… | |
… | |
28 | */ |
28 | */ |
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 */ |
|
|
34 | #define ECB_VERSION 0x00010002 |
|
|
35 | |
33 | #ifdef _WIN32 |
36 | #ifdef _WIN32 |
34 | typedef signed char int8_t; |
37 | typedef signed char int8_t; |
35 | typedef unsigned char uint8_t; |
38 | typedef unsigned char uint8_t; |
36 | typedef signed short int16_t; |
39 | typedef signed short int16_t; |
37 | typedef unsigned short uint16_t; |
40 | typedef unsigned short uint16_t; |
… | |
… | |
42 | typedef unsigned long long uint64_t; |
45 | typedef unsigned long long uint64_t; |
43 | #else /* _MSC_VER || __BORLANDC__ */ |
46 | #else /* _MSC_VER || __BORLANDC__ */ |
44 | typedef signed __int64 int64_t; |
47 | typedef signed __int64 int64_t; |
45 | typedef unsigned __int64 uint64_t; |
48 | typedef unsigned __int64 uint64_t; |
46 | #endif |
49 | #endif |
|
|
50 | #ifdef _WIN64 |
|
|
51 | #define ECB_PTRSIZE 8 |
|
|
52 | typedef uint64_t uintptr_t; |
|
|
53 | typedef int64_t intptr_t; |
|
|
54 | #else |
|
|
55 | #define ECB_PTRSIZE 4 |
|
|
56 | typedef uint32_t uintptr_t; |
|
|
57 | typedef int32_t intptr_t; |
|
|
58 | #endif |
47 | #else |
59 | #else |
48 | #include <inttypes.h> |
60 | #include <inttypes.h> |
|
|
61 | #if UINTMAX_MAX > 0xffffffffU |
|
|
62 | #define ECB_PTRSIZE 8 |
|
|
63 | #else |
|
|
64 | #define ECB_PTRSIZE 4 |
|
|
65 | #endif |
49 | #endif |
66 | #endif |
50 | |
67 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
68 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
69 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enormous grief in return for some better fake benchmark numbers. |
70 | * causing enormous grief in return for some better fake benchmark numbers. |
… | |
… | |
61 | #else |
78 | #else |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
79 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
63 | #endif |
80 | #endif |
64 | #endif |
81 | #endif |
65 | |
82 | |
|
|
83 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
|
|
84 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
85 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
|
|
86 | #define ECB_CPP (__cplusplus+0) |
|
|
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 |
|
|
98 | |
66 | /*****************************************************************************/ |
99 | /*****************************************************************************/ |
67 | |
100 | |
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
101 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
69 | /* 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 */ |
70 | |
103 | |
71 | #if ECB_NO_THREADS |
104 | #if ECB_NO_THREADS |
72 | # define ECB_NO_SMP 1 |
105 | #define ECB_NO_SMP 1 |
73 | #endif |
106 | #endif |
74 | |
107 | |
75 | #if ECB_NO_THREADS || ECB_NO_SMP |
108 | #if ECB_NO_SMP |
76 | #define ECB_MEMORY_FENCE do { } while (0) |
109 | #define ECB_MEMORY_FENCE do { } while (0) |
77 | #endif |
110 | #endif |
78 | |
111 | |
79 | #ifndef ECB_MEMORY_FENCE |
112 | #ifndef ECB_MEMORY_FENCE |
80 | #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 |
81 | #if __i386 || __i386__ |
114 | #if __i386 || __i386__ |
82 | #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") |
83 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
116 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
84 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
117 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
85 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
118 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
119 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
87 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
120 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
88 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
121 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
89 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
122 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
90 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
123 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
91 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
124 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
92 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
125 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
93 | #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") |
94 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
127 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
95 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
128 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
96 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
97 | #elif __sparc || __sparc__ |
130 | #elif __sparc || __sparc__ |
98 | #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") |
99 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
132 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
100 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
101 | #elif defined __s390__ || defined __s390x__ |
134 | #elif defined __s390__ || defined __s390x__ |
102 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
103 | #elif defined __mips__ |
136 | #elif defined __mips__ |
104 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
137 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
105 | #elif defined __alpha__ |
138 | #elif defined __alpha__ |
106 | #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") |
107 | #endif |
145 | #endif |
108 | #endif |
146 | #endif |
109 | #endif |
147 | #endif |
110 | |
148 | |
111 | #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 | |
|
|
154 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
|
|
155 | * without risking compile time errors with other compilers. We *could* |
|
|
156 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
|
|
157 | * around this shit time and again. |
|
|
158 | * #elif defined __clang && __has_feature (cxx_atomic) |
|
|
159 | * // see comment below (stdatomic.h) about the C11 memory model. |
|
|
160 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
161 | */ |
|
|
162 | |
112 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
163 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
113 | #define ECB_MEMORY_FENCE __sync_synchronize () |
164 | #define ECB_MEMORY_FENCE __sync_synchronize () |
114 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
|
|
115 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
|
|
116 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
165 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
117 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
166 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
118 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
167 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
119 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
168 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
120 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
169 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
130 | #define ECB_MEMORY_FENCE __sync () |
179 | #define ECB_MEMORY_FENCE __sync () |
131 | #endif |
180 | #endif |
132 | #endif |
181 | #endif |
133 | |
182 | |
134 | #ifndef ECB_MEMORY_FENCE |
183 | #ifndef ECB_MEMORY_FENCE |
|
|
184 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
185 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
186 | /* not just C11 atomics and atomic accesses */ |
|
|
187 | #include <stdatomic.h> |
|
|
188 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
189 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
190 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
191 | /* for most usages, or gcc and clang have a bug */ |
|
|
192 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
193 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
194 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
195 | #endif |
|
|
196 | #endif |
|
|
197 | |
|
|
198 | #ifndef ECB_MEMORY_FENCE |
135 | #if !ECB_AVOID_PTHREADS |
199 | #if !ECB_AVOID_PTHREADS |
136 | /* |
200 | /* |
137 | * if you get undefined symbol references to pthread_mutex_lock, |
201 | * if you get undefined symbol references to pthread_mutex_lock, |
138 | * or failure to find pthread.h, then you should implement |
202 | * or failure to find pthread.h, then you should implement |
139 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
203 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
… | |
… | |
157 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
221 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
158 | #endif |
222 | #endif |
159 | |
223 | |
160 | /*****************************************************************************/ |
224 | /*****************************************************************************/ |
161 | |
225 | |
162 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
163 | |
|
|
164 | #if __cplusplus |
226 | #if __cplusplus |
165 | #define ecb_inline static inline |
227 | #define ecb_inline static inline |
166 | #elif ECB_GCC_VERSION(2,5) |
228 | #elif ECB_GCC_VERSION(2,5) |
167 | #define ecb_inline static __inline__ |
229 | #define ecb_inline static __inline__ |
168 | #elif ECB_C99 |
230 | #elif ECB_C99 |
… | |
… | |
206 | #elif ECB_GCC_VERSION(3,0) |
268 | #elif ECB_GCC_VERSION(3,0) |
207 | #define ecb_decltype(x) __typeof(x) |
269 | #define ecb_decltype(x) __typeof(x) |
208 | #endif |
270 | #endif |
209 | |
271 | |
210 | #define ecb_noinline ecb_attribute ((__noinline__)) |
272 | #define ecb_noinline ecb_attribute ((__noinline__)) |
211 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
212 | #define ecb_unused ecb_attribute ((__unused__)) |
273 | #define ecb_unused ecb_attribute ((__unused__)) |
213 | #define ecb_const ecb_attribute ((__const__)) |
274 | #define ecb_const ecb_attribute ((__const__)) |
214 | #define ecb_pure ecb_attribute ((__pure__)) |
275 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
276 | |
|
|
277 | #if ECB_C11 |
|
|
278 | #define ecb_noreturn _Noreturn |
|
|
279 | #else |
|
|
280 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
281 | #endif |
215 | |
282 | |
216 | #if ECB_GCC_VERSION(4,3) |
283 | #if ECB_GCC_VERSION(4,3) |
217 | #define ecb_artificial ecb_attribute ((__artificial__)) |
284 | #define ecb_artificial ecb_attribute ((__artificial__)) |
218 | #define ecb_hot ecb_attribute ((__hot__)) |
285 | #define ecb_hot ecb_attribute ((__hot__)) |
219 | #define ecb_cold ecb_attribute ((__cold__)) |
286 | #define ecb_cold ecb_attribute ((__cold__)) |
… | |
… | |
310 | |
377 | |
311 | return r + ecb_ld32 (x); |
378 | return r + ecb_ld32 (x); |
312 | } |
379 | } |
313 | #endif |
380 | #endif |
314 | |
381 | |
|
|
382 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
383 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
384 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
385 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
386 | |
315 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
387 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
316 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
388 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
317 | { |
389 | { |
318 | return ( (x * 0x0802U & 0x22110U) |
390 | return ( (x * 0x0802U & 0x22110U) |
319 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
391 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
… | |
… | |
403 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
475 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
404 | ecb_inline void ecb_unreachable (void) { } |
476 | ecb_inline void ecb_unreachable (void) { } |
405 | #endif |
477 | #endif |
406 | |
478 | |
407 | /* try to tell the compiler that some condition is definitely true */ |
479 | /* try to tell the compiler that some condition is definitely true */ |
408 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
480 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
409 | |
481 | |
410 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
482 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
411 | ecb_inline unsigned char |
483 | ecb_inline unsigned char |
412 | ecb_byteorder_helper (void) |
484 | ecb_byteorder_helper (void) |
413 | { |
485 | { |
414 | const uint32_t u = 0x11223344; |
486 | /* the union code still generates code under pressure in gcc, */ |
415 | return *(unsigned char *)&u; |
487 | /* but less than using pointers, and always seems to */ |
|
|
488 | /* successfully return a constant. */ |
|
|
489 | /* the reason why we have this horrible preprocessor mess */ |
|
|
490 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
491 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
492 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
493 | return 0x44; |
|
|
494 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
495 | return 0x44; |
|
|
496 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
497 | return 0x11; |
|
|
498 | #else |
|
|
499 | union |
|
|
500 | { |
|
|
501 | uint32_t i; |
|
|
502 | uint8_t c; |
|
|
503 | } u = { 0x11223344 }; |
|
|
504 | return u.c; |
|
|
505 | #endif |
416 | } |
506 | } |
417 | |
507 | |
418 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
508 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
419 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
509 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
420 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
510 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
… | |
… | |
451 | } |
541 | } |
452 | #else |
542 | #else |
453 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
543 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
454 | #endif |
544 | #endif |
455 | |
545 | |
|
|
546 | /*******************************************************************************/ |
|
|
547 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
548 | |
|
|
549 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
550 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
551 | #if 0 \ |
|
|
552 | || __i386 || __i386__ \ |
|
|
553 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
554 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
555 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
556 | || defined __s390__ || defined __s390x__ \ |
|
|
557 | || defined __mips__ \ |
|
|
558 | || defined __alpha__ \ |
|
|
559 | || defined __hppa__ \ |
|
|
560 | || defined __ia64__ \ |
|
|
561 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
|
|
562 | #define ECB_STDFP 1 |
|
|
563 | #include <string.h> /* for memcpy */ |
|
|
564 | #else |
|
|
565 | #define ECB_STDFP 0 |
|
|
566 | #include <math.h> /* for frexp*, ldexp* */ |
|
|
567 | #endif |
|
|
568 | |
|
|
569 | #ifndef ECB_NO_LIBM |
|
|
570 | |
|
|
571 | /* convert a float to ieee single/binary32 */ |
|
|
572 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
573 | ecb_function_ uint32_t |
|
|
574 | ecb_float_to_binary32 (float x) |
|
|
575 | { |
|
|
576 | uint32_t r; |
|
|
577 | |
|
|
578 | #if ECB_STDFP |
|
|
579 | memcpy (&r, &x, 4); |
|
|
580 | #else |
|
|
581 | /* slow emulation, works for anything but -0 */ |
|
|
582 | uint32_t m; |
|
|
583 | int e; |
|
|
584 | |
|
|
585 | if (x == 0e0f ) return 0x00000000U; |
|
|
586 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
587 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
588 | if (x != x ) return 0x7fbfffffU; |
|
|
589 | |
|
|
590 | m = frexpf (x, &e) * 0x1000000U; |
|
|
591 | |
|
|
592 | r = m & 0x80000000U; |
|
|
593 | |
|
|
594 | if (r) |
|
|
595 | m = -m; |
|
|
596 | |
|
|
597 | if (e <= -126) |
|
|
598 | { |
|
|
599 | m &= 0xffffffU; |
|
|
600 | m >>= (-125 - e); |
|
|
601 | e = -126; |
|
|
602 | } |
|
|
603 | |
|
|
604 | r |= (e + 126) << 23; |
|
|
605 | r |= m & 0x7fffffU; |
456 | #endif |
606 | #endif |
457 | |
607 | |
|
|
608 | return r; |
|
|
609 | } |
|
|
610 | |
|
|
611 | /* converts an ieee single/binary32 to a float */ |
|
|
612 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
613 | ecb_function_ float |
|
|
614 | ecb_binary32_to_float (uint32_t x) |
|
|
615 | { |
|
|
616 | float r; |
|
|
617 | |
|
|
618 | #if ECB_STDFP |
|
|
619 | memcpy (&r, &x, 4); |
|
|
620 | #else |
|
|
621 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
622 | int neg = x >> 31; |
|
|
623 | int e = (x >> 23) & 0xffU; |
|
|
624 | |
|
|
625 | x &= 0x7fffffU; |
|
|
626 | |
|
|
627 | if (e) |
|
|
628 | x |= 0x800000U; |
|
|
629 | else |
|
|
630 | e = 1; |
|
|
631 | |
|
|
632 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
633 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
634 | |
|
|
635 | r = neg ? -r : r; |
|
|
636 | #endif |
|
|
637 | |
|
|
638 | return r; |
|
|
639 | } |
|
|
640 | |
|
|
641 | /* convert a double to ieee double/binary64 */ |
|
|
642 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
643 | ecb_function_ uint64_t |
|
|
644 | ecb_double_to_binary64 (double x) |
|
|
645 | { |
|
|
646 | uint64_t r; |
|
|
647 | |
|
|
648 | #if ECB_STDFP |
|
|
649 | memcpy (&r, &x, 8); |
|
|
650 | #else |
|
|
651 | /* slow emulation, works for anything but -0 */ |
|
|
652 | uint64_t m; |
|
|
653 | int e; |
|
|
654 | |
|
|
655 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
656 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
657 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
658 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
659 | |
|
|
660 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
661 | |
|
|
662 | r = m & 0x8000000000000000;; |
|
|
663 | |
|
|
664 | if (r) |
|
|
665 | m = -m; |
|
|
666 | |
|
|
667 | if (e <= -1022) |
|
|
668 | { |
|
|
669 | m &= 0x1fffffffffffffU; |
|
|
670 | m >>= (-1021 - e); |
|
|
671 | e = -1022; |
|
|
672 | } |
|
|
673 | |
|
|
674 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
675 | r |= m & 0xfffffffffffffU; |
|
|
676 | #endif |
|
|
677 | |
|
|
678 | return r; |
|
|
679 | } |
|
|
680 | |
|
|
681 | /* converts an ieee double/binary64 to a double */ |
|
|
682 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
683 | ecb_function_ double |
|
|
684 | ecb_binary64_to_double (uint64_t x) |
|
|
685 | { |
|
|
686 | double r; |
|
|
687 | |
|
|
688 | #if ECB_STDFP |
|
|
689 | memcpy (&r, &x, 8); |
|
|
690 | #else |
|
|
691 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
692 | int neg = x >> 63; |
|
|
693 | int e = (x >> 52) & 0x7ffU; |
|
|
694 | |
|
|
695 | x &= 0xfffffffffffffU; |
|
|
696 | |
|
|
697 | if (e) |
|
|
698 | x |= 0x10000000000000U; |
|
|
699 | else |
|
|
700 | e = 1; |
|
|
701 | |
|
|
702 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
703 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
704 | |
|
|
705 | r = neg ? -r : r; |
|
|
706 | #endif |
|
|
707 | |
|
|
708 | return r; |
|
|
709 | } |
|
|
710 | |
|
|
711 | #endif |
|
|
712 | |
|
|
713 | #endif |
|
|
714 | |