| 1 |
/* |
| 2 |
* libecb - http://software.schmorp.de/pkg/libecb |
| 3 |
* |
| 4 |
* Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de> |
| 5 |
* Copyright (©) 2011 Emanuele Giaquinta |
| 6 |
* All rights reserved. |
| 7 |
* |
| 8 |
* Redistribution and use in source and binary forms, with or without modifica- |
| 9 |
* tion, are permitted provided that the following conditions are met: |
| 10 |
* |
| 11 |
* 1. Redistributions of source code must retain the above copyright notice, |
| 12 |
* this list of conditions and the following disclaimer. |
| 13 |
* |
| 14 |
* 2. Redistributions in binary form must reproduce the above copyright |
| 15 |
* notice, this list of conditions and the following disclaimer in the |
| 16 |
* documentation and/or other materials provided with the distribution. |
| 17 |
* |
| 18 |
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED |
| 19 |
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
| 20 |
* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
| 21 |
* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- |
| 22 |
* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
| 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, |
| 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 |
| 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. |
| 39 |
*/ |
| 40 |
|
| 41 |
#ifndef ECB_H |
| 42 |
#define ECB_H |
| 43 |
|
| 44 |
/* 16 bits major, 16 bits minor */ |
| 45 |
#define ECB_VERSION 0x00010006 |
| 46 |
|
| 47 |
#ifdef _WIN32 |
| 48 |
typedef signed char int8_t; |
| 49 |
typedef unsigned char uint8_t; |
| 50 |
typedef signed short int16_t; |
| 51 |
typedef unsigned short uint16_t; |
| 52 |
typedef signed int int32_t; |
| 53 |
typedef unsigned int uint32_t; |
| 54 |
#if __GNUC__ |
| 55 |
typedef signed long long int64_t; |
| 56 |
typedef unsigned long long uint64_t; |
| 57 |
#else /* _MSC_VER || __BORLANDC__ */ |
| 58 |
typedef signed __int64 int64_t; |
| 59 |
typedef unsigned __int64 uint64_t; |
| 60 |
#endif |
| 61 |
#ifdef _WIN64 |
| 62 |
#define ECB_PTRSIZE 8 |
| 63 |
typedef uint64_t uintptr_t; |
| 64 |
typedef int64_t intptr_t; |
| 65 |
#else |
| 66 |
#define ECB_PTRSIZE 4 |
| 67 |
typedef uint32_t uintptr_t; |
| 68 |
typedef int32_t intptr_t; |
| 69 |
#endif |
| 70 |
#else |
| 71 |
#include <inttypes.h> |
| 72 |
#if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
| 73 |
#define ECB_PTRSIZE 8 |
| 74 |
#else |
| 75 |
#define ECB_PTRSIZE 4 |
| 76 |
#endif |
| 77 |
#endif |
| 78 |
|
| 79 |
#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
| 80 |
#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
| 81 |
|
| 82 |
/* work around x32 idiocy by defining proper macros */ |
| 83 |
#if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
| 84 |
#if _ILP32 |
| 85 |
#define ECB_AMD64_X32 1 |
| 86 |
#else |
| 87 |
#define ECB_AMD64 1 |
| 88 |
#endif |
| 89 |
#endif |
| 90 |
|
| 91 |
/* many compilers define _GNUC_ to some versions but then only implement |
| 92 |
* what their idiot authors think are the "more important" extensions, |
| 93 |
* causing enormous grief in return for some better fake benchmark numbers. |
| 94 |
* or so. |
| 95 |
* we try to detect these and simply assume they are not gcc - if they have |
| 96 |
* an issue with that they should have done it right in the first place. |
| 97 |
*/ |
| 98 |
#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
| 99 |
#define ECB_GCC_VERSION(major,minor) 0 |
| 100 |
#else |
| 101 |
#define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 102 |
#endif |
| 103 |
|
| 104 |
#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
| 105 |
|
| 106 |
#if __clang__ && defined __has_builtin |
| 107 |
#define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
| 108 |
#else |
| 109 |
#define ECB_CLANG_BUILTIN(x) 0 |
| 110 |
#endif |
| 111 |
|
| 112 |
#if __clang__ && defined __has_extension |
| 113 |
#define ECB_CLANG_EXTENSION(x) __has_extension (x) |
| 114 |
#else |
| 115 |
#define ECB_CLANG_EXTENSION(x) 0 |
| 116 |
#endif |
| 117 |
|
| 118 |
#define ECB_CPP (__cplusplus+0) |
| 119 |
#define ECB_CPP11 (__cplusplus >= 201103L) |
| 120 |
#define ECB_CPP14 (__cplusplus >= 201402L) |
| 121 |
#define ECB_CPP17 (__cplusplus >= 201703L) |
| 122 |
|
| 123 |
#if ECB_CPP |
| 124 |
#define ECB_C 0 |
| 125 |
#define ECB_STDC_VERSION 0 |
| 126 |
#else |
| 127 |
#define ECB_C 1 |
| 128 |
#define ECB_STDC_VERSION __STDC_VERSION__ |
| 129 |
#endif |
| 130 |
|
| 131 |
#define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
| 132 |
#define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
| 133 |
#define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
| 134 |
|
| 135 |
#if ECB_CPP |
| 136 |
#define ECB_EXTERN_C extern "C" |
| 137 |
#define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
| 138 |
#define ECB_EXTERN_C_END } |
| 139 |
#else |
| 140 |
#define ECB_EXTERN_C extern |
| 141 |
#define ECB_EXTERN_C_BEG |
| 142 |
#define ECB_EXTERN_C_END |
| 143 |
#endif |
| 144 |
|
| 145 |
/*****************************************************************************/ |
| 146 |
|
| 147 |
/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
| 148 |
/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
| 149 |
|
| 150 |
#if ECB_NO_THREADS |
| 151 |
#define ECB_NO_SMP 1 |
| 152 |
#endif |
| 153 |
|
| 154 |
#if ECB_NO_SMP |
| 155 |
#define ECB_MEMORY_FENCE do { } while (0) |
| 156 |
#endif |
| 157 |
|
| 158 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
| 159 |
#if __xlC__ && ECB_CPP |
| 160 |
#include <builtins.h> |
| 161 |
#endif |
| 162 |
|
| 163 |
#if 1400 <= _MSC_VER |
| 164 |
#include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
| 165 |
#endif |
| 166 |
|
| 167 |
#ifndef ECB_MEMORY_FENCE |
| 168 |
#if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 169 |
#define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") |
| 170 |
#if __i386 || __i386__ |
| 171 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
| 172 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 173 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 174 |
#elif ECB_GCC_AMD64 |
| 175 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 176 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 177 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 178 |
#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
| 179 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
| 180 |
#elif defined __ARM_ARCH_2__ \ |
| 181 |
|| defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
| 182 |
|| defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
| 183 |
|| defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
| 184 |
|| defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
| 185 |
|| defined __ARM_ARCH_5TEJ__ |
| 186 |
/* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
| 187 |
#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
| 188 |
|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
| 189 |
|| defined __ARM_ARCH_6T2__ |
| 190 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
| 191 |
#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
| 192 |
|| defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
| 193 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
| 194 |
#elif __aarch64__ |
| 195 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 196 |
#elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
| 197 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
| 198 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
| 199 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
| 200 |
#elif defined __s390__ || defined __s390x__ |
| 201 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
| 202 |
#elif defined __mips__ |
| 203 |
/* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
| 204 |
/* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
| 205 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
| 206 |
#elif defined __alpha__ |
| 207 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
| 208 |
#elif defined __hppa__ |
| 209 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 210 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 211 |
#elif defined __ia64__ |
| 212 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
| 213 |
#elif defined __m68k__ |
| 214 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 215 |
#elif defined __m88k__ |
| 216 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
| 217 |
#elif defined __sh__ |
| 218 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 219 |
#endif |
| 220 |
#endif |
| 221 |
#endif |
| 222 |
|
| 223 |
#ifndef ECB_MEMORY_FENCE |
| 224 |
#if ECB_GCC_VERSION(4,7) |
| 225 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
| 226 |
#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 227 |
#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
| 228 |
#define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
| 229 |
#define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) |
| 230 |
|
| 231 |
#elif ECB_CLANG_EXTENSION(c_atomic) |
| 232 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
| 233 |
#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 234 |
#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
| 235 |
#define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
| 236 |
#define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) |
| 237 |
|
| 238 |
#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 239 |
#define ECB_MEMORY_FENCE __sync_synchronize () |
| 240 |
#elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 241 |
/* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
| 242 |
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 243 |
#define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
| 244 |
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
| 245 |
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 246 |
#elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 247 |
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 248 |
#define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 249 |
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 250 |
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 251 |
#elif defined _WIN32 |
| 252 |
#include <WinNT.h> |
| 253 |
#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
| 254 |
#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 255 |
#include <mbarrier.h> |
| 256 |
#define ECB_MEMORY_FENCE __machine_rw_barrier () |
| 257 |
#define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
| 258 |
#define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |
| 259 |
#define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () |
| 260 |
#elif __xlC__ |
| 261 |
#define ECB_MEMORY_FENCE __sync () |
| 262 |
#endif |
| 263 |
#endif |
| 264 |
|
| 265 |
#ifndef ECB_MEMORY_FENCE |
| 266 |
#if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
| 267 |
/* we assume that these memory fences work on all variables/all memory accesses, */ |
| 268 |
/* not just C11 atomics and atomic accesses */ |
| 269 |
#include <stdatomic.h> |
| 270 |
#define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
| 271 |
#define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
| 272 |
#define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
| 273 |
#endif |
| 274 |
#endif |
| 275 |
|
| 276 |
#ifndef ECB_MEMORY_FENCE |
| 277 |
#if !ECB_AVOID_PTHREADS |
| 278 |
/* |
| 279 |
* if you get undefined symbol references to pthread_mutex_lock, |
| 280 |
* or failure to find pthread.h, then you should implement |
| 281 |
* the ECB_MEMORY_FENCE operations for your cpu/compiler |
| 282 |
* OR provide pthread.h and link against the posix thread library |
| 283 |
* of your system. |
| 284 |
*/ |
| 285 |
#include <pthread.h> |
| 286 |
#define ECB_NEEDS_PTHREADS 1 |
| 287 |
#define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
| 288 |
|
| 289 |
static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
| 290 |
#define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
| 291 |
#endif |
| 292 |
#endif |
| 293 |
|
| 294 |
#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
| 295 |
#define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
| 296 |
#endif |
| 297 |
|
| 298 |
#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 299 |
#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 300 |
#endif |
| 301 |
|
| 302 |
#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE |
| 303 |
#define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ |
| 304 |
#endif |
| 305 |
|
| 306 |
/*****************************************************************************/ |
| 307 |
|
| 308 |
#if ECB_CPP |
| 309 |
#define ecb_inline static inline |
| 310 |
#elif ECB_GCC_VERSION(2,5) |
| 311 |
#define ecb_inline static __inline__ |
| 312 |
#elif ECB_C99 |
| 313 |
#define ecb_inline static inline |
| 314 |
#else |
| 315 |
#define ecb_inline static |
| 316 |
#endif |
| 317 |
|
| 318 |
#if ECB_GCC_VERSION(3,3) |
| 319 |
#define ecb_restrict __restrict__ |
| 320 |
#elif ECB_C99 |
| 321 |
#define ecb_restrict restrict |
| 322 |
#else |
| 323 |
#define ecb_restrict |
| 324 |
#endif |
| 325 |
|
| 326 |
typedef int ecb_bool; |
| 327 |
|
| 328 |
#define ECB_CONCAT_(a, b) a ## b |
| 329 |
#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
| 330 |
#define ECB_STRINGIFY_(a) # a |
| 331 |
#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
| 332 |
#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
| 333 |
|
| 334 |
#define ecb_function_ ecb_inline |
| 335 |
|
| 336 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
| 337 |
#define ecb_attribute(attrlist) __attribute__ (attrlist) |
| 338 |
#else |
| 339 |
#define ecb_attribute(attrlist) |
| 340 |
#endif |
| 341 |
|
| 342 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
| 343 |
#define ecb_is_constant(expr) __builtin_constant_p (expr) |
| 344 |
#else |
| 345 |
/* possible C11 impl for integral types |
| 346 |
typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
| 347 |
#define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
| 348 |
|
| 349 |
#define ecb_is_constant(expr) 0 |
| 350 |
#endif |
| 351 |
|
| 352 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
| 353 |
#define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
| 354 |
#else |
| 355 |
#define ecb_expect(expr,value) (expr) |
| 356 |
#endif |
| 357 |
|
| 358 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
| 359 |
#define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 360 |
#else |
| 361 |
#define ecb_prefetch(addr,rw,locality) |
| 362 |
#endif |
| 363 |
|
| 364 |
/* no emulation for ecb_decltype */ |
| 365 |
#if ECB_CPP11 |
| 366 |
// older implementations might have problems with decltype(x)::type, work around it |
| 367 |
template<class T> struct ecb_decltype_t { typedef T type; }; |
| 368 |
#define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
| 369 |
#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
| 370 |
#define ecb_decltype(x) __typeof__ (x) |
| 371 |
#endif |
| 372 |
|
| 373 |
#if _MSC_VER >= 1300 |
| 374 |
#define ecb_deprecated __declspec (deprecated) |
| 375 |
#else |
| 376 |
#define ecb_deprecated ecb_attribute ((__deprecated__)) |
| 377 |
#endif |
| 378 |
|
| 379 |
#if _MSC_VER >= 1500 |
| 380 |
#define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
| 381 |
#elif ECB_GCC_VERSION(4,5) |
| 382 |
#define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
| 383 |
#else |
| 384 |
#define ecb_deprecated_message(msg) ecb_deprecated |
| 385 |
#endif |
| 386 |
|
| 387 |
#if _MSC_VER >= 1400 |
| 388 |
#define ecb_noinline __declspec (noinline) |
| 389 |
#else |
| 390 |
#define ecb_noinline ecb_attribute ((__noinline__)) |
| 391 |
#endif |
| 392 |
|
| 393 |
#define ecb_unused ecb_attribute ((__unused__)) |
| 394 |
#define ecb_const ecb_attribute ((__const__)) |
| 395 |
#define ecb_pure ecb_attribute ((__pure__)) |
| 396 |
|
| 397 |
#if ECB_C11 || __IBMC_NORETURN |
| 398 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
| 399 |
#define ecb_noreturn _Noreturn |
| 400 |
#elif ECB_CPP11 |
| 401 |
#define ecb_noreturn [[noreturn]] |
| 402 |
#elif _MSC_VER >= 1200 |
| 403 |
/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
| 404 |
#define ecb_noreturn __declspec (noreturn) |
| 405 |
#else |
| 406 |
#define ecb_noreturn ecb_attribute ((__noreturn__)) |
| 407 |
#endif |
| 408 |
|
| 409 |
#if ECB_GCC_VERSION(4,3) |
| 410 |
#define ecb_artificial ecb_attribute ((__artificial__)) |
| 411 |
#define ecb_hot ecb_attribute ((__hot__)) |
| 412 |
#define ecb_cold ecb_attribute ((__cold__)) |
| 413 |
#else |
| 414 |
#define ecb_artificial |
| 415 |
#define ecb_hot |
| 416 |
#define ecb_cold |
| 417 |
#endif |
| 418 |
|
| 419 |
/* put around conditional expressions if you are very sure that the */ |
| 420 |
/* expression is mostly true or mostly false. note that these return */ |
| 421 |
/* booleans, not the expression. */ |
| 422 |
#define ecb_expect_false(expr) ecb_expect (!!(expr), 0) |
| 423 |
#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) |
| 424 |
/* for compatibility to the rest of the world */ |
| 425 |
#define ecb_likely(expr) ecb_expect_true (expr) |
| 426 |
#define ecb_unlikely(expr) ecb_expect_false (expr) |
| 427 |
|
| 428 |
/* count trailing zero bits and count # of one bits */ |
| 429 |
#if ECB_GCC_VERSION(3,4) \ |
| 430 |
|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
| 431 |
&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
| 432 |
&& ECB_CLANG_BUILTIN(__builtin_popcount)) |
| 433 |
/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
| 434 |
#define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
| 435 |
#define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
| 436 |
#define ecb_ctz32(x) __builtin_ctz (x) |
| 437 |
#define ecb_ctz64(x) __builtin_ctzll (x) |
| 438 |
#define ecb_popcount32(x) __builtin_popcount (x) |
| 439 |
/* no popcountll */ |
| 440 |
#else |
| 441 |
ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
| 442 |
ecb_function_ ecb_const int |
| 443 |
ecb_ctz32 (uint32_t x) |
| 444 |
{ |
| 445 |
#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
| 446 |
unsigned long r; |
| 447 |
_BitScanForward (&r, x); |
| 448 |
return (int)r; |
| 449 |
#else |
| 450 |
int r = 0; |
| 451 |
|
| 452 |
x &= ~x + 1; /* this isolates the lowest bit */ |
| 453 |
|
| 454 |
#if ECB_branchless_on_i386 |
| 455 |
r += !!(x & 0xaaaaaaaa) << 0; |
| 456 |
r += !!(x & 0xcccccccc) << 1; |
| 457 |
r += !!(x & 0xf0f0f0f0) << 2; |
| 458 |
r += !!(x & 0xff00ff00) << 3; |
| 459 |
r += !!(x & 0xffff0000) << 4; |
| 460 |
#else |
| 461 |
if (x & 0xaaaaaaaa) r += 1; |
| 462 |
if (x & 0xcccccccc) r += 2; |
| 463 |
if (x & 0xf0f0f0f0) r += 4; |
| 464 |
if (x & 0xff00ff00) r += 8; |
| 465 |
if (x & 0xffff0000) r += 16; |
| 466 |
#endif |
| 467 |
|
| 468 |
return r; |
| 469 |
#endif |
| 470 |
} |
| 471 |
|
| 472 |
ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
| 473 |
ecb_function_ ecb_const int |
| 474 |
ecb_ctz64 (uint64_t x) |
| 475 |
{ |
| 476 |
#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
| 477 |
unsigned long r; |
| 478 |
_BitScanForward64 (&r, x); |
| 479 |
return (int)r; |
| 480 |
#else |
| 481 |
int shift = x & 0xffffffff ? 0 : 32; |
| 482 |
return ecb_ctz32 (x >> shift) + shift; |
| 483 |
#endif |
| 484 |
} |
| 485 |
|
| 486 |
ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
| 487 |
ecb_function_ ecb_const int |
| 488 |
ecb_popcount32 (uint32_t x) |
| 489 |
{ |
| 490 |
x -= (x >> 1) & 0x55555555; |
| 491 |
x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
| 492 |
x = ((x >> 4) + x) & 0x0f0f0f0f; |
| 493 |
x *= 0x01010101; |
| 494 |
|
| 495 |
return x >> 24; |
| 496 |
} |
| 497 |
|
| 498 |
ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
| 499 |
ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
| 500 |
{ |
| 501 |
#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
| 502 |
unsigned long r; |
| 503 |
_BitScanReverse (&r, x); |
| 504 |
return (int)r; |
| 505 |
#else |
| 506 |
int r = 0; |
| 507 |
|
| 508 |
if (x >> 16) { x >>= 16; r += 16; } |
| 509 |
if (x >> 8) { x >>= 8; r += 8; } |
| 510 |
if (x >> 4) { x >>= 4; r += 4; } |
| 511 |
if (x >> 2) { x >>= 2; r += 2; } |
| 512 |
if (x >> 1) { r += 1; } |
| 513 |
|
| 514 |
return r; |
| 515 |
#endif |
| 516 |
} |
| 517 |
|
| 518 |
ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
| 519 |
ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
| 520 |
{ |
| 521 |
#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
| 522 |
unsigned long r; |
| 523 |
_BitScanReverse64 (&r, x); |
| 524 |
return (int)r; |
| 525 |
#else |
| 526 |
int r = 0; |
| 527 |
|
| 528 |
if (x >> 32) { x >>= 32; r += 32; } |
| 529 |
|
| 530 |
return r + ecb_ld32 (x); |
| 531 |
#endif |
| 532 |
} |
| 533 |
#endif |
| 534 |
|
| 535 |
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
| 536 |
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
| 537 |
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
| 538 |
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
| 539 |
|
| 540 |
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
| 541 |
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
| 542 |
{ |
| 543 |
return ( (x * 0x0802U & 0x22110U) |
| 544 |
| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
| 545 |
} |
| 546 |
|
| 547 |
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
| 548 |
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
| 549 |
{ |
| 550 |
x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
| 551 |
x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
| 552 |
x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
| 553 |
x = ( x >> 8 ) | ( x << 8); |
| 554 |
|
| 555 |
return x; |
| 556 |
} |
| 557 |
|
| 558 |
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
| 559 |
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
| 560 |
{ |
| 561 |
x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
| 562 |
x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
| 563 |
x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
| 564 |
x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
| 565 |
x = ( x >> 16 ) | ( x << 16); |
| 566 |
|
| 567 |
return x; |
| 568 |
} |
| 569 |
|
| 570 |
/* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 571 |
/* so for this version we are lazy */ |
| 572 |
ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
| 573 |
ecb_function_ ecb_const int |
| 574 |
ecb_popcount64 (uint64_t x) |
| 575 |
{ |
| 576 |
return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
| 577 |
} |
| 578 |
|
| 579 |
ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
| 580 |
ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
| 581 |
ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
| 582 |
ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
| 583 |
ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
| 584 |
ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
| 585 |
ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
| 586 |
ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
| 587 |
|
| 588 |
ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
| 589 |
ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
| 590 |
ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
| 591 |
ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
| 592 |
ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
| 593 |
ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
| 594 |
ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
| 595 |
ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
| 596 |
|
| 597 |
#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
| 598 |
#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
| 599 |
#define ecb_bswap16(x) __builtin_bswap16 (x) |
| 600 |
#else |
| 601 |
#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
| 602 |
#endif |
| 603 |
#define ecb_bswap32(x) __builtin_bswap32 (x) |
| 604 |
#define ecb_bswap64(x) __builtin_bswap64 (x) |
| 605 |
#elif _MSC_VER |
| 606 |
#include <stdlib.h> |
| 607 |
#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
| 608 |
#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
| 609 |
#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
| 610 |
#else |
| 611 |
ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
| 612 |
ecb_function_ ecb_const uint16_t |
| 613 |
ecb_bswap16 (uint16_t x) |
| 614 |
{ |
| 615 |
return ecb_rotl16 (x, 8); |
| 616 |
} |
| 617 |
|
| 618 |
ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
| 619 |
ecb_function_ ecb_const uint32_t |
| 620 |
ecb_bswap32 (uint32_t x) |
| 621 |
{ |
| 622 |
return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
| 623 |
} |
| 624 |
|
| 625 |
ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
| 626 |
ecb_function_ ecb_const uint64_t |
| 627 |
ecb_bswap64 (uint64_t x) |
| 628 |
{ |
| 629 |
return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 630 |
} |
| 631 |
#endif |
| 632 |
|
| 633 |
#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
| 634 |
#define ecb_unreachable() __builtin_unreachable () |
| 635 |
#else |
| 636 |
/* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 637 |
ecb_inline ecb_noreturn void ecb_unreachable (void); |
| 638 |
ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
| 639 |
#endif |
| 640 |
|
| 641 |
/* try to tell the compiler that some condition is definitely true */ |
| 642 |
#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 643 |
|
| 644 |
ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
| 645 |
ecb_inline ecb_const uint32_t |
| 646 |
ecb_byteorder_helper (void) |
| 647 |
{ |
| 648 |
/* the union code still generates code under pressure in gcc, */ |
| 649 |
/* but less than using pointers, and always seems to */ |
| 650 |
/* successfully return a constant. */ |
| 651 |
/* the reason why we have this horrible preprocessor mess */ |
| 652 |
/* is to avoid it in all cases, at least on common architectures */ |
| 653 |
/* or when using a recent enough gcc version (>= 4.6) */ |
| 654 |
#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
| 655 |
|| ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
| 656 |
#define ECB_LITTLE_ENDIAN 1 |
| 657 |
return 0x44332211; |
| 658 |
#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
| 659 |
|| ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
| 660 |
#define ECB_BIG_ENDIAN 1 |
| 661 |
return 0x11223344; |
| 662 |
#else |
| 663 |
union |
| 664 |
{ |
| 665 |
uint8_t c[4]; |
| 666 |
uint32_t u; |
| 667 |
} u = { 0x11, 0x22, 0x33, 0x44 }; |
| 668 |
return u.u; |
| 669 |
#endif |
| 670 |
} |
| 671 |
|
| 672 |
ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
| 673 |
ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
| 674 |
ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
| 675 |
ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
| 676 |
|
| 677 |
#if ECB_GCC_VERSION(3,0) || ECB_C99 |
| 678 |
#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 679 |
#else |
| 680 |
#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
| 681 |
#endif |
| 682 |
|
| 683 |
#if ECB_CPP |
| 684 |
template<typename T> |
| 685 |
static inline T ecb_div_rd (T val, T div) |
| 686 |
{ |
| 687 |
return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
| 688 |
} |
| 689 |
template<typename T> |
| 690 |
static inline T ecb_div_ru (T val, T div) |
| 691 |
{ |
| 692 |
return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
| 693 |
} |
| 694 |
#else |
| 695 |
#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
| 696 |
#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 697 |
#endif |
| 698 |
|
| 699 |
#if ecb_cplusplus_does_not_suck |
| 700 |
/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
| 701 |
template<typename T, int N> |
| 702 |
static inline int ecb_array_length (const T (&arr)[N]) |
| 703 |
{ |
| 704 |
return N; |
| 705 |
} |
| 706 |
#else |
| 707 |
#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 708 |
#endif |
| 709 |
|
| 710 |
ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
| 711 |
ecb_function_ ecb_const uint32_t |
| 712 |
ecb_binary16_to_binary32 (uint32_t x) |
| 713 |
{ |
| 714 |
unsigned int s = (x & 0x8000) << (31 - 15); |
| 715 |
int e = (x >> 10) & 0x001f; |
| 716 |
unsigned int m = x & 0x03ff; |
| 717 |
|
| 718 |
if (ecb_expect_false (e == 31)) |
| 719 |
/* infinity or NaN */ |
| 720 |
e = 255 - (127 - 15); |
| 721 |
else if (ecb_expect_false (!e)) |
| 722 |
{ |
| 723 |
if (ecb_expect_true (!m)) |
| 724 |
/* zero, handled by code below by forcing e to 0 */ |
| 725 |
e = 0 - (127 - 15); |
| 726 |
else |
| 727 |
{ |
| 728 |
/* subnormal, renormalise */ |
| 729 |
unsigned int s = 10 - ecb_ld32 (m); |
| 730 |
|
| 731 |
m = (m << s) & 0x3ff; /* mask implicit bit */ |
| 732 |
e -= s - 1; |
| 733 |
} |
| 734 |
} |
| 735 |
|
| 736 |
/* e and m now are normalised, or zero, (or inf or nan) */ |
| 737 |
e += 127 - 15; |
| 738 |
|
| 739 |
return s | (e << 23) | (m << (23 - 10)); |
| 740 |
} |
| 741 |
|
| 742 |
ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
| 743 |
ecb_function_ ecb_const uint16_t |
| 744 |
ecb_binary32_to_binary16 (uint32_t x) |
| 745 |
{ |
| 746 |
unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
| 747 |
unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
| 748 |
unsigned int m = x & 0x007fffff; |
| 749 |
|
| 750 |
x &= 0x7fffffff; |
| 751 |
|
| 752 |
/* if it's within range of binary16 normals, use fast path */ |
| 753 |
if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
| 754 |
{ |
| 755 |
/* mantissa round-to-even */ |
| 756 |
m += 0x00000fff + ((m >> (23 - 10)) & 1); |
| 757 |
|
| 758 |
/* handle overflow */ |
| 759 |
if (ecb_expect_false (m >= 0x00800000)) |
| 760 |
{ |
| 761 |
m >>= 1; |
| 762 |
e += 1; |
| 763 |
} |
| 764 |
|
| 765 |
return s | (e << 10) | (m >> (23 - 10)); |
| 766 |
} |
| 767 |
|
| 768 |
/* handle large numbers and infinity */ |
| 769 |
if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
| 770 |
return s | 0x7c00; |
| 771 |
|
| 772 |
/* handle zero, subnormals and small numbers */ |
| 773 |
if (ecb_expect_true (x < 0x38800000)) |
| 774 |
{ |
| 775 |
/* zero */ |
| 776 |
if (ecb_expect_true (!x)) |
| 777 |
return s; |
| 778 |
|
| 779 |
/* handle subnormals */ |
| 780 |
|
| 781 |
/* too small, will be zero */ |
| 782 |
if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
| 783 |
return s; |
| 784 |
|
| 785 |
m |= 0x00800000; /* make implicit bit explicit */ |
| 786 |
|
| 787 |
/* very tricky - we need to round to the nearest e (+10) bit value */ |
| 788 |
{ |
| 789 |
unsigned int bits = 14 - e; |
| 790 |
unsigned int half = (1 << (bits - 1)) - 1; |
| 791 |
unsigned int even = (m >> bits) & 1; |
| 792 |
|
| 793 |
/* if this overflows, we will end up with a normalised number */ |
| 794 |
m = (m + half + even) >> bits; |
| 795 |
} |
| 796 |
|
| 797 |
return s | m; |
| 798 |
} |
| 799 |
|
| 800 |
/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
| 801 |
m >>= 13; |
| 802 |
|
| 803 |
return s | 0x7c00 | m | !m; |
| 804 |
} |
| 805 |
|
| 806 |
/*******************************************************************************/ |
| 807 |
/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
| 808 |
|
| 809 |
/* basically, everything uses "ieee pure-endian" floating point numbers */ |
| 810 |
/* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
| 811 |
#if 0 \ |
| 812 |
|| __i386 || __i386__ \ |
| 813 |
|| ECB_GCC_AMD64 \ |
| 814 |
|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
| 815 |
|| defined __s390__ || defined __s390x__ \ |
| 816 |
|| defined __mips__ \ |
| 817 |
|| defined __alpha__ \ |
| 818 |
|| defined __hppa__ \ |
| 819 |
|| defined __ia64__ \ |
| 820 |
|| defined __m68k__ \ |
| 821 |
|| defined __m88k__ \ |
| 822 |
|| defined __sh__ \ |
| 823 |
|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
| 824 |
|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
| 825 |
|| defined __aarch64__ |
| 826 |
#define ECB_STDFP 1 |
| 827 |
#include <string.h> /* for memcpy */ |
| 828 |
#else |
| 829 |
#define ECB_STDFP 0 |
| 830 |
#endif |
| 831 |
|
| 832 |
#ifndef ECB_NO_LIBM |
| 833 |
|
| 834 |
#include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
| 835 |
|
| 836 |
/* only the oldest of old doesn't have this one. solaris. */ |
| 837 |
#ifdef INFINITY |
| 838 |
#define ECB_INFINITY INFINITY |
| 839 |
#else |
| 840 |
#define ECB_INFINITY HUGE_VAL |
| 841 |
#endif |
| 842 |
|
| 843 |
#ifdef NAN |
| 844 |
#define ECB_NAN NAN |
| 845 |
#else |
| 846 |
#define ECB_NAN ECB_INFINITY |
| 847 |
#endif |
| 848 |
|
| 849 |
#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
| 850 |
#define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
| 851 |
#define ecb_frexpf(x,e) frexpf ((x), (e)) |
| 852 |
#else |
| 853 |
#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
| 854 |
#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
| 855 |
#endif |
| 856 |
|
| 857 |
/* convert a float to ieee single/binary32 */ |
| 858 |
ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
| 859 |
ecb_function_ ecb_const uint32_t |
| 860 |
ecb_float_to_binary32 (float x) |
| 861 |
{ |
| 862 |
uint32_t r; |
| 863 |
|
| 864 |
#if ECB_STDFP |
| 865 |
memcpy (&r, &x, 4); |
| 866 |
#else |
| 867 |
/* slow emulation, works for anything but -0 */ |
| 868 |
uint32_t m; |
| 869 |
int e; |
| 870 |
|
| 871 |
if (x == 0e0f ) return 0x00000000U; |
| 872 |
if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
| 873 |
if (x < -3.40282346638528860e+38f) return 0xff800000U; |
| 874 |
if (x != x ) return 0x7fbfffffU; |
| 875 |
|
| 876 |
m = ecb_frexpf (x, &e) * 0x1000000U; |
| 877 |
|
| 878 |
r = m & 0x80000000U; |
| 879 |
|
| 880 |
if (r) |
| 881 |
m = -m; |
| 882 |
|
| 883 |
if (e <= -126) |
| 884 |
{ |
| 885 |
m &= 0xffffffU; |
| 886 |
m >>= (-125 - e); |
| 887 |
e = -126; |
| 888 |
} |
| 889 |
|
| 890 |
r |= (e + 126) << 23; |
| 891 |
r |= m & 0x7fffffU; |
| 892 |
#endif |
| 893 |
|
| 894 |
return r; |
| 895 |
} |
| 896 |
|
| 897 |
/* converts an ieee single/binary32 to a float */ |
| 898 |
ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
| 899 |
ecb_function_ ecb_const float |
| 900 |
ecb_binary32_to_float (uint32_t x) |
| 901 |
{ |
| 902 |
float r; |
| 903 |
|
| 904 |
#if ECB_STDFP |
| 905 |
memcpy (&r, &x, 4); |
| 906 |
#else |
| 907 |
/* emulation, only works for normals and subnormals and +0 */ |
| 908 |
int neg = x >> 31; |
| 909 |
int e = (x >> 23) & 0xffU; |
| 910 |
|
| 911 |
x &= 0x7fffffU; |
| 912 |
|
| 913 |
if (e) |
| 914 |
x |= 0x800000U; |
| 915 |
else |
| 916 |
e = 1; |
| 917 |
|
| 918 |
/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
| 919 |
r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
| 920 |
|
| 921 |
r = neg ? -r : r; |
| 922 |
#endif |
| 923 |
|
| 924 |
return r; |
| 925 |
} |
| 926 |
|
| 927 |
/* convert a double to ieee double/binary64 */ |
| 928 |
ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
| 929 |
ecb_function_ ecb_const uint64_t |
| 930 |
ecb_double_to_binary64 (double x) |
| 931 |
{ |
| 932 |
uint64_t r; |
| 933 |
|
| 934 |
#if ECB_STDFP |
| 935 |
memcpy (&r, &x, 8); |
| 936 |
#else |
| 937 |
/* slow emulation, works for anything but -0 */ |
| 938 |
uint64_t m; |
| 939 |
int e; |
| 940 |
|
| 941 |
if (x == 0e0 ) return 0x0000000000000000U; |
| 942 |
if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
| 943 |
if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
| 944 |
if (x != x ) return 0X7ff7ffffffffffffU; |
| 945 |
|
| 946 |
m = frexp (x, &e) * 0x20000000000000U; |
| 947 |
|
| 948 |
r = m & 0x8000000000000000;; |
| 949 |
|
| 950 |
if (r) |
| 951 |
m = -m; |
| 952 |
|
| 953 |
if (e <= -1022) |
| 954 |
{ |
| 955 |
m &= 0x1fffffffffffffU; |
| 956 |
m >>= (-1021 - e); |
| 957 |
e = -1022; |
| 958 |
} |
| 959 |
|
| 960 |
r |= ((uint64_t)(e + 1022)) << 52; |
| 961 |
r |= m & 0xfffffffffffffU; |
| 962 |
#endif |
| 963 |
|
| 964 |
return r; |
| 965 |
} |
| 966 |
|
| 967 |
/* converts an ieee double/binary64 to a double */ |
| 968 |
ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
| 969 |
ecb_function_ ecb_const double |
| 970 |
ecb_binary64_to_double (uint64_t x) |
| 971 |
{ |
| 972 |
double r; |
| 973 |
|
| 974 |
#if ECB_STDFP |
| 975 |
memcpy (&r, &x, 8); |
| 976 |
#else |
| 977 |
/* emulation, only works for normals and subnormals and +0 */ |
| 978 |
int neg = x >> 63; |
| 979 |
int e = (x >> 52) & 0x7ffU; |
| 980 |
|
| 981 |
x &= 0xfffffffffffffU; |
| 982 |
|
| 983 |
if (e) |
| 984 |
x |= 0x10000000000000U; |
| 985 |
else |
| 986 |
e = 1; |
| 987 |
|
| 988 |
/* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
| 989 |
r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
| 990 |
|
| 991 |
r = neg ? -r : r; |
| 992 |
#endif |
| 993 |
|
| 994 |
return r; |
| 995 |
} |
| 996 |
|
| 997 |
/* convert a float to ieee half/binary16 */ |
| 998 |
ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
| 999 |
ecb_function_ ecb_const uint16_t |
| 1000 |
ecb_float_to_binary16 (float x) |
| 1001 |
{ |
| 1002 |
return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
| 1003 |
} |
| 1004 |
|
| 1005 |
/* convert an ieee half/binary16 to float */ |
| 1006 |
ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
| 1007 |
ecb_function_ ecb_const float |
| 1008 |
ecb_binary16_to_float (uint16_t x) |
| 1009 |
{ |
| 1010 |
return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
| 1011 |
} |
| 1012 |
|
| 1013 |
#endif |
| 1014 |
|
| 1015 |
#endif |
| 1016 |
|
