| 1 |
/* |
| 2 |
* libecb - http://software.schmorp.de/pkg/libecb |
| 3 |
* |
| 4 |
* Copyright (©) 2009-2015,2018-2021 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 0x0001000c |
| 46 |
|
| 47 |
#include <string.h> /* for memcpy */ |
| 48 |
|
| 49 |
#if defined (_WIN32) && !defined (__MINGW32__) |
| 50 |
typedef signed char int8_t; |
| 51 |
typedef unsigned char uint8_t; |
| 52 |
typedef signed char int_fast8_t; |
| 53 |
typedef unsigned char uint_fast8_t; |
| 54 |
typedef signed short int16_t; |
| 55 |
typedef unsigned short uint16_t; |
| 56 |
typedef signed int int_fast16_t; |
| 57 |
typedef unsigned int uint_fast16_t; |
| 58 |
typedef signed int int32_t; |
| 59 |
typedef unsigned int uint32_t; |
| 60 |
typedef signed int int_fast32_t; |
| 61 |
typedef unsigned int uint_fast32_t; |
| 62 |
#if __GNUC__ |
| 63 |
typedef signed long long int64_t; |
| 64 |
typedef unsigned long long uint64_t; |
| 65 |
#else /* _MSC_VER || __BORLANDC__ */ |
| 66 |
typedef signed __int64 int64_t; |
| 67 |
typedef unsigned __int64 uint64_t; |
| 68 |
#endif |
| 69 |
typedef int64_t int_fast64_t; |
| 70 |
typedef uint64_t uint_fast64_t; |
| 71 |
#ifdef _WIN64 |
| 72 |
#define ECB_PTRSIZE 8 |
| 73 |
typedef uint64_t uintptr_t; |
| 74 |
typedef int64_t intptr_t; |
| 75 |
#else |
| 76 |
#define ECB_PTRSIZE 4 |
| 77 |
typedef uint32_t uintptr_t; |
| 78 |
typedef int32_t intptr_t; |
| 79 |
#endif |
| 80 |
#else |
| 81 |
#include <inttypes.h> |
| 82 |
#if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
| 83 |
#define ECB_PTRSIZE 8 |
| 84 |
#else |
| 85 |
#define ECB_PTRSIZE 4 |
| 86 |
#endif |
| 87 |
#endif |
| 88 |
|
| 89 |
#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
| 90 |
#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
| 91 |
|
| 92 |
#ifndef ECB_OPTIMIZE_SIZE |
| 93 |
#if __OPTIMIZE_SIZE__ |
| 94 |
#define ECB_OPTIMIZE_SIZE 1 |
| 95 |
#else |
| 96 |
#define ECB_OPTIMIZE_SIZE 0 |
| 97 |
#endif |
| 98 |
#endif |
| 99 |
|
| 100 |
/* work around x32 idiocy by defining proper macros */ |
| 101 |
#if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
| 102 |
#if _ILP32 |
| 103 |
#define ECB_AMD64_X32 1 |
| 104 |
#else |
| 105 |
#define ECB_AMD64 1 |
| 106 |
#endif |
| 107 |
#endif |
| 108 |
|
| 109 |
#if ECB_PTRSIZE >= 8 || ECB_AMD64_X32 |
| 110 |
#define ECB_64BIT_NATIVE 1 |
| 111 |
#else |
| 112 |
#define ECB_64BIT_NATIVE 0 |
| 113 |
#endif |
| 114 |
|
| 115 |
/* many compilers define _GNUC_ to some versions but then only implement |
| 116 |
* what their idiot authors think are the "more important" extensions, |
| 117 |
* causing enormous grief in return for some better fake benchmark numbers. |
| 118 |
* or so. |
| 119 |
* we try to detect these and simply assume they are not gcc - if they have |
| 120 |
* an issue with that they should have done it right in the first place. |
| 121 |
*/ |
| 122 |
#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
| 123 |
#define ECB_GCC_VERSION(major,minor) 0 |
| 124 |
#else |
| 125 |
#define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 126 |
#endif |
| 127 |
|
| 128 |
#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
| 129 |
|
| 130 |
#if __clang__ && defined __has_builtin |
| 131 |
#define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
| 132 |
#else |
| 133 |
#define ECB_CLANG_BUILTIN(x) 0 |
| 134 |
#endif |
| 135 |
|
| 136 |
#if __clang__ && defined __has_extension |
| 137 |
#define ECB_CLANG_EXTENSION(x) __has_extension (x) |
| 138 |
#else |
| 139 |
#define ECB_CLANG_EXTENSION(x) 0 |
| 140 |
#endif |
| 141 |
|
| 142 |
#define ECB_CPP (__cplusplus+0) |
| 143 |
#define ECB_CPP11 (__cplusplus >= 201103L) |
| 144 |
#define ECB_CPP14 (__cplusplus >= 201402L) |
| 145 |
#define ECB_CPP17 (__cplusplus >= 201703L) |
| 146 |
|
| 147 |
#if ECB_CPP |
| 148 |
#define ECB_C 0 |
| 149 |
#define ECB_STDC_VERSION 0 |
| 150 |
#else |
| 151 |
#define ECB_C 1 |
| 152 |
#define ECB_STDC_VERSION __STDC_VERSION__ |
| 153 |
#endif |
| 154 |
|
| 155 |
#define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
| 156 |
#define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
| 157 |
#define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
| 158 |
|
| 159 |
#if ECB_CPP |
| 160 |
#define ECB_EXTERN_C extern "C" |
| 161 |
#define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
| 162 |
#define ECB_EXTERN_C_END } |
| 163 |
#else |
| 164 |
#define ECB_EXTERN_C extern |
| 165 |
#define ECB_EXTERN_C_BEG |
| 166 |
#define ECB_EXTERN_C_END |
| 167 |
#endif |
| 168 |
|
| 169 |
/*****************************************************************************/ |
| 170 |
|
| 171 |
/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
| 172 |
/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
| 173 |
|
| 174 |
#if ECB_NO_THREADS |
| 175 |
#define ECB_NO_SMP 1 |
| 176 |
#endif |
| 177 |
|
| 178 |
#if ECB_NO_SMP |
| 179 |
#define ECB_MEMORY_FENCE do { } while (0) |
| 180 |
#endif |
| 181 |
|
| 182 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
| 183 |
#if __xlC__ && ECB_CPP |
| 184 |
#include <builtins.h> |
| 185 |
#endif |
| 186 |
|
| 187 |
#if 1400 <= _MSC_VER |
| 188 |
#include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
| 189 |
#endif |
| 190 |
|
| 191 |
#ifndef ECB_MEMORY_FENCE |
| 192 |
#if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 193 |
#define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") |
| 194 |
#if __i386 || __i386__ |
| 195 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
| 196 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 197 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 198 |
#elif ECB_GCC_AMD64 |
| 199 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 200 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 201 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 202 |
#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
| 203 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
| 204 |
#elif defined __ARM_ARCH_2__ \ |
| 205 |
|| defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
| 206 |
|| defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
| 207 |
|| defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
| 208 |
|| defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
| 209 |
|| defined __ARM_ARCH_5TEJ__ |
| 210 |
/* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
| 211 |
#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
| 212 |
|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
| 213 |
|| defined __ARM_ARCH_6T2__ |
| 214 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
| 215 |
#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
| 216 |
|| defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
| 217 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
| 218 |
#elif __aarch64__ |
| 219 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
| 220 |
#elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
| 221 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
| 222 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
| 223 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
| 224 |
#elif defined __s390__ || defined __s390x__ |
| 225 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
| 226 |
#elif defined __mips__ |
| 227 |
/* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
| 228 |
/* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
| 229 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
| 230 |
#elif defined __alpha__ |
| 231 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
| 232 |
#elif defined __hppa__ |
| 233 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 234 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 235 |
#elif defined __ia64__ |
| 236 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
| 237 |
#elif defined __m68k__ |
| 238 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 239 |
#elif defined __m88k__ |
| 240 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
| 241 |
#elif defined __sh__ |
| 242 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 243 |
#endif |
| 244 |
#endif |
| 245 |
#endif |
| 246 |
|
| 247 |
#ifndef ECB_MEMORY_FENCE |
| 248 |
#if ECB_GCC_VERSION(4,7) |
| 249 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
| 250 |
#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 251 |
#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
| 252 |
#define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
| 253 |
#undef ECB_MEMORY_FENCE_RELAXED |
| 254 |
#define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) |
| 255 |
|
| 256 |
#elif ECB_CLANG_EXTENSION(c_atomic) |
| 257 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
| 258 |
#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 259 |
#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
| 260 |
#define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
| 261 |
#undef ECB_MEMORY_FENCE_RELAXED |
| 262 |
#define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) |
| 263 |
|
| 264 |
#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 265 |
#define ECB_MEMORY_FENCE __sync_synchronize () |
| 266 |
#elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 267 |
/* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
| 268 |
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 269 |
#define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
| 270 |
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
| 271 |
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 272 |
#elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 273 |
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 274 |
#define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 275 |
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 276 |
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 277 |
#elif defined _WIN32 |
| 278 |
#include <WinNT.h> |
| 279 |
#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
| 280 |
#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 281 |
#include <mbarrier.h> |
| 282 |
#define ECB_MEMORY_FENCE __machine_rw_barrier () |
| 283 |
#define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
| 284 |
#define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |
| 285 |
#define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () |
| 286 |
#elif __xlC__ |
| 287 |
#define ECB_MEMORY_FENCE __sync () |
| 288 |
#endif |
| 289 |
#endif |
| 290 |
|
| 291 |
#ifndef ECB_MEMORY_FENCE |
| 292 |
#if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
| 293 |
/* we assume that these memory fences work on all variables/all memory accesses, */ |
| 294 |
/* not just C11 atomics and atomic accesses */ |
| 295 |
#include <stdatomic.h> |
| 296 |
#define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
| 297 |
#define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
| 298 |
#define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
| 299 |
#endif |
| 300 |
#endif |
| 301 |
|
| 302 |
#ifndef ECB_MEMORY_FENCE |
| 303 |
#if !ECB_AVOID_PTHREADS |
| 304 |
/* |
| 305 |
* if you get undefined symbol references to pthread_mutex_lock, |
| 306 |
* or failure to find pthread.h, then you should implement |
| 307 |
* the ECB_MEMORY_FENCE operations for your cpu/compiler |
| 308 |
* OR provide pthread.h and link against the posix thread library |
| 309 |
* of your system. |
| 310 |
*/ |
| 311 |
#include <pthread.h> |
| 312 |
#define ECB_NEEDS_PTHREADS 1 |
| 313 |
#define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
| 314 |
|
| 315 |
static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
| 316 |
#define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
| 317 |
#endif |
| 318 |
#endif |
| 319 |
|
| 320 |
#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
| 321 |
#define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
| 322 |
#endif |
| 323 |
|
| 324 |
#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 325 |
#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 326 |
#endif |
| 327 |
|
| 328 |
#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE |
| 329 |
#define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ |
| 330 |
#endif |
| 331 |
|
| 332 |
/*****************************************************************************/ |
| 333 |
|
| 334 |
#if ECB_CPP |
| 335 |
#define ecb_inline static inline |
| 336 |
#elif ECB_GCC_VERSION(2,5) |
| 337 |
#define ecb_inline static __inline__ |
| 338 |
#elif ECB_C99 |
| 339 |
#define ecb_inline static inline |
| 340 |
#else |
| 341 |
#define ecb_inline static |
| 342 |
#endif |
| 343 |
|
| 344 |
#if ECB_GCC_VERSION(3,3) |
| 345 |
#define ecb_restrict __restrict__ |
| 346 |
#elif ECB_C99 |
| 347 |
#define ecb_restrict restrict |
| 348 |
#else |
| 349 |
#define ecb_restrict |
| 350 |
#endif |
| 351 |
|
| 352 |
typedef int ecb_bool; |
| 353 |
|
| 354 |
#define ECB_CONCAT_(a, b) a ## b |
| 355 |
#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
| 356 |
#define ECB_STRINGIFY_(a) # a |
| 357 |
#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
| 358 |
#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
| 359 |
|
| 360 |
/* This marks larger functions that do not neccessarily need to be inlined */ |
| 361 |
/* The idea is to possibly compile the header twice, */ |
| 362 |
/* once exposing only the declarations, another time to define external functions */ |
| 363 |
/* TODO: possibly static would be best for these at the moment? */ |
| 364 |
#define ecb_function_ ecb_inline |
| 365 |
|
| 366 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
| 367 |
#define ecb_attribute(attrlist) __attribute__ (attrlist) |
| 368 |
#else |
| 369 |
#define ecb_attribute(attrlist) |
| 370 |
#endif |
| 371 |
|
| 372 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
| 373 |
#define ecb_is_constant(expr) __builtin_constant_p (expr) |
| 374 |
#else |
| 375 |
/* possible C11 impl for integral types |
| 376 |
typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
| 377 |
#define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
| 378 |
|
| 379 |
#define ecb_is_constant(expr) 0 |
| 380 |
#endif |
| 381 |
|
| 382 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
| 383 |
#define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
| 384 |
#else |
| 385 |
#define ecb_expect(expr,value) (expr) |
| 386 |
#endif |
| 387 |
|
| 388 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
| 389 |
#define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 390 |
#else |
| 391 |
#define ecb_prefetch(addr,rw,locality) |
| 392 |
#endif |
| 393 |
|
| 394 |
/* no emulation for ecb_decltype */ |
| 395 |
#if ECB_CPP11 |
| 396 |
// older implementations might have problems with decltype(x)::type, work around it |
| 397 |
template<class T> struct ecb_decltype_t { typedef T type; }; |
| 398 |
#define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
| 399 |
#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
| 400 |
#define ecb_decltype(x) __typeof__ (x) |
| 401 |
#endif |
| 402 |
|
| 403 |
#if _MSC_VER >= 1300 |
| 404 |
#define ecb_deprecated __declspec (deprecated) |
| 405 |
#else |
| 406 |
#define ecb_deprecated ecb_attribute ((__deprecated__)) |
| 407 |
#endif |
| 408 |
|
| 409 |
#if _MSC_VER >= 1500 |
| 410 |
#define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
| 411 |
#elif ECB_GCC_VERSION(4,5) |
| 412 |
#define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
| 413 |
#else |
| 414 |
#define ecb_deprecated_message(msg) ecb_deprecated |
| 415 |
#endif |
| 416 |
|
| 417 |
#if _MSC_VER >= 1400 |
| 418 |
#define ecb_noinline __declspec (noinline) |
| 419 |
#else |
| 420 |
#define ecb_noinline ecb_attribute ((__noinline__)) |
| 421 |
#endif |
| 422 |
|
| 423 |
#define ecb_unused ecb_attribute ((__unused__)) |
| 424 |
#define ecb_const ecb_attribute ((__const__)) |
| 425 |
#define ecb_pure ecb_attribute ((__pure__)) |
| 426 |
|
| 427 |
#if ECB_C11 || __IBMC_NORETURN |
| 428 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
| 429 |
#define ecb_noreturn _Noreturn |
| 430 |
#elif ECB_CPP11 |
| 431 |
#define ecb_noreturn [[noreturn]] |
| 432 |
#elif _MSC_VER >= 1200 |
| 433 |
/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
| 434 |
#define ecb_noreturn __declspec (noreturn) |
| 435 |
#else |
| 436 |
#define ecb_noreturn ecb_attribute ((__noreturn__)) |
| 437 |
#endif |
| 438 |
|
| 439 |
#if ECB_GCC_VERSION(4,3) |
| 440 |
#define ecb_artificial ecb_attribute ((__artificial__)) |
| 441 |
#define ecb_hot ecb_attribute ((__hot__)) |
| 442 |
#define ecb_cold ecb_attribute ((__cold__)) |
| 443 |
#else |
| 444 |
#define ecb_artificial |
| 445 |
#define ecb_hot |
| 446 |
#define ecb_cold |
| 447 |
#endif |
| 448 |
|
| 449 |
/* put around conditional expressions if you are very sure that the */ |
| 450 |
/* expression is mostly true or mostly false. note that these return */ |
| 451 |
/* booleans, not the expression. */ |
| 452 |
#define ecb_expect_false(expr) ecb_expect (!!(expr), 0) |
| 453 |
#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) |
| 454 |
/* for compatibility to the rest of the world */ |
| 455 |
#define ecb_likely(expr) ecb_expect_true (expr) |
| 456 |
#define ecb_unlikely(expr) ecb_expect_false (expr) |
| 457 |
|
| 458 |
/* count trailing zero bits and count # of one bits */ |
| 459 |
#if ECB_GCC_VERSION(3,4) \ |
| 460 |
|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
| 461 |
&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
| 462 |
&& ECB_CLANG_BUILTIN(__builtin_popcount)) |
| 463 |
#define ecb_ctz32(x) __builtin_ctz (x) |
| 464 |
#define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x)) |
| 465 |
#define ecb_clz32(x) __builtin_clz (x) |
| 466 |
#define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x)) |
| 467 |
#define ecb_ld32(x) (ecb_clz32 (x) ^ 31) |
| 468 |
#define ecb_ld64(x) (ecb_clz64 (x) ^ 63) |
| 469 |
#define ecb_popcount32(x) __builtin_popcount (x) |
| 470 |
/* ecb_popcount64 is more difficult, see below */ |
| 471 |
#else |
| 472 |
ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
| 473 |
ecb_function_ ecb_const int ecb_ctz32 (uint32_t x) |
| 474 |
{ |
| 475 |
#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
| 476 |
unsigned long r; |
| 477 |
_BitScanForward (&r, x); |
| 478 |
return (int)r; |
| 479 |
#else |
| 480 |
int r; |
| 481 |
|
| 482 |
x &= ~x + 1; /* this isolates the lowest bit */ |
| 483 |
|
| 484 |
#if 1 |
| 485 |
/* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */ |
| 486 |
/* This happens to return 32 for x == 0, but the API does not support this */ |
| 487 |
|
| 488 |
/* -0 marks unused entries */ |
| 489 |
static unsigned char table[64] = |
| 490 |
{ |
| 491 |
32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14, |
| 492 |
10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15, |
| 493 |
31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26, |
| 494 |
30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0 |
| 495 |
}; |
| 496 |
|
| 497 |
/* magic constant results in 33 unique values in the upper 6 bits */ |
| 498 |
x *= 0x0450fbafU; /* == 17 * 65 * 65535 */ |
| 499 |
|
| 500 |
r = table [x >> 26]; |
| 501 |
#elif 0 /* branchless on i386, typically */ |
| 502 |
r = 0; |
| 503 |
r += !!(x & 0xaaaaaaaa) << 0; |
| 504 |
r += !!(x & 0xcccccccc) << 1; |
| 505 |
r += !!(x & 0xf0f0f0f0) << 2; |
| 506 |
r += !!(x & 0xff00ff00) << 3; |
| 507 |
r += !!(x & 0xffff0000) << 4; |
| 508 |
#else /* branchless on modern compilers, typically */ |
| 509 |
r = 0; |
| 510 |
if (x & 0xaaaaaaaa) r += 1; |
| 511 |
if (x & 0xcccccccc) r += 2; |
| 512 |
if (x & 0xf0f0f0f0) r += 4; |
| 513 |
if (x & 0xff00ff00) r += 8; |
| 514 |
if (x & 0xffff0000) r += 16; |
| 515 |
#endif |
| 516 |
|
| 517 |
return r; |
| 518 |
#endif |
| 519 |
} |
| 520 |
|
| 521 |
ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
| 522 |
ecb_function_ ecb_const int ecb_ctz64 (uint64_t x) |
| 523 |
{ |
| 524 |
#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
| 525 |
unsigned long r; |
| 526 |
_BitScanForward64 (&r, x); |
| 527 |
return (int)r; |
| 528 |
#else |
| 529 |
int shift = x & 0xffffffff ? 0 : 32; |
| 530 |
return ecb_ctz32 (x >> shift) + shift; |
| 531 |
#endif |
| 532 |
} |
| 533 |
|
| 534 |
ecb_function_ ecb_const int ecb_clz32 (uint32_t x); |
| 535 |
ecb_function_ ecb_const int ecb_clz32 (uint32_t x) |
| 536 |
{ |
| 537 |
#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
| 538 |
unsigned long r; |
| 539 |
_BitScanReverse (&r, x); |
| 540 |
return 31 - (int)r; |
| 541 |
#else |
| 542 |
|
| 543 |
/* Robert Harley's algorithm from comp.arch 1996-12-07 */ |
| 544 |
/* This happens to return 32 for x == 0, but the API does not support this */ |
| 545 |
|
| 546 |
/* -0 marks unused table elements */ |
| 547 |
static unsigned char table[64] = |
| 548 |
{ |
| 549 |
32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0, |
| 550 |
-0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0, |
| 551 |
17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18, |
| 552 |
5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0 |
| 553 |
}; |
| 554 |
|
| 555 |
/* propagate leftmost 1 bit to the right */ |
| 556 |
x |= x >> 1; |
| 557 |
x |= x >> 2; |
| 558 |
x |= x >> 4; |
| 559 |
x |= x >> 8; |
| 560 |
x |= x >> 16; |
| 561 |
|
| 562 |
/* magic constant results in 33 unique values in the upper 6 bits */ |
| 563 |
x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */ |
| 564 |
|
| 565 |
return table [x >> 26]; |
| 566 |
#endif |
| 567 |
} |
| 568 |
|
| 569 |
ecb_function_ ecb_const int ecb_clz64 (uint64_t x); |
| 570 |
ecb_function_ ecb_const int ecb_clz64 (uint64_t x) |
| 571 |
{ |
| 572 |
#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
| 573 |
unsigned long r; |
| 574 |
_BitScanReverse64 (&r, x); |
| 575 |
return 63 - (int)r; |
| 576 |
#else |
| 577 |
uint32_t l = x >> 32; |
| 578 |
int shift = l ? 0 : 32; |
| 579 |
return ecb_clz32 (l ? l : x) + shift; |
| 580 |
#endif |
| 581 |
} |
| 582 |
|
| 583 |
ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
| 584 |
ecb_function_ ecb_const int ecb_popcount32 (uint32_t x) |
| 585 |
{ |
| 586 |
x -= (x >> 1) & 0x55555555; |
| 587 |
x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
| 588 |
x = ((x >> 4) + x) & 0x0f0f0f0f; |
| 589 |
x *= 0x01010101; |
| 590 |
|
| 591 |
return x >> 24; |
| 592 |
} |
| 593 |
|
| 594 |
ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
| 595 |
ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
| 596 |
{ |
| 597 |
#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
| 598 |
unsigned long r; |
| 599 |
_BitScanReverse (&r, x); |
| 600 |
return (int)r; |
| 601 |
#else |
| 602 |
int r = 0; |
| 603 |
|
| 604 |
if (x >> 16) { x >>= 16; r += 16; } |
| 605 |
if (x >> 8) { x >>= 8; r += 8; } |
| 606 |
if (x >> 4) { x >>= 4; r += 4; } |
| 607 |
if (x >> 2) { x >>= 2; r += 2; } |
| 608 |
if (x >> 1) { r += 1; } |
| 609 |
|
| 610 |
return r; |
| 611 |
#endif |
| 612 |
} |
| 613 |
|
| 614 |
ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
| 615 |
ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
| 616 |
{ |
| 617 |
#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
| 618 |
unsigned long r; |
| 619 |
_BitScanReverse64 (&r, x); |
| 620 |
return (int)r; |
| 621 |
#else |
| 622 |
int r = 0; |
| 623 |
|
| 624 |
if (x >> 32) { x >>= 32; r += 32; } |
| 625 |
|
| 626 |
return r + ecb_ld32 (x); |
| 627 |
#endif |
| 628 |
} |
| 629 |
#endif |
| 630 |
|
| 631 |
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
| 632 |
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
| 633 |
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
| 634 |
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
| 635 |
|
| 636 |
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
| 637 |
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
| 638 |
{ |
| 639 |
return ( (x * 0x0802U & 0x22110U) |
| 640 |
| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
| 641 |
} |
| 642 |
|
| 643 |
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
| 644 |
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
| 645 |
{ |
| 646 |
x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
| 647 |
x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
| 648 |
x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
| 649 |
x = ( x >> 8 ) | ( x << 8); |
| 650 |
|
| 651 |
return x; |
| 652 |
} |
| 653 |
|
| 654 |
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
| 655 |
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
| 656 |
{ |
| 657 |
x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
| 658 |
x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
| 659 |
x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
| 660 |
x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
| 661 |
x = ( x >> 16 ) | ( x << 16); |
| 662 |
|
| 663 |
return x; |
| 664 |
} |
| 665 |
|
| 666 |
ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
| 667 |
ecb_function_ ecb_const int ecb_popcount64 (uint64_t x) |
| 668 |
{ |
| 669 |
/* popcount64 is only available on 64 bit cpus as gcc builtin. */ |
| 670 |
/* also, gcc/clang make this surprisingly difficult to use */ |
| 671 |
#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl)) |
| 672 |
return __builtin_popcountl (x); |
| 673 |
#else |
| 674 |
return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
| 675 |
#endif |
| 676 |
} |
| 677 |
|
| 678 |
ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); } |
| 679 |
ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); } |
| 680 |
ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); } |
| 681 |
ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); } |
| 682 |
ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); } |
| 683 |
ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); } |
| 684 |
ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); } |
| 685 |
ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); } |
| 686 |
|
| 687 |
#if ECB_CPP |
| 688 |
|
| 689 |
inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } |
| 690 |
inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } |
| 691 |
inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); } |
| 692 |
inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); } |
| 693 |
|
| 694 |
inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); } |
| 695 |
inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); } |
| 696 |
inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); } |
| 697 |
inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); } |
| 698 |
|
| 699 |
inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); } |
| 700 |
inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); } |
| 701 |
inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); } |
| 702 |
inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); } |
| 703 |
|
| 704 |
inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); } |
| 705 |
inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); } |
| 706 |
inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); } |
| 707 |
inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); } |
| 708 |
|
| 709 |
inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); } |
| 710 |
inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); } |
| 711 |
inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); } |
| 712 |
|
| 713 |
inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); } |
| 714 |
inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); } |
| 715 |
inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); } |
| 716 |
inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); } |
| 717 |
|
| 718 |
inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); } |
| 719 |
inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); } |
| 720 |
inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); } |
| 721 |
inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); } |
| 722 |
|
| 723 |
#endif |
| 724 |
|
| 725 |
#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
| 726 |
#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
| 727 |
#define ecb_bswap16(x) __builtin_bswap16 (x) |
| 728 |
#else |
| 729 |
#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
| 730 |
#endif |
| 731 |
#define ecb_bswap32(x) __builtin_bswap32 (x) |
| 732 |
#define ecb_bswap64(x) __builtin_bswap64 (x) |
| 733 |
#elif _MSC_VER |
| 734 |
#include <stdlib.h> |
| 735 |
#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
| 736 |
#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
| 737 |
#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
| 738 |
#else |
| 739 |
ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
| 740 |
ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x) |
| 741 |
{ |
| 742 |
return ecb_rotl16 (x, 8); |
| 743 |
} |
| 744 |
|
| 745 |
ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
| 746 |
ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x) |
| 747 |
{ |
| 748 |
return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
| 749 |
} |
| 750 |
|
| 751 |
ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
| 752 |
ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x) |
| 753 |
{ |
| 754 |
return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 755 |
} |
| 756 |
#endif |
| 757 |
|
| 758 |
#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
| 759 |
#define ecb_unreachable() __builtin_unreachable () |
| 760 |
#else |
| 761 |
/* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 762 |
ecb_inline ecb_noreturn void ecb_unreachable (void); |
| 763 |
ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
| 764 |
#endif |
| 765 |
|
| 766 |
/* try to tell the compiler that some condition is definitely true */ |
| 767 |
#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 768 |
|
| 769 |
ecb_inline uint32_t ecb_byteorder_helper (void); |
| 770 |
ecb_inline uint32_t ecb_byteorder_helper (void) |
| 771 |
{ |
| 772 |
/* the union code still generates code under pressure in gcc, */ |
| 773 |
/* but less than using pointers, and always seems to */ |
| 774 |
/* successfully return a constant. */ |
| 775 |
/* the reason why we have this horrible preprocessor mess */ |
| 776 |
/* is to avoid it in all cases, at least on common architectures */ |
| 777 |
/* or when using a recent enough gcc version (>= 4.6) */ |
| 778 |
#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
| 779 |
|| ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
| 780 |
#define ECB_LITTLE_ENDIAN 1 |
| 781 |
return 0x44332211; |
| 782 |
#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
| 783 |
|| ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
| 784 |
#define ECB_BIG_ENDIAN 1 |
| 785 |
return 0x11223344; |
| 786 |
#else |
| 787 |
union |
| 788 |
{ |
| 789 |
uint8_t c[4]; |
| 790 |
uint32_t u; |
| 791 |
} u = { 0x11, 0x22, 0x33, 0x44 }; |
| 792 |
return u.u; |
| 793 |
#endif |
| 794 |
} |
| 795 |
|
| 796 |
ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
| 797 |
ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
| 798 |
|
| 799 |
/*****************************************************************************/ |
| 800 |
/* unaligned load/store */ |
| 801 |
|
| 802 |
ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
| 803 |
ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
| 804 |
ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
| 805 |
|
| 806 |
ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
| 807 |
ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
| 808 |
ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
| 809 |
|
| 810 |
ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
| 811 |
ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
| 812 |
ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
| 813 |
|
| 814 |
ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); } |
| 815 |
ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); } |
| 816 |
ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); } |
| 817 |
|
| 818 |
ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); } |
| 819 |
ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); } |
| 820 |
ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); } |
| 821 |
|
| 822 |
ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
| 823 |
ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
| 824 |
ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
| 825 |
|
| 826 |
ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
| 827 |
ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
| 828 |
ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
| 829 |
|
| 830 |
ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); } |
| 831 |
ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); } |
| 832 |
ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } |
| 833 |
|
| 834 |
ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); } |
| 835 |
ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); } |
| 836 |
ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); } |
| 837 |
|
| 838 |
ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); } |
| 839 |
ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); } |
| 840 |
ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); } |
| 841 |
|
| 842 |
#if ECB_CPP |
| 843 |
|
| 844 |
inline uint8_t ecb_bswap (uint8_t v) { return v; } |
| 845 |
inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); } |
| 846 |
inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); } |
| 847 |
inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); } |
| 848 |
|
| 849 |
template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
| 850 |
template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
| 851 |
template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; } |
| 852 |
template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); } |
| 853 |
template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); } |
| 854 |
template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; } |
| 855 |
template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); } |
| 856 |
template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); } |
| 857 |
|
| 858 |
template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
| 859 |
template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
| 860 |
template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; } |
| 861 |
template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); } |
| 862 |
template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); } |
| 863 |
template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } |
| 864 |
template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } |
| 865 |
template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } |
| 866 |
|
| 867 |
#endif |
| 868 |
|
| 869 |
/*****************************************************************************/ |
| 870 |
/* pointer/integer hashing */ |
| 871 |
|
| 872 |
/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */ |
| 873 |
ecb_function_ ecb_const uint32_t ecb_mix32 (uint32_t v); |
| 874 |
ecb_function_ ecb_const uint32_t ecb_mix32 (uint32_t v) |
| 875 |
{ |
| 876 |
v ^= v >> 16; v *= 0x7feb352dU; |
| 877 |
v ^= v >> 15; v *= 0x846ca68bU; |
| 878 |
v ^= v >> 16; |
| 879 |
return v; |
| 880 |
} |
| 881 |
|
| 882 |
ecb_function_ ecb_const uint32_t ecb_unmix32 (uint32_t v); |
| 883 |
ecb_function_ ecb_const uint32_t ecb_unmix32 (uint32_t v) |
| 884 |
{ |
| 885 |
v ^= v >> 16 ; v *= 0x43021123U; |
| 886 |
v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U; |
| 887 |
v ^= v >> 16 ; |
| 888 |
return v; |
| 889 |
} |
| 890 |
|
| 891 |
/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */ |
| 892 |
ecb_function_ ecb_const uint64_t ecb_mix64 (uint64_t v); |
| 893 |
ecb_function_ ecb_const uint64_t ecb_mix64 (uint64_t v) |
| 894 |
{ |
| 895 |
v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U; |
| 896 |
v ^= v >> 27; v *= 0x94d049bb133111ebU; |
| 897 |
v ^= v >> 31; |
| 898 |
return v; |
| 899 |
} |
| 900 |
|
| 901 |
ecb_function_ ecb_const uint64_t ecb_unmix64 (uint64_t v); |
| 902 |
ecb_function_ ecb_const uint64_t ecb_unmix64 (uint64_t v) |
| 903 |
{ |
| 904 |
v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U; |
| 905 |
v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U; |
| 906 |
v ^= v >> 30 ^ v >> 60; |
| 907 |
return v; |
| 908 |
} |
| 909 |
|
| 910 |
ecb_function_ ecb_const uintptr_t ecb_ptrmix (void *p); |
| 911 |
ecb_function_ ecb_const uintptr_t ecb_ptrmix (void *p) |
| 912 |
{ |
| 913 |
#if ECB_PTRSIZE <= 4 |
| 914 |
return ecb_mix32 ((uint32_t)p); |
| 915 |
#else |
| 916 |
return ecb_mix64 ((uint64_t)p); |
| 917 |
#endif |
| 918 |
} |
| 919 |
|
| 920 |
ecb_function_ ecb_const void *ecb_ptrunmix (uintptr_t v); |
| 921 |
ecb_function_ ecb_const void *ecb_ptrunmix (uintptr_t v) |
| 922 |
{ |
| 923 |
#if ECB_PTRSIZE <= 4 |
| 924 |
return (void *)ecb_unmix32 (v); |
| 925 |
#else |
| 926 |
return (void *)ecb_unmix64 (v); |
| 927 |
#endif |
| 928 |
} |
| 929 |
|
| 930 |
#if ECB_CPP |
| 931 |
|
| 932 |
template<typename T> |
| 933 |
inline uintptr_t ecb_ptrmix (T *p) |
| 934 |
{ |
| 935 |
return ecb_ptrmix (static_cast<void *>(p)); |
| 936 |
} |
| 937 |
|
| 938 |
template<typename T> |
| 939 |
inline T *ecb_ptrunmix (uintptr_t v) |
| 940 |
{ |
| 941 |
return static_cast<T *>(ecb_ptrunmix (v)); |
| 942 |
} |
| 943 |
|
| 944 |
#endif |
| 945 |
|
| 946 |
/*****************************************************************************/ |
| 947 |
/* gray code */ |
| 948 |
|
| 949 |
ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); } |
| 950 |
ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); } |
| 951 |
ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); } |
| 952 |
ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); } |
| 953 |
|
| 954 |
ecb_function_ ecb_const uint8_t ecb_gray_decode8 (uint8_t g); |
| 955 |
ecb_function_ ecb_const uint8_t ecb_gray_decode8 (uint8_t g) |
| 956 |
{ |
| 957 |
g ^= g >> 1; |
| 958 |
g ^= g >> 2; |
| 959 |
g ^= g >> 4; |
| 960 |
|
| 961 |
return g; |
| 962 |
} |
| 963 |
|
| 964 |
ecb_function_ ecb_const uint16_t ecb_gray_decode16 (uint16_t g); |
| 965 |
ecb_function_ ecb_const uint16_t ecb_gray_decode16 (uint16_t g) |
| 966 |
{ |
| 967 |
g ^= g >> 1; |
| 968 |
g ^= g >> 2; |
| 969 |
g ^= g >> 4; |
| 970 |
g ^= g >> 8; |
| 971 |
|
| 972 |
return g; |
| 973 |
} |
| 974 |
|
| 975 |
ecb_function_ ecb_const uint32_t ecb_gray_decode32 (uint32_t g); |
| 976 |
ecb_function_ ecb_const uint32_t ecb_gray_decode32 (uint32_t g) |
| 977 |
{ |
| 978 |
g ^= g >> 1; |
| 979 |
g ^= g >> 2; |
| 980 |
g ^= g >> 4; |
| 981 |
g ^= g >> 8; |
| 982 |
g ^= g >> 16; |
| 983 |
|
| 984 |
return g; |
| 985 |
} |
| 986 |
|
| 987 |
ecb_function_ ecb_const uint64_t ecb_gray_decode64 (uint64_t g); |
| 988 |
ecb_function_ ecb_const uint64_t ecb_gray_decode64 (uint64_t g) |
| 989 |
{ |
| 990 |
g ^= g >> 1; |
| 991 |
g ^= g >> 2; |
| 992 |
g ^= g >> 4; |
| 993 |
g ^= g >> 8; |
| 994 |
g ^= g >> 16; |
| 995 |
g ^= g >> 32; |
| 996 |
|
| 997 |
return g; |
| 998 |
} |
| 999 |
|
| 1000 |
#if ECB_CPP |
| 1001 |
|
| 1002 |
ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); } |
| 1003 |
ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); } |
| 1004 |
ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); } |
| 1005 |
ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); } |
| 1006 |
|
| 1007 |
ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); } |
| 1008 |
ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); } |
| 1009 |
ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); } |
| 1010 |
ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); } |
| 1011 |
|
| 1012 |
#endif |
| 1013 |
|
| 1014 |
/*****************************************************************************/ |
| 1015 |
/* 2d hilbert curves */ |
| 1016 |
|
| 1017 |
/* algorithm from the book Hacker's Delight, modified to not */ |
| 1018 |
/* run into undefined behaviour for n==16 */ |
| 1019 |
ecb_function_ ecb_const uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s); |
| 1020 |
ecb_function_ ecb_const uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s) |
| 1021 |
{ |
| 1022 |
uint32_t comp, swap, cs, t, sr; |
| 1023 |
|
| 1024 |
/* pad s on the left (unused) bits with 01 (no change groups) */ |
| 1025 |
s |= 0x55555555U << n << n; |
| 1026 |
/* "s shift right" */ |
| 1027 |
sr = (s >> 1) & 0x55555555U; |
| 1028 |
/* compute complement and swap info in two-bit groups */ |
| 1029 |
cs = ((s & 0x55555555U) + sr) ^ 0x55555555U; |
| 1030 |
|
| 1031 |
/* parallel prefix xor op to propagate both complement |
| 1032 |
* and swap info together from left to right (there is |
| 1033 |
* no step "cs ^= cs >> 1", so in effect it computes |
| 1034 |
* two independent parallel prefix operations on two |
| 1035 |
* interleaved sets of sixteen bits). |
| 1036 |
*/ |
| 1037 |
cs ^= cs >> 2; |
| 1038 |
cs ^= cs >> 4; |
| 1039 |
cs ^= cs >> 8; |
| 1040 |
cs ^= cs >> 16; |
| 1041 |
|
| 1042 |
/* separate swap and complement bits */ |
| 1043 |
swap = cs & 0x55555555U; |
| 1044 |
comp = (cs >> 1) & 0x55555555U; |
| 1045 |
|
| 1046 |
/* calculate coordinates in odd and even bit positions */ |
| 1047 |
t = (s & swap) ^ comp; |
| 1048 |
s = s ^ sr ^ t ^ (t << 1); |
| 1049 |
|
| 1050 |
/* unpad/clear out any junk on the left */ |
| 1051 |
s = s & ((1 << n << n) - 1); |
| 1052 |
|
| 1053 |
/* Now "unshuffle" to separate the x and y bits. */ |
| 1054 |
t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1); |
| 1055 |
t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2); |
| 1056 |
t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4); |
| 1057 |
t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8); |
| 1058 |
|
| 1059 |
/* now s contains two 16-bit coordinates */ |
| 1060 |
return s; |
| 1061 |
} |
| 1062 |
|
| 1063 |
/* 64 bit, a straightforward extension to the 32 bit case */ |
| 1064 |
ecb_function_ ecb_const uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s); |
| 1065 |
ecb_function_ ecb_const uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s) |
| 1066 |
{ |
| 1067 |
uint64_t comp, swap, cs, t, sr; |
| 1068 |
|
| 1069 |
/* pad s on the left (unused) bits with 01 (no change groups) */ |
| 1070 |
s |= 0x5555555555555555U << n << n; |
| 1071 |
/* "s shift right" */ |
| 1072 |
sr = (s >> 1) & 0x5555555555555555U; |
| 1073 |
/* compute complement and swap info in two-bit groups */ |
| 1074 |
cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U; |
| 1075 |
|
| 1076 |
/* parallel prefix xor op to propagate both complement |
| 1077 |
* and swap info together from left to right (there is |
| 1078 |
* no step "cs ^= cs >> 1", so in effect it computes |
| 1079 |
* two independent parallel prefix operations on two |
| 1080 |
* interleaved sets of thirty-two bits). |
| 1081 |
*/ |
| 1082 |
cs ^= cs >> 2; |
| 1083 |
cs ^= cs >> 4; |
| 1084 |
cs ^= cs >> 8; |
| 1085 |
cs ^= cs >> 16; |
| 1086 |
cs ^= cs >> 32; |
| 1087 |
|
| 1088 |
/* separate swap and complement bits */ |
| 1089 |
swap = cs & 0x5555555555555555U; |
| 1090 |
comp = (cs >> 1) & 0x5555555555555555U; |
| 1091 |
|
| 1092 |
/* calculate coordinates in odd and even bit positions */ |
| 1093 |
t = (s & swap) ^ comp; |
| 1094 |
s = s ^ sr ^ t ^ (t << 1); |
| 1095 |
|
| 1096 |
/* unpad/clear out any junk on the left */ |
| 1097 |
s = s & ((1 << n << n) - 1); |
| 1098 |
|
| 1099 |
/* Now "unshuffle" to separate the x and y bits. */ |
| 1100 |
t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1); |
| 1101 |
t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2); |
| 1102 |
t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4); |
| 1103 |
t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8); |
| 1104 |
t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16); |
| 1105 |
|
| 1106 |
/* now s contains two 32-bit coordinates */ |
| 1107 |
return s; |
| 1108 |
} |
| 1109 |
|
| 1110 |
/* algorithm from the book Hacker's Delight, but a similar algorithm*/ |
| 1111 |
/* is given in https://doi.org/10.1002/spe.4380160103 */ |
| 1112 |
/* this has been slightly improved over the original version */ |
| 1113 |
ecb_function_ ecb_const uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy); |
| 1114 |
ecb_function_ ecb_const uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy) |
| 1115 |
{ |
| 1116 |
uint32_t row; |
| 1117 |
uint32_t state = 0; |
| 1118 |
uint32_t s = 0; |
| 1119 |
|
| 1120 |
do |
| 1121 |
{ |
| 1122 |
--n; |
| 1123 |
|
| 1124 |
row = 4 * state |
| 1125 |
| (2 & (xy >> n >> 15)) |
| 1126 |
| (1 & (xy >> n )); |
| 1127 |
|
| 1128 |
/* these funky constants are lookup tables for two-bit values */ |
| 1129 |
s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; |
| 1130 |
state = (0x8fe65831U >> 2 * row) & 3; |
| 1131 |
} |
| 1132 |
while (n > 0); |
| 1133 |
|
| 1134 |
return s; |
| 1135 |
} |
| 1136 |
|
| 1137 |
/* 64 bit, essentially the same as 32 bit */ |
| 1138 |
ecb_function_ ecb_const uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy); |
| 1139 |
ecb_function_ ecb_const uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy) |
| 1140 |
{ |
| 1141 |
uint32_t row; |
| 1142 |
uint32_t state = 0; |
| 1143 |
uint64_t s = 0; |
| 1144 |
|
| 1145 |
do |
| 1146 |
{ |
| 1147 |
--n; |
| 1148 |
|
| 1149 |
row = 4 * state |
| 1150 |
| (2 & (xy >> n >> 31)) |
| 1151 |
| (1 & (xy >> n )); |
| 1152 |
|
| 1153 |
/* these funky constants are lookup tables for two-bit values */ |
| 1154 |
s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; |
| 1155 |
state = (0x8fe65831U >> 2 * row) & 3; |
| 1156 |
} |
| 1157 |
while (n > 0); |
| 1158 |
|
| 1159 |
return s; |
| 1160 |
} |
| 1161 |
|
| 1162 |
/*****************************************************************************/ |
| 1163 |
/* division */ |
| 1164 |
|
| 1165 |
#if ECB_GCC_VERSION(3,0) || ECB_C99 |
| 1166 |
/* C99 tightened the definition of %, so we can use a more efficient version */ |
| 1167 |
#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 1168 |
#else |
| 1169 |
#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
| 1170 |
#endif |
| 1171 |
|
| 1172 |
#if ECB_CPP |
| 1173 |
template<typename T> |
| 1174 |
static inline T ecb_div_rd (T val, T div) |
| 1175 |
{ |
| 1176 |
return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
| 1177 |
} |
| 1178 |
template<typename T> |
| 1179 |
static inline T ecb_div_ru (T val, T div) |
| 1180 |
{ |
| 1181 |
return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
| 1182 |
} |
| 1183 |
#else |
| 1184 |
#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
| 1185 |
#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 1186 |
#endif |
| 1187 |
|
| 1188 |
/*****************************************************************************/ |
| 1189 |
/* array length */ |
| 1190 |
|
| 1191 |
#if ecb_cplusplus_does_not_suck |
| 1192 |
/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
| 1193 |
template<typename T, int N> |
| 1194 |
static inline int ecb_array_length (const T (&arr)[N]) |
| 1195 |
{ |
| 1196 |
return N; |
| 1197 |
} |
| 1198 |
#else |
| 1199 |
#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 1200 |
#endif |
| 1201 |
|
| 1202 |
/*****************************************************************************/ |
| 1203 |
/* IEEE 754-2008 half float conversions */ |
| 1204 |
|
| 1205 |
ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
| 1206 |
ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x) |
| 1207 |
{ |
| 1208 |
unsigned int s = (x & 0x8000) << (31 - 15); |
| 1209 |
int e = (x >> 10) & 0x001f; |
| 1210 |
unsigned int m = x & 0x03ff; |
| 1211 |
|
| 1212 |
if (ecb_expect_false (e == 31)) |
| 1213 |
/* infinity or NaN */ |
| 1214 |
e = 255 - (127 - 15); |
| 1215 |
else if (ecb_expect_false (!e)) |
| 1216 |
{ |
| 1217 |
if (ecb_expect_true (!m)) |
| 1218 |
/* zero, handled by code below by forcing e to 0 */ |
| 1219 |
e = 0 - (127 - 15); |
| 1220 |
else |
| 1221 |
{ |
| 1222 |
/* subnormal, renormalise */ |
| 1223 |
unsigned int s = 10 - ecb_ld32 (m); |
| 1224 |
|
| 1225 |
m = (m << s) & 0x3ff; /* mask implicit bit */ |
| 1226 |
e -= s - 1; |
| 1227 |
} |
| 1228 |
} |
| 1229 |
|
| 1230 |
/* e and m now are normalised, or zero, (or inf or nan) */ |
| 1231 |
e += 127 - 15; |
| 1232 |
|
| 1233 |
return s | (e << 23) | (m << (23 - 10)); |
| 1234 |
} |
| 1235 |
|
| 1236 |
ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
| 1237 |
ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x) |
| 1238 |
{ |
| 1239 |
unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
| 1240 |
int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
| 1241 |
unsigned int m = x & 0x007fffff; |
| 1242 |
|
| 1243 |
x &= 0x7fffffff; |
| 1244 |
|
| 1245 |
/* if it's within range of binary16 normals, use fast path */ |
| 1246 |
if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
| 1247 |
{ |
| 1248 |
/* mantissa round-to-even */ |
| 1249 |
m += 0x00000fff + ((m >> (23 - 10)) & 1); |
| 1250 |
|
| 1251 |
/* handle overflow */ |
| 1252 |
if (ecb_expect_false (m >= 0x00800000)) |
| 1253 |
{ |
| 1254 |
m >>= 1; |
| 1255 |
e += 1; |
| 1256 |
} |
| 1257 |
|
| 1258 |
return s | (e << 10) | (m >> (23 - 10)); |
| 1259 |
} |
| 1260 |
|
| 1261 |
/* handle large numbers and infinity */ |
| 1262 |
if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
| 1263 |
return s | 0x7c00; |
| 1264 |
|
| 1265 |
/* handle zero, subnormals and small numbers */ |
| 1266 |
if (ecb_expect_true (x < 0x38800000)) |
| 1267 |
{ |
| 1268 |
/* zero */ |
| 1269 |
if (ecb_expect_true (!x)) |
| 1270 |
return s; |
| 1271 |
|
| 1272 |
/* handle subnormals */ |
| 1273 |
|
| 1274 |
/* too small, will be zero */ |
| 1275 |
if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
| 1276 |
return s; |
| 1277 |
|
| 1278 |
m |= 0x00800000; /* make implicit bit explicit */ |
| 1279 |
|
| 1280 |
/* very tricky - we need to round to the nearest e (+10) bit value */ |
| 1281 |
{ |
| 1282 |
unsigned int bits = 14 - e; |
| 1283 |
unsigned int half = (1 << (bits - 1)) - 1; |
| 1284 |
unsigned int even = (m >> bits) & 1; |
| 1285 |
|
| 1286 |
/* if this overflows, we will end up with a normalised number */ |
| 1287 |
m = (m + half + even) >> bits; |
| 1288 |
} |
| 1289 |
|
| 1290 |
return s | m; |
| 1291 |
} |
| 1292 |
|
| 1293 |
/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
| 1294 |
m >>= 13; |
| 1295 |
|
| 1296 |
return s | 0x7c00 | m | !m; |
| 1297 |
} |
| 1298 |
|
| 1299 |
/*******************************************************************************/ |
| 1300 |
/* fast integer to ascii */ |
| 1301 |
|
| 1302 |
/* |
| 1303 |
* This code is pretty complicated because it is general. The idea behind it, |
| 1304 |
* however, is pretty simple: first, the number is multiplied with a scaling |
| 1305 |
* factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point |
| 1306 |
* number with the first digit in the upper bits. |
| 1307 |
* Then this digit is converted to text and masked out. The resulting number |
| 1308 |
* is then multiplied by 10, by multiplying the fixed point representation |
| 1309 |
* by 5 and shifting the (binary) decimal point one to the right, so a 4.28 |
| 1310 |
* format becomes 5.27, 6.26 and so on. |
| 1311 |
* The rest involves only advancing the pointer if we already generated a |
| 1312 |
* non-zero digit, so leading zeroes are overwritten. |
| 1313 |
*/ |
| 1314 |
|
| 1315 |
/* simply return a mask with "bits" bits set */ |
| 1316 |
#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) |
| 1317 |
|
| 1318 |
/* oputput a single digit. maskvalue is 10**digitidx */ |
| 1319 |
#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ |
| 1320 |
if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ |
| 1321 |
{ \ |
| 1322 |
char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ |
| 1323 |
*ptr = digit + '0'; /* output it */ \ |
| 1324 |
nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ |
| 1325 |
ptr += nz; /* output digit only if non-zero digit seen */ \ |
| 1326 |
x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \ |
| 1327 |
} |
| 1328 |
|
| 1329 |
/* convert integer to fixed point format and multiply out digits, highest first */ |
| 1330 |
/* requires magic constants: max. digits and number of bits after the decimal point */ |
| 1331 |
#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \ |
| 1332 |
ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ |
| 1333 |
{ \ |
| 1334 |
char nz = lz; /* non-zero digit seen? */ \ |
| 1335 |
/* convert to x.bits fixed-point */ \ |
| 1336 |
type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \ |
| 1337 |
/* output up to 10 digits */ \ |
| 1338 |
ecb_i2a_digit (type,bits,digitmask, 1, 0); \ |
| 1339 |
ecb_i2a_digit (type,bits,digitmask, 10, 1); \ |
| 1340 |
ecb_i2a_digit (type,bits,digitmask, 100, 2); \ |
| 1341 |
ecb_i2a_digit (type,bits,digitmask, 1000, 3); \ |
| 1342 |
ecb_i2a_digit (type,bits,digitmask, 10000, 4); \ |
| 1343 |
ecb_i2a_digit (type,bits,digitmask, 100000, 5); \ |
| 1344 |
ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \ |
| 1345 |
ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \ |
| 1346 |
ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ |
| 1347 |
ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ |
| 1348 |
return ptr; \ |
| 1349 |
} |
| 1350 |
|
| 1351 |
/* predefined versions of the above, for various digits */ |
| 1352 |
/* ecb_i2a_xN = almost N digits, limit defined by macro */ |
| 1353 |
/* ecb_i2a_N = up to N digits, leading zeroes suppressed */ |
| 1354 |
/* ecb_i2a_0N = exactly N digits, including leading zeroes */ |
| 1355 |
|
| 1356 |
/* non-leading-zero versions, limited range */ |
| 1357 |
#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */ |
| 1358 |
#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */ |
| 1359 |
ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) |
| 1360 |
ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) |
| 1361 |
|
| 1362 |
/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */ |
| 1363 |
ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) |
| 1364 |
ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) |
| 1365 |
ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) |
| 1366 |
ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) |
| 1367 |
ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) |
| 1368 |
ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) |
| 1369 |
ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) |
| 1370 |
ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) |
| 1371 |
|
| 1372 |
/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */ |
| 1373 |
ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) |
| 1374 |
ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) |
| 1375 |
ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) |
| 1376 |
ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) |
| 1377 |
ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) |
| 1378 |
ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1) |
| 1379 |
ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1) |
| 1380 |
ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1) |
| 1381 |
|
| 1382 |
#define ECB_I2A_I32_DIGITS 11 |
| 1383 |
#define ECB_I2A_U32_DIGITS 10 |
| 1384 |
#define ECB_I2A_I64_DIGITS 20 |
| 1385 |
#define ECB_I2A_U64_DIGITS 21 |
| 1386 |
#define ECB_I2A_MAX_DIGITS 21 |
| 1387 |
|
| 1388 |
ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u); |
| 1389 |
ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u) |
| 1390 |
{ |
| 1391 |
#if ECB_64BIT_NATIVE |
| 1392 |
if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
| 1393 |
ptr = ecb_i2a_x10 (ptr, u); |
| 1394 |
else /* x10 almost, but not fully, covers 32 bit */ |
| 1395 |
{ |
| 1396 |
uint32_t u1 = u % 1000000000; |
| 1397 |
uint32_t u2 = u / 1000000000; |
| 1398 |
|
| 1399 |
*ptr++ = u2 + '0'; |
| 1400 |
ptr = ecb_i2a_09 (ptr, u1); |
| 1401 |
} |
| 1402 |
#else |
| 1403 |
if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) |
| 1404 |
ecb_i2a_x5 (ptr, u); |
| 1405 |
else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000)) |
| 1406 |
{ |
| 1407 |
uint32_t u1 = u % 10000; |
| 1408 |
uint32_t u2 = u / 10000; |
| 1409 |
|
| 1410 |
ptr = ecb_i2a_x5 (ptr, u2); |
| 1411 |
ptr = ecb_i2a_04 (ptr, u1); |
| 1412 |
} |
| 1413 |
else |
| 1414 |
{ |
| 1415 |
uint32_t u1 = u % 10000; |
| 1416 |
uint32_t ua = u / 10000; |
| 1417 |
uint32_t u2 = ua % 10000; |
| 1418 |
uint32_t u3 = ua / 10000; |
| 1419 |
|
| 1420 |
ptr = ecb_i2a_2 (ptr, u3); |
| 1421 |
ptr = ecb_i2a_04 (ptr, u2); |
| 1422 |
ptr = ecb_i2a_04 (ptr, u1); |
| 1423 |
} |
| 1424 |
#endif |
| 1425 |
|
| 1426 |
return ptr; |
| 1427 |
} |
| 1428 |
|
| 1429 |
ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v); |
| 1430 |
ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v) |
| 1431 |
{ |
| 1432 |
*ptr = '-'; ptr += v < 0; |
| 1433 |
uint32_t u = v < 0 ? -(uint32_t)v : v; |
| 1434 |
|
| 1435 |
#if ECB_64BIT_NATIVE |
| 1436 |
ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */ |
| 1437 |
#else |
| 1438 |
ptr = ecb_i2a_u32 (ptr, u); |
| 1439 |
#endif |
| 1440 |
|
| 1441 |
return ptr; |
| 1442 |
} |
| 1443 |
|
| 1444 |
ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u); |
| 1445 |
ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u) |
| 1446 |
{ |
| 1447 |
#if ECB_64BIT_NATIVE |
| 1448 |
if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
| 1449 |
ptr = ecb_i2a_x10 (ptr, u); |
| 1450 |
else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) |
| 1451 |
{ |
| 1452 |
uint64_t u1 = u % 1000000000; |
| 1453 |
uint64_t u2 = u / 1000000000; |
| 1454 |
|
| 1455 |
ptr = ecb_i2a_x10 (ptr, u2); |
| 1456 |
ptr = ecb_i2a_09 (ptr, u1); |
| 1457 |
} |
| 1458 |
else |
| 1459 |
{ |
| 1460 |
uint64_t u1 = u % 1000000000; |
| 1461 |
uint64_t ua = u / 1000000000; |
| 1462 |
uint64_t u2 = ua % 1000000000; |
| 1463 |
uint64_t u3 = ua / 1000000000; |
| 1464 |
|
| 1465 |
ptr = ecb_i2a_2 (ptr, u3); |
| 1466 |
ptr = ecb_i2a_09 (ptr, u2); |
| 1467 |
ptr = ecb_i2a_09 (ptr, u1); |
| 1468 |
} |
| 1469 |
#else |
| 1470 |
if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) |
| 1471 |
ptr = ecb_i2a_x5 (ptr, u); |
| 1472 |
else |
| 1473 |
{ |
| 1474 |
uint64_t u1 = u % 10000; |
| 1475 |
uint64_t u2 = u / 10000; |
| 1476 |
|
| 1477 |
ptr = ecb_i2a_u64 (ptr, u2); |
| 1478 |
ptr = ecb_i2a_04 (ptr, u1); |
| 1479 |
} |
| 1480 |
#endif |
| 1481 |
|
| 1482 |
return ptr; |
| 1483 |
} |
| 1484 |
|
| 1485 |
ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v); |
| 1486 |
ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v) |
| 1487 |
{ |
| 1488 |
*ptr = '-'; ptr += v < 0; |
| 1489 |
uint64_t u = v < 0 ? -(uint64_t)v : v; |
| 1490 |
|
| 1491 |
#if ECB_64BIT_NATIVE |
| 1492 |
if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
| 1493 |
ptr = ecb_i2a_x10 (ptr, u); |
| 1494 |
else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) |
| 1495 |
{ |
| 1496 |
uint64_t u1 = u % 1000000000; |
| 1497 |
uint64_t u2 = u / 1000000000; |
| 1498 |
|
| 1499 |
ptr = ecb_i2a_x10 (ptr, u2); |
| 1500 |
ptr = ecb_i2a_09 (ptr, u1); |
| 1501 |
} |
| 1502 |
else |
| 1503 |
{ |
| 1504 |
uint64_t u1 = u % 1000000000; |
| 1505 |
uint64_t ua = u / 1000000000; |
| 1506 |
uint64_t u2 = ua % 1000000000; |
| 1507 |
uint64_t u3 = ua / 1000000000; |
| 1508 |
|
| 1509 |
/* 2**31 is 19 digits, so the top is exactly one digit */ |
| 1510 |
*ptr++ = u3 + '0'; |
| 1511 |
ptr = ecb_i2a_09 (ptr, u2); |
| 1512 |
ptr = ecb_i2a_09 (ptr, u1); |
| 1513 |
} |
| 1514 |
#else |
| 1515 |
ptr = ecb_i2a_u64 (ptr, u); |
| 1516 |
#endif |
| 1517 |
|
| 1518 |
return ptr; |
| 1519 |
} |
| 1520 |
|
| 1521 |
/*******************************************************************************/ |
| 1522 |
/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
| 1523 |
|
| 1524 |
/* basically, everything uses "ieee pure-endian" floating point numbers */ |
| 1525 |
/* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
| 1526 |
#if 0 \ |
| 1527 |
|| __i386 || __i386__ \ |
| 1528 |
|| ECB_GCC_AMD64 \ |
| 1529 |
|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
| 1530 |
|| defined __s390__ || defined __s390x__ \ |
| 1531 |
|| defined __mips__ \ |
| 1532 |
|| defined __alpha__ \ |
| 1533 |
|| defined __hppa__ \ |
| 1534 |
|| defined __ia64__ \ |
| 1535 |
|| defined __m68k__ \ |
| 1536 |
|| defined __m88k__ \ |
| 1537 |
|| defined __sh__ \ |
| 1538 |
|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
| 1539 |
|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
| 1540 |
|| defined __aarch64__ |
| 1541 |
#define ECB_STDFP 1 |
| 1542 |
#else |
| 1543 |
#define ECB_STDFP 0 |
| 1544 |
#endif |
| 1545 |
|
| 1546 |
#ifndef ECB_NO_LIBM |
| 1547 |
|
| 1548 |
#include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
| 1549 |
|
| 1550 |
/* only the oldest of old doesn't have this one. solaris. */ |
| 1551 |
#ifdef INFINITY |
| 1552 |
#define ECB_INFINITY INFINITY |
| 1553 |
#else |
| 1554 |
#define ECB_INFINITY HUGE_VAL |
| 1555 |
#endif |
| 1556 |
|
| 1557 |
#ifdef NAN |
| 1558 |
#define ECB_NAN NAN |
| 1559 |
#else |
| 1560 |
#define ECB_NAN ECB_INFINITY |
| 1561 |
#endif |
| 1562 |
|
| 1563 |
#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
| 1564 |
#define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
| 1565 |
#define ecb_frexpf(x,e) frexpf ((x), (e)) |
| 1566 |
#else |
| 1567 |
#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
| 1568 |
#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
| 1569 |
#endif |
| 1570 |
|
| 1571 |
/* convert a float to ieee single/binary32 */ |
| 1572 |
ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
| 1573 |
ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x) |
| 1574 |
{ |
| 1575 |
uint32_t r; |
| 1576 |
|
| 1577 |
#if ECB_STDFP |
| 1578 |
memcpy (&r, &x, 4); |
| 1579 |
#else |
| 1580 |
/* slow emulation, works for anything but -0 */ |
| 1581 |
uint32_t m; |
| 1582 |
int e; |
| 1583 |
|
| 1584 |
if (x == 0e0f ) return 0x00000000U; |
| 1585 |
if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
| 1586 |
if (x < -3.40282346638528860e+38f) return 0xff800000U; |
| 1587 |
if (x != x ) return 0x7fbfffffU; |
| 1588 |
|
| 1589 |
m = ecb_frexpf (x, &e) * 0x1000000U; |
| 1590 |
|
| 1591 |
r = m & 0x80000000U; |
| 1592 |
|
| 1593 |
if (r) |
| 1594 |
m = -m; |
| 1595 |
|
| 1596 |
if (e <= -126) |
| 1597 |
{ |
| 1598 |
m &= 0xffffffU; |
| 1599 |
m >>= (-125 - e); |
| 1600 |
e = -126; |
| 1601 |
} |
| 1602 |
|
| 1603 |
r |= (e + 126) << 23; |
| 1604 |
r |= m & 0x7fffffU; |
| 1605 |
#endif |
| 1606 |
|
| 1607 |
return r; |
| 1608 |
} |
| 1609 |
|
| 1610 |
/* converts an ieee single/binary32 to a float */ |
| 1611 |
ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
| 1612 |
ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x) |
| 1613 |
{ |
| 1614 |
float r; |
| 1615 |
|
| 1616 |
#if ECB_STDFP |
| 1617 |
memcpy (&r, &x, 4); |
| 1618 |
#else |
| 1619 |
/* emulation, only works for normals and subnormals and +0 */ |
| 1620 |
int neg = x >> 31; |
| 1621 |
int e = (x >> 23) & 0xffU; |
| 1622 |
|
| 1623 |
x &= 0x7fffffU; |
| 1624 |
|
| 1625 |
if (e) |
| 1626 |
x |= 0x800000U; |
| 1627 |
else |
| 1628 |
e = 1; |
| 1629 |
|
| 1630 |
/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
| 1631 |
r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
| 1632 |
|
| 1633 |
r = neg ? -r : r; |
| 1634 |
#endif |
| 1635 |
|
| 1636 |
return r; |
| 1637 |
} |
| 1638 |
|
| 1639 |
/* convert a double to ieee double/binary64 */ |
| 1640 |
ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
| 1641 |
ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x) |
| 1642 |
{ |
| 1643 |
uint64_t r; |
| 1644 |
|
| 1645 |
#if ECB_STDFP |
| 1646 |
memcpy (&r, &x, 8); |
| 1647 |
#else |
| 1648 |
/* slow emulation, works for anything but -0 */ |
| 1649 |
uint64_t m; |
| 1650 |
int e; |
| 1651 |
|
| 1652 |
if (x == 0e0 ) return 0x0000000000000000U; |
| 1653 |
if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
| 1654 |
if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
| 1655 |
if (x != x ) return 0X7ff7ffffffffffffU; |
| 1656 |
|
| 1657 |
m = frexp (x, &e) * 0x20000000000000U; |
| 1658 |
|
| 1659 |
r = m & 0x8000000000000000;; |
| 1660 |
|
| 1661 |
if (r) |
| 1662 |
m = -m; |
| 1663 |
|
| 1664 |
if (e <= -1022) |
| 1665 |
{ |
| 1666 |
m &= 0x1fffffffffffffU; |
| 1667 |
m >>= (-1021 - e); |
| 1668 |
e = -1022; |
| 1669 |
} |
| 1670 |
|
| 1671 |
r |= ((uint64_t)(e + 1022)) << 52; |
| 1672 |
r |= m & 0xfffffffffffffU; |
| 1673 |
#endif |
| 1674 |
|
| 1675 |
return r; |
| 1676 |
} |
| 1677 |
|
| 1678 |
/* converts an ieee double/binary64 to a double */ |
| 1679 |
ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
| 1680 |
ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x) |
| 1681 |
{ |
| 1682 |
double r; |
| 1683 |
|
| 1684 |
#if ECB_STDFP |
| 1685 |
memcpy (&r, &x, 8); |
| 1686 |
#else |
| 1687 |
/* emulation, only works for normals and subnormals and +0 */ |
| 1688 |
int neg = x >> 63; |
| 1689 |
int e = (x >> 52) & 0x7ffU; |
| 1690 |
|
| 1691 |
x &= 0xfffffffffffffU; |
| 1692 |
|
| 1693 |
if (e) |
| 1694 |
x |= 0x10000000000000U; |
| 1695 |
else |
| 1696 |
e = 1; |
| 1697 |
|
| 1698 |
/* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
| 1699 |
r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
| 1700 |
|
| 1701 |
r = neg ? -r : r; |
| 1702 |
#endif |
| 1703 |
|
| 1704 |
return r; |
| 1705 |
} |
| 1706 |
|
| 1707 |
/* convert a float to ieee half/binary16 */ |
| 1708 |
ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
| 1709 |
ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x) |
| 1710 |
{ |
| 1711 |
return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
| 1712 |
} |
| 1713 |
|
| 1714 |
/* convert an ieee half/binary16 to float */ |
| 1715 |
ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
| 1716 |
ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x) |
| 1717 |
{ |
| 1718 |
return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
| 1719 |
} |
| 1720 |
|
| 1721 |
#endif |
| 1722 |
|
| 1723 |
#endif |
| 1724 |
|
