| 1 |
/* |
| 2 |
* libecb - http://software.schmorp.de/pkg/libecb |
| 3 |
* |
| 4 |
* Copyright (©) 2009-2013 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 |
|
| 30 |
#ifndef ECB_H |
| 31 |
#define ECB_H |
| 32 |
|
| 33 |
/* 16 bits major, 16 bits minor */ |
| 34 |
#define ECB_VERSION 0x00010003 |
| 35 |
|
| 36 |
#ifdef _WIN32 |
| 37 |
typedef signed char int8_t; |
| 38 |
typedef unsigned char uint8_t; |
| 39 |
typedef signed short int16_t; |
| 40 |
typedef unsigned short uint16_t; |
| 41 |
typedef signed int int32_t; |
| 42 |
typedef unsigned int uint32_t; |
| 43 |
#if __GNUC__ |
| 44 |
typedef signed long long int64_t; |
| 45 |
typedef unsigned long long uint64_t; |
| 46 |
#else /* _MSC_VER || __BORLANDC__ */ |
| 47 |
typedef signed __int64 int64_t; |
| 48 |
typedef unsigned __int64 uint64_t; |
| 49 |
#endif |
| 50 |
#ifdef _WIN64 |
| 51 |
#define ECB_PTRSIZE 8 |
| 52 |
typedef uint64_t uintptr_t; |
| 53 |
typedef int64_t intptr_t; |
| 54 |
#else |
| 55 |
#define ECB_PTRSIZE 4 |
| 56 |
typedef uint32_t uintptr_t; |
| 57 |
typedef int32_t intptr_t; |
| 58 |
#endif |
| 59 |
#else |
| 60 |
#include <inttypes.h> |
| 61 |
#if UINTMAX_MAX > 0xffffffffU |
| 62 |
#define ECB_PTRSIZE 8 |
| 63 |
#else |
| 64 |
#define ECB_PTRSIZE 4 |
| 65 |
#endif |
| 66 |
#endif |
| 67 |
|
| 68 |
/* work around x32 idiocy by defining proper macros */ |
| 69 |
#if __x86_64 || _M_AMD64 |
| 70 |
#if _ILP32 |
| 71 |
#define ECB_AMD64_X32 1 |
| 72 |
#else |
| 73 |
#define ECB_AMD64 1 |
| 74 |
#endif |
| 75 |
#endif |
| 76 |
|
| 77 |
/* many compilers define _GNUC_ to some versions but then only implement |
| 78 |
* what their idiot authors think are the "more important" extensions, |
| 79 |
* causing enormous grief in return for some better fake benchmark numbers. |
| 80 |
* or so. |
| 81 |
* we try to detect these and simply assume they are not gcc - if they have |
| 82 |
* an issue with that they should have done it right in the first place. |
| 83 |
*/ |
| 84 |
#ifndef ECB_GCC_VERSION |
| 85 |
#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
| 86 |
#define ECB_GCC_VERSION(major,minor) 0 |
| 87 |
#else |
| 88 |
#define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 89 |
#endif |
| 90 |
#endif |
| 91 |
|
| 92 |
#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
| 93 |
#define ECB_C99 (__STDC_VERSION__ >= 199901L) |
| 94 |
#define ECB_C11 (__STDC_VERSION__ >= 201112L) |
| 95 |
#define ECB_CPP (__cplusplus+0) |
| 96 |
#define ECB_CPP11 (__cplusplus >= 201103L) |
| 97 |
|
| 98 |
#if ECB_CPP |
| 99 |
#define ECB_EXTERN_C extern "C" |
| 100 |
#define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
| 101 |
#define ECB_EXTERN_C_END } |
| 102 |
#else |
| 103 |
#define ECB_EXTERN_C extern |
| 104 |
#define ECB_EXTERN_C_BEG |
| 105 |
#define ECB_EXTERN_C_END |
| 106 |
#endif |
| 107 |
|
| 108 |
/*****************************************************************************/ |
| 109 |
|
| 110 |
/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
| 111 |
/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
| 112 |
|
| 113 |
#if ECB_NO_THREADS |
| 114 |
#define ECB_NO_SMP 1 |
| 115 |
#endif |
| 116 |
|
| 117 |
#if ECB_NO_SMP |
| 118 |
#define ECB_MEMORY_FENCE do { } while (0) |
| 119 |
#endif |
| 120 |
|
| 121 |
#ifndef ECB_MEMORY_FENCE |
| 122 |
#if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 123 |
#if __i386 || __i386__ |
| 124 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
| 125 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 126 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 127 |
#elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
| 128 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 129 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 130 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 131 |
#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
| 132 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
| 133 |
#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
| 134 |
|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
| 135 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
| 136 |
#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
| 137 |
|| defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
| 138 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
| 139 |
#elif __sparc || __sparc__ |
| 140 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
| 141 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
| 142 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
| 143 |
#elif defined __s390__ || defined __s390x__ |
| 144 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
| 145 |
#elif defined __mips__ |
| 146 |
/* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
| 147 |
/* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
| 148 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
| 149 |
#elif defined __alpha__ |
| 150 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
| 151 |
#elif defined __hppa__ |
| 152 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 153 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 154 |
#elif defined __ia64__ |
| 155 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
| 156 |
#elif defined __m68k__ |
| 157 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 158 |
#elif defined __m88k__ |
| 159 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
| 160 |
#elif defined __sh__ |
| 161 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 162 |
#endif |
| 163 |
#endif |
| 164 |
#endif |
| 165 |
|
| 166 |
#ifndef ECB_MEMORY_FENCE |
| 167 |
#if ECB_GCC_VERSION(4,7) |
| 168 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
| 169 |
#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 170 |
|
| 171 |
/* The __has_feature syntax from clang is so misdesigned that we cannot use it |
| 172 |
* without risking compile time errors with other compilers. We *could* |
| 173 |
* define our own ecb_clang_has_feature, but I just can't be bothered to work |
| 174 |
* around this shit time and again. |
| 175 |
* #elif defined __clang && __has_feature (cxx_atomic) |
| 176 |
* // see comment below (stdatomic.h) about the C11 memory model. |
| 177 |
* #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 178 |
*/ |
| 179 |
|
| 180 |
#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 181 |
#define ECB_MEMORY_FENCE __sync_synchronize () |
| 182 |
#elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 183 |
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 184 |
#define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 185 |
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 186 |
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 187 |
#elif defined _WIN32 |
| 188 |
#include <WinNT.h> |
| 189 |
#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
| 190 |
#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 191 |
#include <mbarrier.h> |
| 192 |
#define ECB_MEMORY_FENCE __machine_rw_barrier () |
| 193 |
#define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
| 194 |
#define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
| 195 |
#elif __xlC__ |
| 196 |
#define ECB_MEMORY_FENCE __sync () |
| 197 |
#endif |
| 198 |
#endif |
| 199 |
|
| 200 |
#ifndef ECB_MEMORY_FENCE |
| 201 |
#if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
| 202 |
/* we assume that these memory fences work on all variables/all memory accesses, */ |
| 203 |
/* not just C11 atomics and atomic accesses */ |
| 204 |
#include <stdatomic.h> |
| 205 |
/* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
| 206 |
/* any fence other than seq_cst, which isn't very efficient for us. */ |
| 207 |
/* Why that is, we don't know - either the C11 memory model is quite useless */ |
| 208 |
/* for most usages, or gcc and clang have a bug */ |
| 209 |
/* I *currently* lean towards the latter, and inefficiently implement */ |
| 210 |
/* all three of ecb's fences as a seq_cst fence */ |
| 211 |
#define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
| 212 |
#endif |
| 213 |
#endif |
| 214 |
|
| 215 |
#ifndef ECB_MEMORY_FENCE |
| 216 |
#if !ECB_AVOID_PTHREADS |
| 217 |
/* |
| 218 |
* if you get undefined symbol references to pthread_mutex_lock, |
| 219 |
* or failure to find pthread.h, then you should implement |
| 220 |
* the ECB_MEMORY_FENCE operations for your cpu/compiler |
| 221 |
* OR provide pthread.h and link against the posix thread library |
| 222 |
* of your system. |
| 223 |
*/ |
| 224 |
#include <pthread.h> |
| 225 |
#define ECB_NEEDS_PTHREADS 1 |
| 226 |
#define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
| 227 |
|
| 228 |
static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
| 229 |
#define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
| 230 |
#endif |
| 231 |
#endif |
| 232 |
|
| 233 |
#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
| 234 |
#define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
| 235 |
#endif |
| 236 |
|
| 237 |
#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 238 |
#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 239 |
#endif |
| 240 |
|
| 241 |
/*****************************************************************************/ |
| 242 |
|
| 243 |
#if __cplusplus |
| 244 |
#define ecb_inline static inline |
| 245 |
#elif ECB_GCC_VERSION(2,5) |
| 246 |
#define ecb_inline static __inline__ |
| 247 |
#elif ECB_C99 |
| 248 |
#define ecb_inline static inline |
| 249 |
#else |
| 250 |
#define ecb_inline static |
| 251 |
#endif |
| 252 |
|
| 253 |
#if ECB_GCC_VERSION(3,3) |
| 254 |
#define ecb_restrict __restrict__ |
| 255 |
#elif ECB_C99 |
| 256 |
#define ecb_restrict restrict |
| 257 |
#else |
| 258 |
#define ecb_restrict |
| 259 |
#endif |
| 260 |
|
| 261 |
typedef int ecb_bool; |
| 262 |
|
| 263 |
#define ECB_CONCAT_(a, b) a ## b |
| 264 |
#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
| 265 |
#define ECB_STRINGIFY_(a) # a |
| 266 |
#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
| 267 |
|
| 268 |
#define ecb_function_ ecb_inline |
| 269 |
|
| 270 |
#if ECB_GCC_VERSION(3,1) |
| 271 |
#define ecb_attribute(attrlist) __attribute__(attrlist) |
| 272 |
#define ecb_is_constant(expr) __builtin_constant_p (expr) |
| 273 |
#define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
| 274 |
#define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 275 |
#else |
| 276 |
#define ecb_attribute(attrlist) |
| 277 |
#define ecb_is_constant(expr) 0 |
| 278 |
#define ecb_expect(expr,value) (expr) |
| 279 |
#define ecb_prefetch(addr,rw,locality) |
| 280 |
#endif |
| 281 |
|
| 282 |
/* no emulation for ecb_decltype */ |
| 283 |
#if ECB_GCC_VERSION(4,5) |
| 284 |
#define ecb_decltype(x) __decltype(x) |
| 285 |
#elif ECB_GCC_VERSION(3,0) |
| 286 |
#define ecb_decltype(x) __typeof(x) |
| 287 |
#endif |
| 288 |
|
| 289 |
#define ecb_noinline ecb_attribute ((__noinline__)) |
| 290 |
#define ecb_unused ecb_attribute ((__unused__)) |
| 291 |
#define ecb_const ecb_attribute ((__const__)) |
| 292 |
#define ecb_pure ecb_attribute ((__pure__)) |
| 293 |
|
| 294 |
#if ECB_C11 |
| 295 |
#define ecb_noreturn _Noreturn |
| 296 |
#else |
| 297 |
#define ecb_noreturn ecb_attribute ((__noreturn__)) |
| 298 |
#endif |
| 299 |
|
| 300 |
#if ECB_GCC_VERSION(4,3) |
| 301 |
#define ecb_artificial ecb_attribute ((__artificial__)) |
| 302 |
#define ecb_hot ecb_attribute ((__hot__)) |
| 303 |
#define ecb_cold ecb_attribute ((__cold__)) |
| 304 |
#else |
| 305 |
#define ecb_artificial |
| 306 |
#define ecb_hot |
| 307 |
#define ecb_cold |
| 308 |
#endif |
| 309 |
|
| 310 |
/* put around conditional expressions if you are very sure that the */ |
| 311 |
/* expression is mostly true or mostly false. note that these return */ |
| 312 |
/* booleans, not the expression. */ |
| 313 |
#define ecb_expect_false(expr) ecb_expect (!!(expr), 0) |
| 314 |
#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) |
| 315 |
/* for compatibility to the rest of the world */ |
| 316 |
#define ecb_likely(expr) ecb_expect_true (expr) |
| 317 |
#define ecb_unlikely(expr) ecb_expect_false (expr) |
| 318 |
|
| 319 |
/* count trailing zero bits and count # of one bits */ |
| 320 |
#if ECB_GCC_VERSION(3,4) |
| 321 |
/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
| 322 |
#define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
| 323 |
#define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
| 324 |
#define ecb_ctz32(x) __builtin_ctz (x) |
| 325 |
#define ecb_ctz64(x) __builtin_ctzll (x) |
| 326 |
#define ecb_popcount32(x) __builtin_popcount (x) |
| 327 |
/* no popcountll */ |
| 328 |
#else |
| 329 |
ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
| 330 |
ecb_function_ int |
| 331 |
ecb_ctz32 (uint32_t x) |
| 332 |
{ |
| 333 |
int r = 0; |
| 334 |
|
| 335 |
x &= ~x + 1; /* this isolates the lowest bit */ |
| 336 |
|
| 337 |
#if ECB_branchless_on_i386 |
| 338 |
r += !!(x & 0xaaaaaaaa) << 0; |
| 339 |
r += !!(x & 0xcccccccc) << 1; |
| 340 |
r += !!(x & 0xf0f0f0f0) << 2; |
| 341 |
r += !!(x & 0xff00ff00) << 3; |
| 342 |
r += !!(x & 0xffff0000) << 4; |
| 343 |
#else |
| 344 |
if (x & 0xaaaaaaaa) r += 1; |
| 345 |
if (x & 0xcccccccc) r += 2; |
| 346 |
if (x & 0xf0f0f0f0) r += 4; |
| 347 |
if (x & 0xff00ff00) r += 8; |
| 348 |
if (x & 0xffff0000) r += 16; |
| 349 |
#endif |
| 350 |
|
| 351 |
return r; |
| 352 |
} |
| 353 |
|
| 354 |
ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
| 355 |
ecb_function_ int |
| 356 |
ecb_ctz64 (uint64_t x) |
| 357 |
{ |
| 358 |
int shift = x & 0xffffffffU ? 0 : 32; |
| 359 |
return ecb_ctz32 (x >> shift) + shift; |
| 360 |
} |
| 361 |
|
| 362 |
ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
| 363 |
ecb_function_ int |
| 364 |
ecb_popcount32 (uint32_t x) |
| 365 |
{ |
| 366 |
x -= (x >> 1) & 0x55555555; |
| 367 |
x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
| 368 |
x = ((x >> 4) + x) & 0x0f0f0f0f; |
| 369 |
x *= 0x01010101; |
| 370 |
|
| 371 |
return x >> 24; |
| 372 |
} |
| 373 |
|
| 374 |
ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
| 375 |
ecb_function_ int ecb_ld32 (uint32_t x) |
| 376 |
{ |
| 377 |
int r = 0; |
| 378 |
|
| 379 |
if (x >> 16) { x >>= 16; r += 16; } |
| 380 |
if (x >> 8) { x >>= 8; r += 8; } |
| 381 |
if (x >> 4) { x >>= 4; r += 4; } |
| 382 |
if (x >> 2) { x >>= 2; r += 2; } |
| 383 |
if (x >> 1) { r += 1; } |
| 384 |
|
| 385 |
return r; |
| 386 |
} |
| 387 |
|
| 388 |
ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
| 389 |
ecb_function_ int ecb_ld64 (uint64_t x) |
| 390 |
{ |
| 391 |
int r = 0; |
| 392 |
|
| 393 |
if (x >> 32) { x >>= 32; r += 32; } |
| 394 |
|
| 395 |
return r + ecb_ld32 (x); |
| 396 |
} |
| 397 |
#endif |
| 398 |
|
| 399 |
ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
| 400 |
ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
| 401 |
ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
| 402 |
ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
| 403 |
|
| 404 |
ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
| 405 |
ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
| 406 |
{ |
| 407 |
return ( (x * 0x0802U & 0x22110U) |
| 408 |
| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
| 409 |
} |
| 410 |
|
| 411 |
ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
| 412 |
ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
| 413 |
{ |
| 414 |
x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
| 415 |
x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
| 416 |
x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
| 417 |
x = ( x >> 8 ) | ( x << 8); |
| 418 |
|
| 419 |
return x; |
| 420 |
} |
| 421 |
|
| 422 |
ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
| 423 |
ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
| 424 |
{ |
| 425 |
x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
| 426 |
x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
| 427 |
x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
| 428 |
x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
| 429 |
x = ( x >> 16 ) | ( x << 16); |
| 430 |
|
| 431 |
return x; |
| 432 |
} |
| 433 |
|
| 434 |
/* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 435 |
/* so for this version we are lazy */ |
| 436 |
ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
| 437 |
ecb_function_ int |
| 438 |
ecb_popcount64 (uint64_t x) |
| 439 |
{ |
| 440 |
return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
| 441 |
} |
| 442 |
|
| 443 |
ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
| 444 |
ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
| 445 |
ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
| 446 |
ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
| 447 |
ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
| 448 |
ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
| 449 |
ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
| 450 |
ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
| 451 |
|
| 452 |
ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
| 453 |
ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
| 454 |
ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
| 455 |
ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
| 456 |
ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
| 457 |
ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
| 458 |
ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
| 459 |
ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
| 460 |
|
| 461 |
#if ECB_GCC_VERSION(4,3) |
| 462 |
#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
| 463 |
#define ecb_bswap32(x) __builtin_bswap32 (x) |
| 464 |
#define ecb_bswap64(x) __builtin_bswap64 (x) |
| 465 |
#else |
| 466 |
ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
| 467 |
ecb_function_ uint16_t |
| 468 |
ecb_bswap16 (uint16_t x) |
| 469 |
{ |
| 470 |
return ecb_rotl16 (x, 8); |
| 471 |
} |
| 472 |
|
| 473 |
ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
| 474 |
ecb_function_ uint32_t |
| 475 |
ecb_bswap32 (uint32_t x) |
| 476 |
{ |
| 477 |
return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
| 478 |
} |
| 479 |
|
| 480 |
ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
| 481 |
ecb_function_ uint64_t |
| 482 |
ecb_bswap64 (uint64_t x) |
| 483 |
{ |
| 484 |
return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 485 |
} |
| 486 |
#endif |
| 487 |
|
| 488 |
#if ECB_GCC_VERSION(4,5) |
| 489 |
#define ecb_unreachable() __builtin_unreachable () |
| 490 |
#else |
| 491 |
/* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 492 |
ecb_inline void ecb_unreachable (void) ecb_noreturn; |
| 493 |
ecb_inline void ecb_unreachable (void) { } |
| 494 |
#endif |
| 495 |
|
| 496 |
/* try to tell the compiler that some condition is definitely true */ |
| 497 |
#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 498 |
|
| 499 |
ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
| 500 |
ecb_inline unsigned char |
| 501 |
ecb_byteorder_helper (void) |
| 502 |
{ |
| 503 |
/* the union code still generates code under pressure in gcc, */ |
| 504 |
/* but less than using pointers, and always seems to */ |
| 505 |
/* successfully return a constant. */ |
| 506 |
/* the reason why we have this horrible preprocessor mess */ |
| 507 |
/* is to avoid it in all cases, at least on common architectures */ |
| 508 |
/* or when using a recent enough gcc version (>= 4.6) */ |
| 509 |
#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
| 510 |
return 0x44; |
| 511 |
#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
| 512 |
return 0x44; |
| 513 |
#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
| 514 |
return 0x11; |
| 515 |
#else |
| 516 |
union |
| 517 |
{ |
| 518 |
uint32_t i; |
| 519 |
uint8_t c; |
| 520 |
} u = { 0x11223344 }; |
| 521 |
return u.c; |
| 522 |
#endif |
| 523 |
} |
| 524 |
|
| 525 |
ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
| 526 |
ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
| 527 |
ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
| 528 |
ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
| 529 |
|
| 530 |
#if ECB_GCC_VERSION(3,0) || ECB_C99 |
| 531 |
#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 532 |
#else |
| 533 |
#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
| 534 |
#endif |
| 535 |
|
| 536 |
#if __cplusplus |
| 537 |
template<typename T> |
| 538 |
static inline T ecb_div_rd (T val, T div) |
| 539 |
{ |
| 540 |
return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
| 541 |
} |
| 542 |
template<typename T> |
| 543 |
static inline T ecb_div_ru (T val, T div) |
| 544 |
{ |
| 545 |
return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
| 546 |
} |
| 547 |
#else |
| 548 |
#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
| 549 |
#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 550 |
#endif |
| 551 |
|
| 552 |
#if ecb_cplusplus_does_not_suck |
| 553 |
/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
| 554 |
template<typename T, int N> |
| 555 |
static inline int ecb_array_length (const T (&arr)[N]) |
| 556 |
{ |
| 557 |
return N; |
| 558 |
} |
| 559 |
#else |
| 560 |
#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 561 |
#endif |
| 562 |
|
| 563 |
/*******************************************************************************/ |
| 564 |
/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
| 565 |
|
| 566 |
/* basically, everything uses "ieee pure-endian" floating point numbers */ |
| 567 |
/* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
| 568 |
#if 0 \ |
| 569 |
|| __i386 || __i386__ \ |
| 570 |
|| __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
| 571 |
|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
| 572 |
|| defined __arm__ && defined __ARM_EABI__ \ |
| 573 |
|| defined __s390__ || defined __s390x__ \ |
| 574 |
|| defined __mips__ \ |
| 575 |
|| defined __alpha__ \ |
| 576 |
|| defined __hppa__ \ |
| 577 |
|| defined __ia64__ \ |
| 578 |
|| defined __m68k__ \ |
| 579 |
|| defined __m88k__ \ |
| 580 |
|| defined __sh__ \ |
| 581 |
|| defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
| 582 |
#define ECB_STDFP 1 |
| 583 |
#include <string.h> /* for memcpy */ |
| 584 |
#else |
| 585 |
#define ECB_STDFP 0 |
| 586 |
#endif |
| 587 |
|
| 588 |
#ifndef ECB_NO_LIBM |
| 589 |
|
| 590 |
#include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
| 591 |
|
| 592 |
/* only the oldest of old doesn't have this one. solaris. */ |
| 593 |
#ifdef INFINITY |
| 594 |
#define ECB_INFINITY INFINITY |
| 595 |
#else |
| 596 |
#define ECB_INFINITY HUGE_VAL |
| 597 |
#endif |
| 598 |
|
| 599 |
#ifdef NAN |
| 600 |
#define ECB_NAN NAN |
| 601 |
#else |
| 602 |
#define ECB_NAN ECB_INFINITY |
| 603 |
#endif |
| 604 |
|
| 605 |
/* converts an ieee half/binary16 to a float */ |
| 606 |
ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
| 607 |
ecb_function_ float |
| 608 |
ecb_binary16_to_float (uint16_t x) |
| 609 |
{ |
| 610 |
int e = (x >> 10) & 0x1f; |
| 611 |
int m = x & 0x3ff; |
| 612 |
float r; |
| 613 |
|
| 614 |
if (!e ) r = ldexpf (m , -24); |
| 615 |
else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
| 616 |
else if (m ) r = ECB_NAN; |
| 617 |
else r = ECB_INFINITY; |
| 618 |
|
| 619 |
return x & 0x8000 ? -r : r; |
| 620 |
} |
| 621 |
|
| 622 |
/* convert a float to ieee single/binary32 */ |
| 623 |
ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
| 624 |
ecb_function_ uint32_t |
| 625 |
ecb_float_to_binary32 (float x) |
| 626 |
{ |
| 627 |
uint32_t r; |
| 628 |
|
| 629 |
#if ECB_STDFP |
| 630 |
memcpy (&r, &x, 4); |
| 631 |
#else |
| 632 |
/* slow emulation, works for anything but -0 */ |
| 633 |
uint32_t m; |
| 634 |
int e; |
| 635 |
|
| 636 |
if (x == 0e0f ) return 0x00000000U; |
| 637 |
if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
| 638 |
if (x < -3.40282346638528860e+38f) return 0xff800000U; |
| 639 |
if (x != x ) return 0x7fbfffffU; |
| 640 |
|
| 641 |
m = frexpf (x, &e) * 0x1000000U; |
| 642 |
|
| 643 |
r = m & 0x80000000U; |
| 644 |
|
| 645 |
if (r) |
| 646 |
m = -m; |
| 647 |
|
| 648 |
if (e <= -126) |
| 649 |
{ |
| 650 |
m &= 0xffffffU; |
| 651 |
m >>= (-125 - e); |
| 652 |
e = -126; |
| 653 |
} |
| 654 |
|
| 655 |
r |= (e + 126) << 23; |
| 656 |
r |= m & 0x7fffffU; |
| 657 |
#endif |
| 658 |
|
| 659 |
return r; |
| 660 |
} |
| 661 |
|
| 662 |
/* converts an ieee single/binary32 to a float */ |
| 663 |
ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
| 664 |
ecb_function_ float |
| 665 |
ecb_binary32_to_float (uint32_t x) |
| 666 |
{ |
| 667 |
float r; |
| 668 |
|
| 669 |
#if ECB_STDFP |
| 670 |
memcpy (&r, &x, 4); |
| 671 |
#else |
| 672 |
/* emulation, only works for normals and subnormals and +0 */ |
| 673 |
int neg = x >> 31; |
| 674 |
int e = (x >> 23) & 0xffU; |
| 675 |
|
| 676 |
x &= 0x7fffffU; |
| 677 |
|
| 678 |
if (e) |
| 679 |
x |= 0x800000U; |
| 680 |
else |
| 681 |
e = 1; |
| 682 |
|
| 683 |
/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
| 684 |
r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
| 685 |
|
| 686 |
r = neg ? -r : r; |
| 687 |
#endif |
| 688 |
|
| 689 |
return r; |
| 690 |
} |
| 691 |
|
| 692 |
/* convert a double to ieee double/binary64 */ |
| 693 |
ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
| 694 |
ecb_function_ uint64_t |
| 695 |
ecb_double_to_binary64 (double x) |
| 696 |
{ |
| 697 |
uint64_t r; |
| 698 |
|
| 699 |
#if ECB_STDFP |
| 700 |
memcpy (&r, &x, 8); |
| 701 |
#else |
| 702 |
/* slow emulation, works for anything but -0 */ |
| 703 |
uint64_t m; |
| 704 |
int e; |
| 705 |
|
| 706 |
if (x == 0e0 ) return 0x0000000000000000U; |
| 707 |
if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
| 708 |
if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
| 709 |
if (x != x ) return 0X7ff7ffffffffffffU; |
| 710 |
|
| 711 |
m = frexp (x, &e) * 0x20000000000000U; |
| 712 |
|
| 713 |
r = m & 0x8000000000000000;; |
| 714 |
|
| 715 |
if (r) |
| 716 |
m = -m; |
| 717 |
|
| 718 |
if (e <= -1022) |
| 719 |
{ |
| 720 |
m &= 0x1fffffffffffffU; |
| 721 |
m >>= (-1021 - e); |
| 722 |
e = -1022; |
| 723 |
} |
| 724 |
|
| 725 |
r |= ((uint64_t)(e + 1022)) << 52; |
| 726 |
r |= m & 0xfffffffffffffU; |
| 727 |
#endif |
| 728 |
|
| 729 |
return r; |
| 730 |
} |
| 731 |
|
| 732 |
/* converts an ieee double/binary64 to a double */ |
| 733 |
ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
| 734 |
ecb_function_ double |
| 735 |
ecb_binary64_to_double (uint64_t x) |
| 736 |
{ |
| 737 |
double r; |
| 738 |
|
| 739 |
#if ECB_STDFP |
| 740 |
memcpy (&r, &x, 8); |
| 741 |
#else |
| 742 |
/* emulation, only works for normals and subnormals and +0 */ |
| 743 |
int neg = x >> 63; |
| 744 |
int e = (x >> 52) & 0x7ffU; |
| 745 |
|
| 746 |
x &= 0xfffffffffffffU; |
| 747 |
|
| 748 |
if (e) |
| 749 |
x |= 0x10000000000000U; |
| 750 |
else |
| 751 |
e = 1; |
| 752 |
|
| 753 |
/* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
| 754 |
r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
| 755 |
|
| 756 |
r = neg ? -r : r; |
| 757 |
#endif |
| 758 |
|
| 759 |
return r; |
| 760 |
} |
| 761 |
|
| 762 |
#endif |
| 763 |
|
| 764 |
#endif |
| 765 |
|
