| 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 (©) 2009-2013 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") |
| 121 | #elif __sparc || __sparc__ |
147 | #elif (__sparc || __sparc__) && !__sparcv8 |
| 122 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
| 123 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
149 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
| 124 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
150 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
| 125 | #elif defined __s390__ || defined __s390x__ |
151 | #elif defined __s390__ || defined __s390x__ |
| 126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
152 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
| 127 | #elif defined __mips__ |
153 | #elif defined __mips__ |
|
|
154 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
155 | /* 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") |
156 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
| 129 | #elif defined __alpha__ |
157 | #elif defined __alpha__ |
| 130 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
158 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
| 131 | #elif defined __hppa__ |
159 | #elif defined __hppa__ |
| 132 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
160 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
161 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 134 | #elif defined __ia64__ |
162 | #elif defined __ia64__ |
| 135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
163 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
164 | #elif defined __m68k__ |
|
|
165 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
166 | #elif defined __m88k__ |
|
|
167 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
168 | #elif defined __sh__ |
|
|
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 136 | #endif |
170 | #endif |
| 137 | #endif |
171 | #endif |
| 138 | #endif |
172 | #endif |
| 139 | |
173 | |
| 140 | #ifndef ECB_MEMORY_FENCE |
174 | #ifndef ECB_MEMORY_FENCE |
| 141 | #if ECB_GCC_VERSION(4,7) |
175 | #if ECB_GCC_VERSION(4,7) |
| 142 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
176 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
| 143 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
177 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
178 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
179 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
180 | |
|
|
181 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
|
|
182 | * without risking compile time errors with other compilers. We *could* |
|
|
183 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
|
|
184 | * around this shit time and again. |
| 144 | #elif defined __clang && __has_feature (cxx_atomic) |
185 | * #elif defined __clang && __has_feature (cxx_atomic) |
| 145 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
186 | * // see comment below (stdatomic.h) about the C11 memory model. |
| 146 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
187 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
188 | * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
189 | * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
190 | */ |
|
|
191 | |
| 147 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
192 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 148 | #define ECB_MEMORY_FENCE __sync_synchronize () |
193 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
194 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
|
|
195 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
196 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
197 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
198 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
199 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 149 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
200 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 150 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
201 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 151 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
202 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 152 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
203 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 153 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
204 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| … | |
… | |
| 173 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
224 | /* 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 */ |
225 | /* 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 */ |
226 | /* for most usages, or gcc and clang have a bug */ |
| 176 | /* I *currently* lean towards the latter, and inefficiently implement */ |
227 | /* I *currently* lean towards the latter, and inefficiently implement */ |
| 177 | /* all three of ecb's fences as a seq_cst fence */ |
228 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
229 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
230 | /* for all __atomic_thread_fence's except seq_cst */ |
| 178 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
231 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
| 179 | #endif |
232 | #endif |
| 180 | #endif |
233 | #endif |
| 181 | |
234 | |
| 182 | #ifndef ECB_MEMORY_FENCE |
235 | #ifndef ECB_MEMORY_FENCE |
| … | |
… | |
| 239 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
292 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
| 240 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
293 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
| 241 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
294 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 242 | #else |
295 | #else |
| 243 | #define ecb_attribute(attrlist) |
296 | #define ecb_attribute(attrlist) |
|
|
297 | |
|
|
298 | /* possible C11 impl for integral types |
|
|
299 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
300 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
301 | |
| 244 | #define ecb_is_constant(expr) 0 |
302 | #define ecb_is_constant(expr) 0 |
| 245 | #define ecb_expect(expr,value) (expr) |
303 | #define ecb_expect(expr,value) (expr) |
| 246 | #define ecb_prefetch(addr,rw,locality) |
304 | #define ecb_prefetch(addr,rw,locality) |
| 247 | #endif |
305 | #endif |
| 248 | |
306 | |
| … | |
… | |
| 459 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
517 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
| 460 | ecb_inline void ecb_unreachable (void) { } |
518 | ecb_inline void ecb_unreachable (void) { } |
| 461 | #endif |
519 | #endif |
| 462 | |
520 | |
| 463 | /* try to tell the compiler that some condition is definitely true */ |
521 | /* try to tell the compiler that some condition is definitely true */ |
| 464 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
522 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 465 | |
523 | |
| 466 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
524 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
| 467 | ecb_inline unsigned char |
525 | ecb_inline unsigned char |
| 468 | ecb_byteorder_helper (void) |
526 | ecb_byteorder_helper (void) |
| 469 | { |
527 | { |
| 470 | const uint32_t u = 0x11223344; |
528 | /* the union code still generates code under pressure in gcc, */ |
| 471 | return *(unsigned char *)&u; |
529 | /* but less than using pointers, and always seems to */ |
|
|
530 | /* successfully return a constant. */ |
|
|
531 | /* the reason why we have this horrible preprocessor mess */ |
|
|
532 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
533 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
534 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
535 | return 0x44; |
|
|
536 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
537 | return 0x44; |
|
|
538 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
539 | return 0x11; |
|
|
540 | #else |
|
|
541 | union |
|
|
542 | { |
|
|
543 | uint32_t i; |
|
|
544 | uint8_t c; |
|
|
545 | } u = { 0x11223344 }; |
|
|
546 | return u.c; |
|
|
547 | #endif |
| 472 | } |
548 | } |
| 473 | |
549 | |
| 474 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
550 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
| 475 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
551 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
| 476 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
552 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
| … | |
… | |
| 507 | } |
583 | } |
| 508 | #else |
584 | #else |
| 509 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
585 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 510 | #endif |
586 | #endif |
| 511 | |
587 | |
|
|
588 | /*******************************************************************************/ |
|
|
589 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
590 | |
|
|
591 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
592 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
593 | #if 0 \ |
|
|
594 | || __i386 || __i386__ \ |
|
|
595 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
596 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
597 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
598 | || defined __s390__ || defined __s390x__ \ |
|
|
599 | || defined __mips__ \ |
|
|
600 | || defined __alpha__ \ |
|
|
601 | || defined __hppa__ \ |
|
|
602 | || defined __ia64__ \ |
|
|
603 | || defined __m68k__ \ |
|
|
604 | || defined __m88k__ \ |
|
|
605 | || defined __sh__ \ |
|
|
606 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
|
|
607 | #define ECB_STDFP 1 |
|
|
608 | #include <string.h> /* for memcpy */ |
|
|
609 | #else |
|
|
610 | #define ECB_STDFP 0 |
|
|
611 | #endif |
|
|
612 | |
|
|
613 | #ifndef ECB_NO_LIBM |
|
|
614 | |
|
|
615 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
616 | |
|
|
617 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
618 | #ifdef INFINITY |
|
|
619 | #define ECB_INFINITY INFINITY |
|
|
620 | #else |
|
|
621 | #define ECB_INFINITY HUGE_VAL |
| 512 | #endif |
622 | #endif |
| 513 | |
623 | |
|
|
624 | #ifdef NAN |
|
|
625 | #define ECB_NAN NAN |
|
|
626 | #else |
|
|
627 | #define ECB_NAN ECB_INFINITY |
|
|
628 | #endif |
|
|
629 | |
|
|
630 | /* converts an ieee half/binary16 to a float */ |
|
|
631 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
632 | ecb_function_ float |
|
|
633 | ecb_binary16_to_float (uint16_t x) |
|
|
634 | { |
|
|
635 | int e = (x >> 10) & 0x1f; |
|
|
636 | int m = x & 0x3ff; |
|
|
637 | float r; |
|
|
638 | |
|
|
639 | if (!e ) r = ldexpf (m , -24); |
|
|
640 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
641 | else if (m ) r = ECB_NAN; |
|
|
642 | else r = ECB_INFINITY; |
|
|
643 | |
|
|
644 | return x & 0x8000 ? -r : r; |
|
|
645 | } |
|
|
646 | |
|
|
647 | /* convert a float to ieee single/binary32 */ |
|
|
648 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
649 | ecb_function_ uint32_t |
|
|
650 | ecb_float_to_binary32 (float x) |
|
|
651 | { |
|
|
652 | uint32_t r; |
|
|
653 | |
|
|
654 | #if ECB_STDFP |
|
|
655 | memcpy (&r, &x, 4); |
|
|
656 | #else |
|
|
657 | /* slow emulation, works for anything but -0 */ |
|
|
658 | uint32_t m; |
|
|
659 | int e; |
|
|
660 | |
|
|
661 | if (x == 0e0f ) return 0x00000000U; |
|
|
662 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
663 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
664 | if (x != x ) return 0x7fbfffffU; |
|
|
665 | |
|
|
666 | m = frexpf (x, &e) * 0x1000000U; |
|
|
667 | |
|
|
668 | r = m & 0x80000000U; |
|
|
669 | |
|
|
670 | if (r) |
|
|
671 | m = -m; |
|
|
672 | |
|
|
673 | if (e <= -126) |
|
|
674 | { |
|
|
675 | m &= 0xffffffU; |
|
|
676 | m >>= (-125 - e); |
|
|
677 | e = -126; |
|
|
678 | } |
|
|
679 | |
|
|
680 | r |= (e + 126) << 23; |
|
|
681 | r |= m & 0x7fffffU; |
|
|
682 | #endif |
|
|
683 | |
|
|
684 | return r; |
|
|
685 | } |
|
|
686 | |
|
|
687 | /* converts an ieee single/binary32 to a float */ |
|
|
688 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
689 | ecb_function_ float |
|
|
690 | ecb_binary32_to_float (uint32_t x) |
|
|
691 | { |
|
|
692 | float r; |
|
|
693 | |
|
|
694 | #if ECB_STDFP |
|
|
695 | memcpy (&r, &x, 4); |
|
|
696 | #else |
|
|
697 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
698 | int neg = x >> 31; |
|
|
699 | int e = (x >> 23) & 0xffU; |
|
|
700 | |
|
|
701 | x &= 0x7fffffU; |
|
|
702 | |
|
|
703 | if (e) |
|
|
704 | x |= 0x800000U; |
|
|
705 | else |
|
|
706 | e = 1; |
|
|
707 | |
|
|
708 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
709 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
710 | |
|
|
711 | r = neg ? -r : r; |
|
|
712 | #endif |
|
|
713 | |
|
|
714 | return r; |
|
|
715 | } |
|
|
716 | |
|
|
717 | /* convert a double to ieee double/binary64 */ |
|
|
718 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
719 | ecb_function_ uint64_t |
|
|
720 | ecb_double_to_binary64 (double x) |
|
|
721 | { |
|
|
722 | uint64_t r; |
|
|
723 | |
|
|
724 | #if ECB_STDFP |
|
|
725 | memcpy (&r, &x, 8); |
|
|
726 | #else |
|
|
727 | /* slow emulation, works for anything but -0 */ |
|
|
728 | uint64_t m; |
|
|
729 | int e; |
|
|
730 | |
|
|
731 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
732 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
733 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
734 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
735 | |
|
|
736 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
737 | |
|
|
738 | r = m & 0x8000000000000000;; |
|
|
739 | |
|
|
740 | if (r) |
|
|
741 | m = -m; |
|
|
742 | |
|
|
743 | if (e <= -1022) |
|
|
744 | { |
|
|
745 | m &= 0x1fffffffffffffU; |
|
|
746 | m >>= (-1021 - e); |
|
|
747 | e = -1022; |
|
|
748 | } |
|
|
749 | |
|
|
750 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
751 | r |= m & 0xfffffffffffffU; |
|
|
752 | #endif |
|
|
753 | |
|
|
754 | return r; |
|
|
755 | } |
|
|
756 | |
|
|
757 | /* converts an ieee double/binary64 to a double */ |
|
|
758 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
759 | ecb_function_ double |
|
|
760 | ecb_binary64_to_double (uint64_t x) |
|
|
761 | { |
|
|
762 | double r; |
|
|
763 | |
|
|
764 | #if ECB_STDFP |
|
|
765 | memcpy (&r, &x, 8); |
|
|
766 | #else |
|
|
767 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
768 | int neg = x >> 63; |
|
|
769 | int e = (x >> 52) & 0x7ffU; |
|
|
770 | |
|
|
771 | x &= 0xfffffffffffffU; |
|
|
772 | |
|
|
773 | if (e) |
|
|
774 | x |= 0x10000000000000U; |
|
|
775 | else |
|
|
776 | e = 1; |
|
|
777 | |
|
|
778 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
779 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
780 | |
|
|
781 | r = neg ? -r : r; |
|
|
782 | #endif |
|
|
783 | |
|
|
784 | return r; |
|
|
785 | } |
|
|
786 | |
|
|
787 | #endif |
|
|
788 | |
|
|
789 | #endif |
|
|
790 | |
