |
154 | #elif defined __ia64__ |
144 | #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") |
145 | #endif |
162 | #endif |
146 | #endif |
163 | #endif |
147 | #endif |
164 | #endif |
148 | |
165 | |
149 | #ifndef ECB_MEMORY_FENCE |
166 | #ifndef ECB_MEMORY_FENCE |
150 | #if ECB_GCC_VERSION(4,7) |
167 | #if ECB_GCC_VERSION(4,7) |
151 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
168 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
152 | #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. |
153 | /*#elif defined __clang && __has_feature (cxx_atomic)*/ |
175 | * #elif defined __clang && __has_feature (cxx_atomic) |
154 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
176 | * // see comment below (stdatomic.h) about the C11 memory model. |
155 | /*#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 | |
156 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
180 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
157 | #define ECB_MEMORY_FENCE __sync_synchronize () |
181 | #define ECB_MEMORY_FENCE __sync_synchronize () |
158 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
182 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
159 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
183 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
160 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
184 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
… | |
… | |
475 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
499 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
476 | ecb_inline unsigned char |
500 | ecb_inline unsigned char |
477 | ecb_byteorder_helper (void) |
501 | ecb_byteorder_helper (void) |
478 | { |
502 | { |
479 | /* the union code still generates code under pressure in gcc, */ |
503 | /* the union code still generates code under pressure in gcc, */ |
480 | /* but less than using pointers, and always seem to */ |
504 | /* but less than using pointers, and always seems to */ |
481 | /* successfully return a constant. */ |
505 | /* successfully return a constant. */ |
482 | /* the reason why we have this horrible preprocessor mess */ |
506 | /* the reason why we have this horrible preprocessor mess */ |
483 | /* 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 */ |
484 | /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ |
508 | /* or when using a recent enough gcc version (>= 4.6) */ |
485 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
509 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
486 | return 0x44; |
510 | return 0x44; |
487 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
511 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
488 | return 0x44; |
512 | return 0x44; |
489 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
513 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
490 | retrurn 0x11; |
514 | return 0x11; |
491 | #else |
515 | #else |
492 | union |
516 | union |
493 | { |
517 | { |
494 | uint32_t i; |
518 | uint32_t i; |
495 | uint8_t c; |
519 | uint8_t c; |
… | |
… | |
549 | || defined __s390__ || defined __s390x__ \ |
573 | || defined __s390__ || defined __s390x__ \ |
550 | || defined __mips__ \ |
574 | || defined __mips__ \ |
551 | || defined __alpha__ \ |
575 | || defined __alpha__ \ |
552 | || defined __hppa__ \ |
576 | || defined __hppa__ \ |
553 | || defined __ia64__ \ |
577 | || defined __ia64__ \ |
|
|
578 | || defined __m68k__ \ |
|
|
579 | || defined __m88k__ \ |
|
|
580 | || defined __sh__ \ |
554 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
581 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
555 | #define ECB_STDFP 1 |
582 | #define ECB_STDFP 1 |
556 | #include <string.h> /* for memcpy */ |
583 | #include <string.h> /* for memcpy */ |
557 | #else |
584 | #else |
558 | #define ECB_STDFP 0 |
585 | #define ECB_STDFP 0 |
559 | #endif |
586 | #endif |
560 | |
587 | |
561 | #ifndef ECB_NO_LIBM |
588 | #ifndef ECB_NO_LIBM |
562 | |
589 | |
563 | #if __STDC_IEC_559__ || ECB_STDFP |
590 | #include <math.h> /* for frexp*, ldexp* */ |
564 | /* we assume this is defined for most C and many C++ compilers */ |
591 | |
565 | ecb_inline ecb_bool ecb_float_ieee (void) ecb_const; |
592 | /* converts an ieee half/binary16 to a float */ |
566 | ecb_inline ecb_bool ecb_float_ieee (void) { return 1; } |
593 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
567 | ecb_inline ecb_bool ecb_double_ieee (void) ecb_const; |
594 | ecb_function_ float |
568 | ecb_inline ecb_bool ecb_double_ieee (void) { return 1; } |
595 | ecb_binary16_to_float (uint16_t x) |
569 | #elif ECB_CPP |
596 | { |
570 | #include <limits> |
597 | int e = (x >> 10) & 0x1f; |
571 | ecb_inline ecb_bool ecb_float_ieee (void) ecb_const; |
598 | int m = x & 0x3ff; |
572 | ecb_inline ecb_bool ecb_float_ieee (void) { return std::numeric_limits<float >::is_iec559; } |
599 | float r; |
573 | ecb_inline ecb_bool ecb_double_ieee (void) ecb_const; |
600 | |
574 | ecb_inline ecb_bool ecb_double_ieee (void) { return std::numeric_limits<double>::is_iec559; } |
601 | if (!e) |
575 | #else |
602 | r = ldexpf (m , -24); |
576 | ecb_inline ecb_bool ecb_float_ieee (void) ecb_const; |
603 | else if (e != 31) |
577 | ecb_inline ecb_bool ecb_float_ieee (void) { return 0; } |
604 | r = ldexpf (m + 0x400, e - 25); |
578 | ecb_inline ecb_bool ecb_double_ieee (void) ecb_const; |
605 | else if (m) |
579 | ecb_inline ecb_bool ecb_double_ieee (void) { return 0; } |
606 | #ifdef NAN |
|
|
607 | r = NAN; |
580 | #endif |
608 | #endif |
|
|
609 | else |
|
|
610 | r = INFINITY; |
|
|
611 | |
|
|
612 | return x & 0x8000 ? -r : r; |
|
|
613 | } |
581 | |
614 | |
582 | /* convert a float to ieee single/binary32 */ |
615 | /* convert a float to ieee single/binary32 */ |
583 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
616 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
584 | ecb_function_ uint32_t |
617 | ecb_function_ uint32_t |
585 | ecb_float_to_binary32 (float x) |
618 | ecb_float_to_binary32 (float x) |
… | |
… | |
588 | |
621 | |
589 | #if ECB_STDFP |
622 | #if ECB_STDFP |
590 | memcpy (&r, &x, 4); |
623 | memcpy (&r, &x, 4); |
591 | #else |
624 | #else |
592 | /* slow emulation, works for anything but -0 */ |
625 | /* slow emulation, works for anything but -0 */ |
593 | ECB_EXTERN_C float frexpf (float v, int *e); |
|
|
594 | uint32_t m; |
626 | uint32_t m; |
595 | int e; |
627 | int e; |
596 | |
628 | |
597 | if (x == 0e0f ) return 0; |
629 | if (x == 0e0f ) return 0x00000000U; |
598 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
630 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
599 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
631 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
600 | if (x != x ) return 0x7fbfffffU; |
632 | if (x != x ) return 0x7fbfffffU; |
601 | |
633 | |
602 | m = frexpf (x, &e) * 0x1000000U; |
634 | m = frexpf (x, &e) * 0x1000000U; |
… | |
… | |
604 | r = m & 0x80000000U; |
636 | r = m & 0x80000000U; |
605 | |
637 | |
606 | if (r) |
638 | if (r) |
607 | m = -m; |
639 | m = -m; |
608 | |
640 | |
609 | if (e < -125) |
641 | if (e <= -126) |
610 | { |
642 | { |
611 | m &= 0xffffffU; |
643 | m &= 0xffffffU; |
612 | m >>= (-125 - e); |
644 | m >>= (-125 - e); |
613 | e = -126; |
645 | e = -126; |
614 | } |
646 | } |
… | |
… | |
629 | |
661 | |
630 | #if ECB_STDFP |
662 | #if ECB_STDFP |
631 | memcpy (&r, &x, 4); |
663 | memcpy (&r, &x, 4); |
632 | #else |
664 | #else |
633 | /* emulation, only works for normals and subnormals and +0 */ |
665 | /* 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; |
666 | int neg = x >> 31; |
637 | int e = (x >> 23) & 0xffU; |
667 | int e = (x >> 23) & 0xffU; |
638 | |
668 | |
639 | x &= 0x7fffffU; |
669 | x &= 0x7fffffU; |
640 | |
670 | |
… | |
… | |
642 | x |= 0x800000U; |
672 | x |= 0x800000U; |
643 | else |
673 | else |
644 | e = 1; |
674 | e = 1; |
645 | |
675 | |
646 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
676 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
647 | r = ldexpf (x * (1.f / 0x1000000U), e - 126); |
677 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
648 | |
678 | |
649 | r = neg ? -r : r; |
679 | r = neg ? -r : r; |
650 | #endif |
680 | #endif |
651 | |
681 | |
652 | return r; |
682 | return r; |
… | |
… | |
661 | |
691 | |
662 | #if ECB_STDFP |
692 | #if ECB_STDFP |
663 | memcpy (&r, &x, 8); |
693 | memcpy (&r, &x, 8); |
664 | #else |
694 | #else |
665 | /* slow emulation, works for anything but -0 */ |
695 | /* slow emulation, works for anything but -0 */ |
666 | ECB_EXTERN_C double frexp (double v, int *e); |
|
|
667 | uint64_t m; |
696 | uint64_t m; |
668 | int e; |
697 | int e; |
669 | |
698 | |
670 | if (x == 0e0 ) return 0; |
699 | if (x == 0e0 ) return 0x0000000000000000U; |
671 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
700 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
672 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
701 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
673 | if (x != x ) return 0X7ff7ffffffffffffU; |
702 | if (x != x ) return 0X7ff7ffffffffffffU; |
674 | |
703 | |
675 | m = frexpf (x, &e) * 0x1000000U; |
|
|
676 | m = frexp (x, &e) * 0x20000000000000U; |
704 | m = frexp (x, &e) * 0x20000000000000U; |
677 | |
705 | |
678 | r = m & 0x8000000000000000;; |
706 | r = m & 0x8000000000000000;; |
679 | |
707 | |
680 | if (r) |
708 | if (r) |
681 | m = -m; |
709 | m = -m; |
682 | |
710 | |
683 | if (e < -1021) |
711 | if (e <= -1022) |
684 | { |
712 | { |
685 | m &= 0x1fffffffffffffU; |
713 | m &= 0x1fffffffffffffU; |
686 | m >>= (-1021 - e); |
714 | m >>= (-1021 - e); |
687 | e = -1022; |
715 | e = -1022; |
688 | } |
716 | } |
… | |
… | |
703 | |
731 | |
704 | #if ECB_STDFP |
732 | #if ECB_STDFP |
705 | memcpy (&r, &x, 8); |
733 | memcpy (&r, &x, 8); |
706 | #else |
734 | #else |
707 | /* emulation, only works for normals and subnormals and +0 */ |
735 | /* 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; |
736 | int neg = x >> 63; |
711 | int e = (x >> 52) & 0x7ffU; |
737 | int e = (x >> 52) & 0x7ffU; |
712 | |
738 | |
713 | x &= 0xfffffffffffffU; |
739 | x &= 0xfffffffffffffU; |
714 | |
740 | |
715 | if (e) |
741 | if (e) |
716 | x |= 0x10000000000000U; |
742 | x |= 0x10000000000000U; |
717 | else |
743 | else |
718 | e = 1; |
744 | e = 1; |
719 | |
745 | |
720 | /* we distrust ldexpf a bit and do the 2**-53 scaling by an extra multiply */ |
746 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
721 | r = ldexp (x * (1. / 0x20000000000000U), e - 1022); |
747 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
722 | |
748 | |
723 | r = neg ? -r : r; |
749 | r = neg ? -r : r; |
724 | #endif |
750 | #endif |
725 | |
751 | |
726 | return r; |
752 | return r; |