1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2011 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2012 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: |
… | |
… | |
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 0x00010001 |
|
|
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. |
54 | * or so. |
71 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
72 | * we try to detect these and simply assume they are not gcc - if they have |
56 | * an issue with that they should have done it right in the first place. |
73 | * an issue with that they should have done it right in the first place. |
57 | */ |
74 | */ |
58 | #ifndef ECB_GCC_VERSION |
75 | #ifndef ECB_GCC_VERSION |
59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
76 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
60 | #define ECB_GCC_VERSION(major,minor) 0 |
77 | #define ECB_GCC_VERSION(major,minor) 0 |
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 || ECB_NO_SMP |
104 | #if ECB_NO_THREADS |
|
|
105 | #define ECB_NO_SMP 1 |
|
|
106 | #endif |
|
|
107 | |
|
|
108 | #if ECB_NO_SMP |
72 | #define ECB_MEMORY_FENCE do { } while (0) |
109 | #define ECB_MEMORY_FENCE do { } while (0) |
73 | #define ECB_MEMORY_FENCE_ACQUIRE do { } while (0) |
|
|
74 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) |
|
|
75 | #endif |
110 | #endif |
76 | |
111 | |
77 | #ifndef ECB_MEMORY_FENCE |
112 | #ifndef ECB_MEMORY_FENCE |
78 | #if ECB_GCC_VERSION(2,5) |
113 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
79 | #if __x86 |
114 | #if __i386 || __i386__ |
80 | #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") |
81 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
116 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
82 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
117 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
83 | #elif __amd64 |
118 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
84 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
119 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
85 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
120 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
86 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
121 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
122 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
|
|
123 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
87 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
124 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
88 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) \ |
125 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
|
|
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
89 | || defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
127 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
90 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
128 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
91 | #define ECB_MEMORY_FENCE \ |
129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
92 | do { \ |
130 | #elif __sparc || __sparc__ |
93 | int null = 0; \ |
131 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
94 | __asm__ __volatile__ ("mcr p15,0,%0,c6,c10,5", : "=&r" (null) : : "memory"); \ |
132 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
95 | while (0) |
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
134 | #elif defined __s390__ || defined __s390x__ |
|
|
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
136 | #elif defined __mips__ |
|
|
137 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
138 | #elif defined __alpha__ |
|
|
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") |
96 | #endif |
145 | #endif |
97 | #endif |
146 | #endif |
98 | #endif |
147 | #endif |
99 | |
148 | |
100 | #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 | /*#elif defined __clang && __has_feature (cxx_atomic)*/ |
|
|
154 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
155 | /*#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)*/ |
101 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) |
156 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
102 | #define ECB_MEMORY_FENCE __sync_synchronize () |
157 | #define ECB_MEMORY_FENCE __sync_synchronize () |
103 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
|
|
104 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
|
|
105 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
158 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
106 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
159 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
107 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
160 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
108 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
161 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
109 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
162 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
110 | #elif defined(_WIN32) |
163 | #elif defined _WIN32 |
111 | #include <WinNT.h> |
164 | #include <WinNT.h> |
112 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
165 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
166 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
167 | #include <mbarrier.h> |
|
|
168 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
169 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
170 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
171 | #elif __xlC__ |
|
|
172 | #define ECB_MEMORY_FENCE __sync () |
|
|
173 | #endif |
|
|
174 | #endif |
|
|
175 | |
|
|
176 | #ifndef ECB_MEMORY_FENCE |
|
|
177 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
178 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
179 | /* not just C11 atomics and atomic accesses */ |
|
|
180 | #include <stdatomic.h> |
|
|
181 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
182 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
183 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
184 | /* for most usages, or gcc and clang have a bug */ |
|
|
185 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
186 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
187 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
113 | #endif |
188 | #endif |
114 | #endif |
189 | #endif |
115 | |
190 | |
116 | #ifndef ECB_MEMORY_FENCE |
191 | #ifndef ECB_MEMORY_FENCE |
117 | #if !ECB_AVOID_PTHREADS |
192 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
129 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
204 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
130 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
205 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
131 | #endif |
206 | #endif |
132 | #endif |
207 | #endif |
133 | |
208 | |
134 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
209 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
135 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
210 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
136 | #endif |
211 | #endif |
137 | |
212 | |
138 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
213 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
139 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
214 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
140 | #endif |
215 | #endif |
141 | |
216 | |
142 | /*****************************************************************************/ |
217 | /*****************************************************************************/ |
143 | |
|
|
144 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
145 | |
218 | |
146 | #if __cplusplus |
219 | #if __cplusplus |
147 | #define ecb_inline static inline |
220 | #define ecb_inline static inline |
148 | #elif ECB_GCC_VERSION(2,5) |
221 | #elif ECB_GCC_VERSION(2,5) |
149 | #define ecb_inline static __inline__ |
222 | #define ecb_inline static __inline__ |
… | |
… | |
188 | #elif ECB_GCC_VERSION(3,0) |
261 | #elif ECB_GCC_VERSION(3,0) |
189 | #define ecb_decltype(x) __typeof(x) |
262 | #define ecb_decltype(x) __typeof(x) |
190 | #endif |
263 | #endif |
191 | |
264 | |
192 | #define ecb_noinline ecb_attribute ((__noinline__)) |
265 | #define ecb_noinline ecb_attribute ((__noinline__)) |
193 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
194 | #define ecb_unused ecb_attribute ((__unused__)) |
266 | #define ecb_unused ecb_attribute ((__unused__)) |
195 | #define ecb_const ecb_attribute ((__const__)) |
267 | #define ecb_const ecb_attribute ((__const__)) |
196 | #define ecb_pure ecb_attribute ((__pure__)) |
268 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
269 | |
|
|
270 | #if ECB_C11 |
|
|
271 | #define ecb_noreturn _Noreturn |
|
|
272 | #else |
|
|
273 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
274 | #endif |
197 | |
275 | |
198 | #if ECB_GCC_VERSION(4,3) |
276 | #if ECB_GCC_VERSION(4,3) |
199 | #define ecb_artificial ecb_attribute ((__artificial__)) |
277 | #define ecb_artificial ecb_attribute ((__artificial__)) |
200 | #define ecb_hot ecb_attribute ((__hot__)) |
278 | #define ecb_hot ecb_attribute ((__hot__)) |
201 | #define ecb_cold ecb_attribute ((__cold__)) |
279 | #define ecb_cold ecb_attribute ((__cold__)) |
… | |
… | |
292 | |
370 | |
293 | return r + ecb_ld32 (x); |
371 | return r + ecb_ld32 (x); |
294 | } |
372 | } |
295 | #endif |
373 | #endif |
296 | |
374 | |
|
|
375 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
376 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
377 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
378 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
379 | |
|
|
380 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
381 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
382 | { |
|
|
383 | return ( (x * 0x0802U & 0x22110U) |
|
|
384 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
385 | } |
|
|
386 | |
|
|
387 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
388 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
389 | { |
|
|
390 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
391 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
392 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
393 | x = ( x >> 8 ) | ( x << 8); |
|
|
394 | |
|
|
395 | return x; |
|
|
396 | } |
|
|
397 | |
|
|
398 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
399 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
400 | { |
|
|
401 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
402 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
403 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
404 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
405 | x = ( x >> 16 ) | ( x << 16); |
|
|
406 | |
|
|
407 | return x; |
|
|
408 | } |
|
|
409 | |
297 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
410 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
298 | /* so for this version we are lazy */ |
411 | /* so for this version we are lazy */ |
299 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
412 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
300 | ecb_function_ int |
413 | ecb_function_ int |
301 | ecb_popcount64 (uint64_t x) |
414 | ecb_popcount64 (uint64_t x) |
… | |
… | |
350 | |
463 | |
351 | #if ECB_GCC_VERSION(4,5) |
464 | #if ECB_GCC_VERSION(4,5) |
352 | #define ecb_unreachable() __builtin_unreachable () |
465 | #define ecb_unreachable() __builtin_unreachable () |
353 | #else |
466 | #else |
354 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
467 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
355 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
468 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
356 | ecb_function_ void ecb_unreachable (void) { } |
469 | ecb_inline void ecb_unreachable (void) { } |
357 | #endif |
470 | #endif |
358 | |
471 | |
359 | /* try to tell the compiler that some condition is definitely true */ |
472 | /* try to tell the compiler that some condition is definitely true */ |
360 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
473 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
361 | |
474 | |
362 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
475 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
363 | ecb_function_ unsigned char |
476 | ecb_inline unsigned char |
364 | ecb_byteorder_helper (void) |
477 | ecb_byteorder_helper (void) |
365 | { |
478 | { |
366 | const uint32_t u = 0x11223344; |
479 | /* the union code still generates code under pressure in gcc, */ |
367 | return *(unsigned char *)&u; |
480 | /* but less than using pointers, and always seem to */ |
|
|
481 | /* successfully return a constant. */ |
|
|
482 | /* the reason why we have this horrible preprocessor mess */ |
|
|
483 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
484 | /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ |
|
|
485 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
486 | return 0x44; |
|
|
487 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
488 | return 0x44; |
|
|
489 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
490 | retrurn 0x11; |
|
|
491 | #else |
|
|
492 | union |
|
|
493 | { |
|
|
494 | uint32_t i; |
|
|
495 | uint8_t c; |
|
|
496 | } u = { 0x11223344 }; |
|
|
497 | return u.c; |
|
|
498 | #endif |
368 | } |
499 | } |
369 | |
500 | |
370 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
501 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
371 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
502 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
372 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
503 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
373 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
504 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
374 | |
505 | |
375 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
506 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
376 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
507 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
377 | #else |
508 | #else |
378 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
509 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
510 | #endif |
|
|
511 | |
|
|
512 | #if __cplusplus |
|
|
513 | template<typename T> |
|
|
514 | static inline T ecb_div_rd (T val, T div) |
|
|
515 | { |
|
|
516 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
517 | } |
|
|
518 | template<typename T> |
|
|
519 | static inline T ecb_div_ru (T val, T div) |
|
|
520 | { |
|
|
521 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
522 | } |
|
|
523 | #else |
|
|
524 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
525 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
379 | #endif |
526 | #endif |
380 | |
527 | |
381 | #if ecb_cplusplus_does_not_suck |
528 | #if ecb_cplusplus_does_not_suck |
382 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
529 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
383 | template<typename T, int N> |
530 | template<typename T, int N> |
… | |
… | |
387 | } |
534 | } |
388 | #else |
535 | #else |
389 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
536 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
390 | #endif |
537 | #endif |
391 | |
538 | |
|
|
539 | /*******************************************************************************/ |
|
|
540 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
541 | |
|
|
542 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
543 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
544 | #if 0 \ |
|
|
545 | || __i386 || __i386__ \ |
|
|
546 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
547 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
548 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
549 | || defined __s390__ || defined __s390x__ \ |
|
|
550 | || defined __mips__ \ |
|
|
551 | || defined __alpha__ \ |
|
|
552 | || defined __hppa__ \ |
|
|
553 | || defined __ia64__ \ |
|
|
554 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
|
|
555 | #define ECB_STDFP 1 |
|
|
556 | #include <string.h> /* for memcpy */ |
|
|
557 | #else |
|
|
558 | #define ECB_STDFP 0 |
|
|
559 | #include <math.h> /* for frexp*, ldexp* */ |
|
|
560 | #endif |
|
|
561 | |
|
|
562 | #ifndef ECB_NO_LIBM |
|
|
563 | |
|
|
564 | /* convert a float to ieee single/binary32 */ |
|
|
565 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
566 | ecb_function_ uint32_t |
|
|
567 | ecb_float_to_binary32 (float x) |
|
|
568 | { |
|
|
569 | uint32_t r; |
|
|
570 | |
|
|
571 | #if ECB_STDFP |
|
|
572 | memcpy (&r, &x, 4); |
|
|
573 | #else |
|
|
574 | /* slow emulation, works for anything but -0 */ |
|
|
575 | uint32_t m; |
|
|
576 | int e; |
|
|
577 | |
|
|
578 | if (x == 0e0f ) return 0x00000000U; |
|
|
579 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
580 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
581 | if (x != x ) return 0x7fbfffffU; |
|
|
582 | |
|
|
583 | m = frexpf (x, &e) * 0x1000000U; |
|
|
584 | |
|
|
585 | r = m & 0x80000000U; |
|
|
586 | |
|
|
587 | if (r) |
|
|
588 | m = -m; |
|
|
589 | |
|
|
590 | if (e <= -126) |
|
|
591 | { |
|
|
592 | m &= 0xffffffU; |
|
|
593 | m >>= (-125 - e); |
|
|
594 | e = -126; |
|
|
595 | } |
|
|
596 | |
|
|
597 | r |= (e + 126) << 23; |
|
|
598 | r |= m & 0x7fffffU; |
392 | #endif |
599 | #endif |
393 | |
600 | |
|
|
601 | return r; |
|
|
602 | } |
|
|
603 | |
|
|
604 | /* converts an ieee single/binary32 to a float */ |
|
|
605 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
606 | ecb_function_ float |
|
|
607 | ecb_binary32_to_float (uint32_t x) |
|
|
608 | { |
|
|
609 | float r; |
|
|
610 | |
|
|
611 | #if ECB_STDFP |
|
|
612 | memcpy (&r, &x, 4); |
|
|
613 | #else |
|
|
614 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
615 | int neg = x >> 31; |
|
|
616 | int e = (x >> 23) & 0xffU; |
|
|
617 | |
|
|
618 | x &= 0x7fffffU; |
|
|
619 | |
|
|
620 | if (e) |
|
|
621 | x |= 0x800000U; |
|
|
622 | else |
|
|
623 | e = 1; |
|
|
624 | |
|
|
625 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
626 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
627 | |
|
|
628 | r = neg ? -r : r; |
|
|
629 | #endif |
|
|
630 | |
|
|
631 | return r; |
|
|
632 | } |
|
|
633 | |
|
|
634 | /* convert a double to ieee double/binary64 */ |
|
|
635 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
636 | ecb_function_ uint64_t |
|
|
637 | ecb_double_to_binary64 (double x) |
|
|
638 | { |
|
|
639 | uint64_t r; |
|
|
640 | |
|
|
641 | #if ECB_STDFP |
|
|
642 | memcpy (&r, &x, 8); |
|
|
643 | #else |
|
|
644 | /* slow emulation, works for anything but -0 */ |
|
|
645 | uint64_t m; |
|
|
646 | int e; |
|
|
647 | |
|
|
648 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
649 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
650 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
651 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
652 | |
|
|
653 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
654 | |
|
|
655 | r = m & 0x8000000000000000;; |
|
|
656 | |
|
|
657 | if (r) |
|
|
658 | m = -m; |
|
|
659 | |
|
|
660 | if (e <= -1022) |
|
|
661 | { |
|
|
662 | m &= 0x1fffffffffffffU; |
|
|
663 | m >>= (-1021 - e); |
|
|
664 | e = -1022; |
|
|
665 | } |
|
|
666 | |
|
|
667 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
668 | r |= m & 0xfffffffffffffU; |
|
|
669 | #endif |
|
|
670 | |
|
|
671 | return r; |
|
|
672 | } |
|
|
673 | |
|
|
674 | /* converts an ieee double/binary64 to a double */ |
|
|
675 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
676 | ecb_function_ double |
|
|
677 | ecb_binary64_to_double (uint64_t x) |
|
|
678 | { |
|
|
679 | double r; |
|
|
680 | |
|
|
681 | #if ECB_STDFP |
|
|
682 | memcpy (&r, &x, 8); |
|
|
683 | #else |
|
|
684 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
685 | int neg = x >> 63; |
|
|
686 | int e = (x >> 52) & 0x7ffU; |
|
|
687 | |
|
|
688 | x &= 0xfffffffffffffU; |
|
|
689 | |
|
|
690 | if (e) |
|
|
691 | x |= 0x10000000000000U; |
|
|
692 | else |
|
|
693 | e = 1; |
|
|
694 | |
|
|
695 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
696 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
697 | |
|
|
698 | r = neg ? -r : r; |
|
|
699 | #endif |
|
|
700 | |
|
|
701 | return r; |
|
|
702 | } |
|
|
703 | |
|
|
704 | #endif |
|
|
705 | |
|
|
706 | #endif |
|
|
707 | |