| 1 | /* |
1 | /* |
| 2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
| 3 | * |
3 | * |
| 4 | * Copyright (©) 2009-2013 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2014 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: |
| … | |
… | |
| 23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
| 24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
| 25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
| 26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
| 27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
28 | * |
|
|
29 | * Alternatively, the contents of this file may be used under the terms of |
|
|
30 | * the GNU General Public License ("GPL") version 2 or any later version, |
|
|
31 | * in which case the provisions of the GPL are applicable instead of |
|
|
32 | * the above. If you wish to allow the use of your version of this file |
|
|
33 | * only under the terms of the GPL and not to allow others to use your |
|
|
34 | * version of this file under the BSD license, indicate your decision |
|
|
35 | * by deleting the provisions above and replace them with the notice |
|
|
36 | * and other provisions required by the GPL. If you do not delete the |
|
|
37 | * provisions above, a recipient may use your version of this file under |
|
|
38 | * either the BSD or the GPL. |
| 28 | */ |
39 | */ |
| 29 | |
40 | |
| 30 | #ifndef ECB_H |
41 | #ifndef ECB_H |
| 31 | #define ECB_H |
42 | #define ECB_H |
| 32 | |
43 | |
| 33 | /* 16 bits major, 16 bits minor */ |
44 | /* 16 bits major, 16 bits minor */ |
| 34 | #define ECB_VERSION 0x00010003 |
45 | #define ECB_VERSION 0x00010004 |
| 35 | |
46 | |
| 36 | #ifdef _WIN32 |
47 | #ifdef _WIN32 |
| 37 | typedef signed char int8_t; |
48 | typedef signed char int8_t; |
| 38 | typedef unsigned char uint8_t; |
49 | typedef unsigned char uint8_t; |
| 39 | typedef signed short int16_t; |
50 | typedef signed short int16_t; |
| … | |
… | |
| 79 | * causing enormous grief in return for some better fake benchmark numbers. |
90 | * causing enormous grief in return for some better fake benchmark numbers. |
| 80 | * or so. |
91 | * or so. |
| 81 | * we try to detect these and simply assume they are not gcc - if they have |
92 | * we try to detect these and simply assume they are not gcc - if they have |
| 82 | * an issue with that they should have done it right in the first place. |
93 | * an issue with that they should have done it right in the first place. |
| 83 | */ |
94 | */ |
| 84 | #ifndef ECB_GCC_VERSION |
|
|
| 85 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
95 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
| 86 | #define ECB_GCC_VERSION(major,minor) 0 |
96 | #define ECB_GCC_VERSION(major,minor) 0 |
| 87 | #else |
97 | #else |
| 88 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
98 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 89 | #endif |
99 | #endif |
| 90 | #endif |
|
|
| 91 | |
100 | |
| 92 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
101 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
| 93 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
102 | |
| 94 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
103 | #if __clang__ && defined __has_builtin |
|
|
104 | #define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
|
|
105 | #else |
|
|
106 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
107 | #endif |
|
|
108 | |
|
|
109 | #if __clang__ && defined __has_extension |
|
|
110 | #define ECB_CLANG_EXTENSION(x) __has_extension (x) |
|
|
111 | #else |
|
|
112 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
113 | #endif |
|
|
114 | |
| 95 | #define ECB_CPP (__cplusplus+0) |
115 | #define ECB_CPP (__cplusplus+0) |
| 96 | #define ECB_CPP11 (__cplusplus >= 201103L) |
116 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
117 | |
|
|
118 | #if ECB_CPP |
|
|
119 | #define ECB_C 0 |
|
|
120 | #define ECB_STDC_VERSION 0 |
|
|
121 | #else |
|
|
122 | #define ECB_C 1 |
|
|
123 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
124 | #endif |
|
|
125 | |
|
|
126 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
127 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
| 97 | |
128 | |
| 98 | #if ECB_CPP |
129 | #if ECB_CPP |
| 99 | #define ECB_EXTERN_C extern "C" |
130 | #define ECB_EXTERN_C extern "C" |
| 100 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
131 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
| 101 | #define ECB_EXTERN_C_END } |
132 | #define ECB_EXTERN_C_END } |
| … | |
… | |
| 134 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
165 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
| 135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
166 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
| 136 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
167 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
| 137 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
168 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
| 138 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
170 | #elif __aarch64__ |
|
|
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 139 | #elif (__sparc || __sparc__) && !__sparcv8 |
172 | #elif (__sparc || __sparc__) && !__sparcv8 |
| 140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
173 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
| 141 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
174 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
| 142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
175 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
| 143 | #elif defined __s390__ || defined __s390x__ |
176 | #elif defined __s390__ || defined __s390x__ |
| … | |
… | |
| 165 | |
198 | |
| 166 | #ifndef ECB_MEMORY_FENCE |
199 | #ifndef ECB_MEMORY_FENCE |
| 167 | #if ECB_GCC_VERSION(4,7) |
200 | #if ECB_GCC_VERSION(4,7) |
| 168 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
201 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
| 169 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
202 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
203 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
204 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
| 170 | |
205 | |
| 171 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
206 | #elif ECB_CLANG_EXTENSION(c_atomic) |
| 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. |
|
|
| 175 | * #elif defined __clang && __has_feature (cxx_atomic) |
|
|
| 176 | * // see comment below (stdatomic.h) about the C11 memory model. |
207 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
| 177 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
208 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 178 | */ |
209 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
210 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
| 179 | |
211 | |
| 180 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
212 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 181 | #define ECB_MEMORY_FENCE __sync_synchronize () |
213 | #define ECB_MEMORY_FENCE __sync_synchronize () |
| 182 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
214 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 183 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
215 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
| … | |
… | |
| 212 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
244 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
| 213 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
245 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
| 214 | /* for most usages, or gcc and clang have a bug */ |
246 | /* for most usages, or gcc and clang have a bug */ |
| 215 | /* I *currently* lean towards the latter, and inefficiently implement */ |
247 | /* I *currently* lean towards the latter, and inefficiently implement */ |
| 216 | /* all three of ecb's fences as a seq_cst fence */ |
248 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
249 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
250 | /* for all __atomic_thread_fence's except seq_cst */ |
| 217 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
251 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
| 218 | #endif |
252 | #endif |
| 219 | #endif |
253 | #endif |
| 220 | |
254 | |
| 221 | #ifndef ECB_MEMORY_FENCE |
255 | #ifndef ECB_MEMORY_FENCE |
| … | |
… | |
| 244 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
278 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 245 | #endif |
279 | #endif |
| 246 | |
280 | |
| 247 | /*****************************************************************************/ |
281 | /*****************************************************************************/ |
| 248 | |
282 | |
| 249 | #if __cplusplus |
283 | #if ECB_CPP |
| 250 | #define ecb_inline static inline |
284 | #define ecb_inline static inline |
| 251 | #elif ECB_GCC_VERSION(2,5) |
285 | #elif ECB_GCC_VERSION(2,5) |
| 252 | #define ecb_inline static __inline__ |
286 | #define ecb_inline static __inline__ |
| 253 | #elif ECB_C99 |
287 | #elif ECB_C99 |
| 254 | #define ecb_inline static inline |
288 | #define ecb_inline static inline |
| … | |
… | |
| 271 | #define ECB_STRINGIFY_(a) # a |
305 | #define ECB_STRINGIFY_(a) # a |
| 272 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
306 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
| 273 | |
307 | |
| 274 | #define ecb_function_ ecb_inline |
308 | #define ecb_function_ ecb_inline |
| 275 | |
309 | |
| 276 | #if ECB_GCC_VERSION(3,1) |
310 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
| 277 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
311 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
312 | #else |
|
|
313 | #define ecb_attribute(attrlist) |
|
|
314 | #endif |
|
|
315 | |
|
|
316 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
| 278 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
317 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
318 | #else |
|
|
319 | /* possible C11 impl for integral types |
|
|
320 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
321 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
322 | |
|
|
323 | #define ecb_is_constant(expr) 0 |
|
|
324 | #endif |
|
|
325 | |
|
|
326 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
| 279 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
327 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
328 | #else |
|
|
329 | #define ecb_expect(expr,value) (expr) |
|
|
330 | #endif |
|
|
331 | |
|
|
332 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
| 280 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
333 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 281 | #else |
334 | #else |
| 282 | #define ecb_attribute(attrlist) |
|
|
| 283 | #define ecb_is_constant(expr) 0 |
|
|
| 284 | #define ecb_expect(expr,value) (expr) |
|
|
| 285 | #define ecb_prefetch(addr,rw,locality) |
335 | #define ecb_prefetch(addr,rw,locality) |
| 286 | #endif |
336 | #endif |
| 287 | |
337 | |
| 288 | /* no emulation for ecb_decltype */ |
338 | /* no emulation for ecb_decltype */ |
| 289 | #if ECB_GCC_VERSION(4,5) |
339 | #if ECB_CPP11 |
|
|
340 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
341 | template<class T> struct ecb_decltype_t { typedef T type; }; |
| 290 | #define ecb_decltype(x) __decltype(x) |
342 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
| 291 | #elif ECB_GCC_VERSION(3,0) |
343 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
| 292 | #define ecb_decltype(x) __typeof(x) |
344 | #define ecb_decltype(x) __typeof__ (x) |
|
|
345 | #endif |
|
|
346 | |
|
|
347 | #if _MSC_VER >= 1300 |
|
|
348 | #define ecb_deprecated __declspec (deprecated) |
|
|
349 | #else |
|
|
350 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
| 293 | #endif |
351 | #endif |
| 294 | |
352 | |
| 295 | #define ecb_noinline ecb_attribute ((__noinline__)) |
353 | #define ecb_noinline ecb_attribute ((__noinline__)) |
| 296 | #define ecb_unused ecb_attribute ((__unused__)) |
354 | #define ecb_unused ecb_attribute ((__unused__)) |
| 297 | #define ecb_const ecb_attribute ((__const__)) |
355 | #define ecb_const ecb_attribute ((__const__)) |
| 298 | #define ecb_pure ecb_attribute ((__pure__)) |
356 | #define ecb_pure ecb_attribute ((__pure__)) |
| 299 | |
357 | |
| 300 | #if ECB_C11 |
358 | /* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */ |
|
|
359 | #if ECB_C11 || __IBMC_NORETURN |
|
|
360 | /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */ |
| 301 | #define ecb_noreturn _Noreturn |
361 | #define ecb_noreturn _Noreturn |
| 302 | #else |
362 | #else |
| 303 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
363 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
| 304 | #endif |
364 | #endif |
| 305 | |
365 | |
| … | |
… | |
| 321 | /* for compatibility to the rest of the world */ |
381 | /* for compatibility to the rest of the world */ |
| 322 | #define ecb_likely(expr) ecb_expect_true (expr) |
382 | #define ecb_likely(expr) ecb_expect_true (expr) |
| 323 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
383 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
| 324 | |
384 | |
| 325 | /* count trailing zero bits and count # of one bits */ |
385 | /* count trailing zero bits and count # of one bits */ |
| 326 | #if ECB_GCC_VERSION(3,4) |
386 | #if ECB_GCC_VERSION(3,4) \ |
|
|
387 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
388 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
389 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
| 327 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
390 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
| 328 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
391 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
| 329 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
392 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
| 330 | #define ecb_ctz32(x) __builtin_ctz (x) |
393 | #define ecb_ctz32(x) __builtin_ctz (x) |
| 331 | #define ecb_ctz64(x) __builtin_ctzll (x) |
394 | #define ecb_ctz64(x) __builtin_ctzll (x) |
| … | |
… | |
| 462 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
525 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
| 463 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
526 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
| 464 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
527 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
| 465 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
528 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
| 466 | |
529 | |
| 467 | #if ECB_GCC_VERSION(4,3) |
530 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
| 468 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
531 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
| 469 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
532 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
| 470 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
533 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
| 471 | #else |
534 | #else |
| 472 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
535 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
| … | |
… | |
| 489 | { |
552 | { |
| 490 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
553 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 491 | } |
554 | } |
| 492 | #endif |
555 | #endif |
| 493 | |
556 | |
| 494 | #if ECB_GCC_VERSION(4,5) |
557 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
| 495 | #define ecb_unreachable() __builtin_unreachable () |
558 | #define ecb_unreachable() __builtin_unreachable () |
| 496 | #else |
559 | #else |
| 497 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
560 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 498 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
561 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
| 499 | ecb_inline void ecb_unreachable (void) { } |
562 | ecb_inline void ecb_unreachable (void) { } |
| … | |
… | |
| 537 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
600 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 538 | #else |
601 | #else |
| 539 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
602 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
| 540 | #endif |
603 | #endif |
| 541 | |
604 | |
| 542 | #if __cplusplus |
605 | #if ECB_CPP |
| 543 | template<typename T> |
606 | template<typename T> |
| 544 | static inline T ecb_div_rd (T val, T div) |
607 | static inline T ecb_div_rd (T val, T div) |
| 545 | { |
608 | { |
| 546 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
609 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
| 547 | } |
610 | } |
| … | |
… | |
| 573 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
636 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
| 574 | #if 0 \ |
637 | #if 0 \ |
| 575 | || __i386 || __i386__ \ |
638 | || __i386 || __i386__ \ |
| 576 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
639 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
| 577 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
640 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
| 578 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
| 579 | || defined __s390__ || defined __s390x__ \ |
641 | || defined __s390__ || defined __s390x__ \ |
| 580 | || defined __mips__ \ |
642 | || defined __mips__ \ |
| 581 | || defined __alpha__ \ |
643 | || defined __alpha__ \ |
| 582 | || defined __hppa__ \ |
644 | || defined __hppa__ \ |
| 583 | || defined __ia64__ \ |
645 | || defined __ia64__ \ |
| 584 | || defined __m68k__ \ |
646 | || defined __m68k__ \ |
| 585 | || defined __m88k__ \ |
647 | || defined __m88k__ \ |
| 586 | || defined __sh__ \ |
648 | || defined __sh__ \ |
| 587 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
649 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
650 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
651 | || defined __aarch64__ |
| 588 | #define ECB_STDFP 1 |
652 | #define ECB_STDFP 1 |
| 589 | #include <string.h> /* for memcpy */ |
653 | #include <string.h> /* for memcpy */ |
| 590 | #else |
654 | #else |
| 591 | #define ECB_STDFP 0 |
655 | #define ECB_STDFP 0 |
| 592 | #endif |
656 | #endif |
| … | |
… | |
| 606 | #define ECB_NAN NAN |
670 | #define ECB_NAN NAN |
| 607 | #else |
671 | #else |
| 608 | #define ECB_NAN ECB_INFINITY |
672 | #define ECB_NAN ECB_INFINITY |
| 609 | #endif |
673 | #endif |
| 610 | |
674 | |
|
|
675 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
|
|
676 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
677 | #else |
|
|
678 | #define ecb_ldexpf(x,e) (float) ldexp ((x), (e)) |
|
|
679 | #endif |
|
|
680 | |
| 611 | /* converts an ieee half/binary16 to a float */ |
681 | /* converts an ieee half/binary16 to a float */ |
| 612 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
682 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
| 613 | ecb_function_ float |
683 | ecb_function_ float |
| 614 | ecb_binary16_to_float (uint16_t x) |
684 | ecb_binary16_to_float (uint16_t x) |
| 615 | { |
685 | { |
| 616 | int e = (x >> 10) & 0x1f; |
686 | int e = (x >> 10) & 0x1f; |
| 617 | int m = x & 0x3ff; |
687 | int m = x & 0x3ff; |
| 618 | float r; |
688 | float r; |
| 619 | |
689 | |
| 620 | if (!e ) r = ldexpf (m , -24); |
690 | if (!e ) r = ecb_ldexpf (m , -24); |
| 621 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
691 | else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25); |
| 622 | else if (m ) r = ECB_NAN; |
692 | else if (m ) r = ECB_NAN; |
| 623 | else r = ECB_INFINITY; |
693 | else r = ECB_INFINITY; |
| 624 | |
694 | |
| 625 | return x & 0x8000 ? -r : r; |
695 | return x & 0x8000 ? -r : r; |
| 626 | } |
696 | } |
| … | |
… | |
| 685 | x |= 0x800000U; |
755 | x |= 0x800000U; |
| 686 | else |
756 | else |
| 687 | e = 1; |
757 | e = 1; |
| 688 | |
758 | |
| 689 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
759 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
| 690 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
760 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
| 691 | |
761 | |
| 692 | r = neg ? -r : r; |
762 | r = neg ? -r : r; |
| 693 | #endif |
763 | #endif |
| 694 | |
764 | |
| 695 | return r; |
765 | return r; |
