Revision 1.165 by

Revision 1.214 by

1 | /* | 1 | /* |
---|---|---|---|

2 | * libecb - http://software.schmorp.de/pkg/libecb | 2 | * libecb - http://software.schmorp.de/pkg/libecb |

3 | * | 3 | * |

4 | * Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de> | 4 | * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de> |

5 | * Copyright (©) 2011 Emanuele Giaquinta | 5 | * Copyright (©) 2011 Emanuele Giaquinta |

6 | * All rights reserved. | 6 | * All rights reserved. |

7 | * | 7 | * |

8 | * Redistribution and use in source and binary forms, with or without modifica- | 8 | * Redistribution and use in source and binary forms, with or without modifica- |

9 | * tion, are permitted provided that the following conditions are met: | 9 | * tion, are permitted provided that the following conditions are met: |

… | … | ||

40 | 40 | ||

41 | #ifndef ECB_H | 41 | #ifndef ECB_H |

42 | #define ECB_H | 42 | #define ECB_H |

43 | 43 | ||

44 | /* 16 bits major, 16 bits minor */ | 44 | /* 16 bits major, 16 bits minor */ |

45 | #define ECB_VERSION 0x00010004 | 45 | #define ECB_VERSION 0x0001000c |

46 | 46 | ||

47 | #ifdef _WIN32 | 47 | #include <string.h> /* for memcpy */ |

48 | |||

49 | #if defined (_WIN32) && !defined (__MINGW32__) | ||

48 | typedef signed char int8_t; | 50 | typedef signed char int8_t; |

49 | typedef unsigned char uint8_t; | 51 | typedef unsigned char uint8_t; |

52 | typedef signed char int_fast8_t; | ||

53 | typedef unsigned char uint_fast8_t; | ||

50 | typedef signed short int16_t; | 54 | typedef signed short int16_t; |

51 | typedef unsigned short uint16_t; | 55 | typedef unsigned short uint16_t; |

56 | typedef signed int int_fast16_t; | ||

57 | typedef unsigned int uint_fast16_t; | ||

52 | typedef signed int int32_t; | 58 | typedef signed int int32_t; |

53 | typedef unsigned int uint32_t; | 59 | typedef unsigned int uint32_t; |

60 | typedef signed int int_fast32_t; | ||

61 | typedef unsigned int uint_fast32_t; | ||

54 | #if __GNUC__ | 62 | #if __GNUC__ |

55 | typedef signed long long int64_t; | 63 | typedef signed long long int64_t; |

56 | typedef unsigned long long uint64_t; | 64 | typedef unsigned long long uint64_t; |

57 | #else /* _MSC_VER || __BORLANDC__ */ | 65 | #else /* _MSC_VER || __BORLANDC__ */ |

58 | typedef signed __int64 int64_t; | 66 | typedef signed __int64 int64_t; |

59 | typedef unsigned __int64 uint64_t; | 67 | typedef unsigned __int64 uint64_t; |

60 | #endif | 68 | #endif |

69 | typedef int64_t int_fast64_t; | ||

70 | typedef uint64_t uint_fast64_t; | ||

61 | #ifdef _WIN64 | 71 | #ifdef _WIN64 |

62 | #define ECB_PTRSIZE 8 | 72 | #define ECB_PTRSIZE 8 |

63 | typedef uint64_t uintptr_t; | 73 | typedef uint64_t uintptr_t; |

64 | typedef int64_t intptr_t; | 74 | typedef int64_t intptr_t; |

65 | #else | 75 | #else |

… | … | ||

67 | typedef uint32_t uintptr_t; | 77 | typedef uint32_t uintptr_t; |

68 | typedef int32_t intptr_t; | 78 | typedef int32_t intptr_t; |

69 | #endif | 79 | #endif |

70 | #else | 80 | #else |

71 | #include <inttypes.h> | 81 | #include <inttypes.h> |

72 | #if UINTMAX_MAX > 0xffffffffU | 82 | #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |

73 | #define ECB_PTRSIZE 8 | 83 | #define ECB_PTRSIZE 8 |

74 | #else | 84 | #else |

75 | #define ECB_PTRSIZE 4 | 85 | #define ECB_PTRSIZE 4 |

76 | #endif | 86 | #endif |

77 | #endif | 87 | #endif |

78 | 88 | ||

79 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) | 89 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |

80 | #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) | 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 | ||

81 | 99 | ||

82 | /* work around x32 idiocy by defining proper macros */ | 100 | /* work around x32 idiocy by defining proper macros */ |

83 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 | 101 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |

84 | #if _ILP32 | 102 | #if _ILP32 |

85 | #define ECB_AMD64_X32 1 | 103 | #define ECB_AMD64_X32 1 |

86 | #else | 104 | #else |

87 | #define ECB_AMD64 1 | 105 | #define ECB_AMD64 1 |

88 | #endif | 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 | ||

89 | #endif | 113 | #endif |

90 | 114 | ||

91 | /* many compilers define _GNUC_ to some versions but then only implement | 115 | /* many compilers define _GNUC_ to some versions but then only implement |

92 | * what their idiot authors think are the "more important" extensions, | 116 | * what their idiot authors think are the "more important" extensions, |

93 | * causing enormous grief in return for some better fake benchmark numbers. | 117 | * causing enormous grief in return for some better fake benchmark numbers. |

… | … | ||

115 | #define ECB_CLANG_EXTENSION(x) 0 | 139 | #define ECB_CLANG_EXTENSION(x) 0 |

116 | #endif | 140 | #endif |

117 | 141 | ||

118 | #define ECB_CPP (__cplusplus+0) | 142 | #define ECB_CPP (__cplusplus+0) |

119 | #define ECB_CPP11 (__cplusplus >= 201103L) | 143 | #define ECB_CPP11 (__cplusplus >= 201103L) |

144 | #define ECB_CPP14 (__cplusplus >= 201402L) | ||

145 | #define ECB_CPP17 (__cplusplus >= 201703L) | ||

120 | 146 | ||

121 | #if ECB_CPP | 147 | #if ECB_CPP |

122 | #define ECB_C 0 | 148 | #define ECB_C 0 |

123 | #define ECB_STDC_VERSION 0 | 149 | #define ECB_STDC_VERSION 0 |

124 | #else | 150 | #else |

… | … | ||

126 | #define ECB_STDC_VERSION __STDC_VERSION__ | 152 | #define ECB_STDC_VERSION __STDC_VERSION__ |

127 | #endif | 153 | #endif |

128 | 154 | ||

129 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) | 155 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |

130 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) | 156 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |

157 | #define ECB_C17 (ECB_STDC_VERSION >= 201710L) | ||

131 | 158 | ||

132 | #if ECB_CPP | 159 | #if ECB_CPP |

133 | #define ECB_EXTERN_C extern "C" | 160 | #define ECB_EXTERN_C extern "C" |

134 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { | 161 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |

135 | #define ECB_EXTERN_C_END } | 162 | #define ECB_EXTERN_C_END } |

… | … | ||

155 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ | 182 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |

156 | #if __xlC__ && ECB_CPP | 183 | #if __xlC__ && ECB_CPP |

157 | #include <builtins.h> | 184 | #include <builtins.h> |

158 | #endif | 185 | #endif |

159 | 186 | ||

187 | #if 1400 <= _MSC_VER | ||

188 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ | ||

189 | #endif | ||

190 | |||

160 | #ifndef ECB_MEMORY_FENCE | 191 | #ifndef ECB_MEMORY_FENCE |

161 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 | 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") | ||

162 | #if __i386 || __i386__ | 194 | #if __i386 || __i386__ |

163 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") | 195 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |

164 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") | 196 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |

165 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") | 197 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |

166 | #elif ECB_GCC_AMD64 | 198 | #elif ECB_GCC_AMD64 |

167 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") | 199 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |

168 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") | 200 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |

169 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") | 201 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |

170 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ | 202 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |

171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") | 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 */ | ||

172 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ | 211 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |

173 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ | 212 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |

213 | || defined __ARM_ARCH_6T2__ | ||

174 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") | 214 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |

175 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ | 215 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |

176 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ | 216 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |

177 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") | 217 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |

178 | #elif __aarch64__ | 218 | #elif __aarch64__ |

179 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") | 219 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |

180 | #elif (__sparc || __sparc__) && !__sparcv8 | 220 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |

181 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") | 221 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |

182 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") | 222 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |

183 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") | 223 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |

184 | #elif defined __s390__ || defined __s390x__ | 224 | #elif defined __s390__ || defined __s390x__ |

185 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") | 225 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |

… | … | ||

208 | #if ECB_GCC_VERSION(4,7) | 248 | #if ECB_GCC_VERSION(4,7) |

209 | /* see comment below (stdatomic.h) about the C11 memory model. */ | 249 | /* see comment below (stdatomic.h) about the C11 memory model. */ |

210 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) | 250 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |

211 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) | 251 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |

212 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) | 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) | ||

213 | 255 | ||

214 | #elif ECB_CLANG_EXTENSION(c_atomic) | 256 | #elif ECB_CLANG_EXTENSION(c_atomic) |

215 | /* see comment below (stdatomic.h) about the C11 memory model. */ | 257 | /* see comment below (stdatomic.h) about the C11 memory model. */ |

216 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) | 258 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |

217 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) | 259 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |

218 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) | 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) | ||

219 | 263 | ||

220 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ | 264 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |

221 | #define ECB_MEMORY_FENCE __sync_synchronize () | 265 | #define ECB_MEMORY_FENCE __sync_synchronize () |

222 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ | 266 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |

223 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ | 267 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |

… | … | ||

233 | #elif defined _WIN32 | 277 | #elif defined _WIN32 |

234 | #include <WinNT.h> | 278 | #include <WinNT.h> |

235 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ | 279 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |

236 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 | 280 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |

237 | #include <mbarrier.h> | 281 | #include <mbarrier.h> |

238 | #define ECB_MEMORY_FENCE __machine_rw_barrier () | 282 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |

239 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () | 283 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |

240 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () | 284 | #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |

285 | #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () | ||

241 | #elif __xlC__ | 286 | #elif __xlC__ |

242 | #define ECB_MEMORY_FENCE __sync () | 287 | #define ECB_MEMORY_FENCE __sync () |

243 | #endif | 288 | #endif |

244 | #endif | 289 | #endif |

245 | 290 | ||

246 | #ifndef ECB_MEMORY_FENCE | 291 | #ifndef ECB_MEMORY_FENCE |

247 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ | 292 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |

248 | /* we assume that these memory fences work on all variables/all memory accesses, */ | 293 | /* we assume that these memory fences work on all variables/all memory accesses, */ |

249 | /* not just C11 atomics and atomic accesses */ | 294 | /* not just C11 atomics and atomic accesses */ |

250 | #include <stdatomic.h> | 295 | #include <stdatomic.h> |

251 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ | ||

252 | /* any fence other than seq_cst, which isn't very efficient for us. */ | ||

253 | /* Why that is, we don't know - either the C11 memory model is quite useless */ | ||

254 | /* for most usages, or gcc and clang have a bug */ | ||

255 | /* I *currently* lean towards the latter, and inefficiently implement */ | ||

256 | /* all three of ecb's fences as a seq_cst fence */ | ||

257 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ | ||

258 | /* for all __atomic_thread_fence's except seq_cst */ | ||

259 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) | 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) | ||

260 | #endif | 299 | #endif |

261 | #endif | 300 | #endif |

262 | 301 | ||

263 | #ifndef ECB_MEMORY_FENCE | 302 | #ifndef ECB_MEMORY_FENCE |

264 | #if !ECB_AVOID_PTHREADS | 303 | #if !ECB_AVOID_PTHREADS |

… | … | ||

284 | 323 | ||

285 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE | 324 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |

286 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE | 325 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |

287 | #endif | 326 | #endif |

288 | 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 | |||

289 | /*****************************************************************************/ | 332 | /*****************************************************************************/ |

290 | 333 | ||

291 | #if ECB_CPP | 334 | #if ECB_CPP |

292 | #define ecb_inline static inline | 335 | #define ecb_inline static inline |

293 | #elif ECB_GCC_VERSION(2,5) | 336 | #elif ECB_GCC_VERSION(2,5) |

… | … | ||

312 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) | 355 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |

313 | #define ECB_STRINGIFY_(a) # a | 356 | #define ECB_STRINGIFY_(a) # a |

314 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) | 357 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |

315 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) | 358 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |

316 | 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? */ | ||

317 | #define ecb_function_ ecb_inline | 364 | #define ecb_function_ ecb_inline |

318 | 365 | ||

319 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) | 366 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |

320 | #define ecb_attribute(attrlist) __attribute__ (attrlist) | 367 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |

321 | #else | 368 | #else |

… | … | ||

411 | /* count trailing zero bits and count # of one bits */ | 458 | /* count trailing zero bits and count # of one bits */ |

412 | #if ECB_GCC_VERSION(3,4) \ | 459 | #if ECB_GCC_VERSION(3,4) \ |

413 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ | 460 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |

414 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ | 461 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |

415 | && ECB_CLANG_BUILTIN(__builtin_popcount)) | 462 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |

416 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ | ||

417 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) | ||

418 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) | ||

419 | #define ecb_ctz32(x) __builtin_ctz (x) | 463 | #define ecb_ctz32(x) __builtin_ctz (x) |

464 | #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x)) | ||

420 | #define ecb_ctz64(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) | ||

421 | #define ecb_popcount32(x) __builtin_popcount (x) | 469 | #define ecb_popcount32(x) __builtin_popcount (x) |

422 | /* no popcountll */ | 470 | /* ecb_popcount64 is more difficult, see below */ |

423 | #else | 471 | #else |

424 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); | 472 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |

425 | ecb_function_ ecb_const int | 473 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x) |

426 | ecb_ctz32 (uint32_t x) | ||

427 | { | 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 | ||

428 | int r = 0; | 480 | int r; |

429 | 481 | ||

430 | x &= ~x + 1; /* this isolates the lowest bit */ | 482 | x &= ~x + 1; /* this isolates the lowest bit */ |

431 | 483 | ||

432 | #if ECB_branchless_on_i386 | 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; | ||

433 | r += !!(x & 0xaaaaaaaa) << 0; | 503 | r += !!(x & 0xaaaaaaaa) << 0; |

434 | r += !!(x & 0xcccccccc) << 1; | 504 | r += !!(x & 0xcccccccc) << 1; |

435 | r += !!(x & 0xf0f0f0f0) << 2; | 505 | r += !!(x & 0xf0f0f0f0) << 2; |

436 | r += !!(x & 0xff00ff00) << 3; | 506 | r += !!(x & 0xff00ff00) << 3; |

437 | r += !!(x & 0xffff0000) << 4; | 507 | r += !!(x & 0xffff0000) << 4; |

438 | #else | 508 | #else /* branchless on modern compilers, typically */ |

509 | r = 0; | ||

439 | if (x & 0xaaaaaaaa) r += 1; | 510 | if (x & 0xaaaaaaaa) r += 1; |

440 | if (x & 0xcccccccc) r += 2; | 511 | if (x & 0xcccccccc) r += 2; |

441 | if (x & 0xf0f0f0f0) r += 4; | 512 | if (x & 0xf0f0f0f0) r += 4; |

442 | if (x & 0xff00ff00) r += 8; | 513 | if (x & 0xff00ff00) r += 8; |

443 | if (x & 0xffff0000) r += 16; | 514 | if (x & 0xffff0000) r += 16; |

444 | #endif | 515 | #endif |

445 | 516 | ||

446 | return r; | 517 | return r; |

518 | #endif | ||

447 | } | 519 | } |

448 | 520 | ||

449 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); | 521 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |

450 | ecb_function_ ecb_const int | 522 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x) |

451 | ecb_ctz64 (uint64_t x) | ||

452 | { | 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 | ||

453 | int shift = x & 0xffffffffU ? 0 : 32; | 529 | int shift = x & 0xffffffff ? 0 : 32; |

454 | return ecb_ctz32 (x >> shift) + shift; | 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 (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 (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 | ||

455 | } | 581 | } |

456 | 582 | ||

457 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); | 583 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |

458 | ecb_function_ ecb_const int | 584 | ecb_function_ ecb_const int |

459 | ecb_popcount32 (uint32_t x) | 585 | ecb_popcount32 (uint32_t x) |

… | … | ||

467 | } | 593 | } |

468 | 594 | ||

469 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); | 595 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |

470 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) | 596 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |

471 | { | 597 | { |

598 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) | ||

599 | unsigned long r; | ||

600 | _BitScanReverse (&r, x); | ||

601 | return (int)r; | ||

602 | #else | ||

472 | int r = 0; | 603 | int r = 0; |

473 | 604 | ||

474 | if (x >> 16) { x >>= 16; r += 16; } | 605 | if (x >> 16) { x >>= 16; r += 16; } |

475 | if (x >> 8) { x >>= 8; r += 8; } | 606 | if (x >> 8) { x >>= 8; r += 8; } |

476 | if (x >> 4) { x >>= 4; r += 4; } | 607 | if (x >> 4) { x >>= 4; r += 4; } |

477 | if (x >> 2) { x >>= 2; r += 2; } | 608 | if (x >> 2) { x >>= 2; r += 2; } |

478 | if (x >> 1) { r += 1; } | 609 | if (x >> 1) { r += 1; } |

479 | 610 | ||

480 | return r; | 611 | return r; |

612 | #endif | ||

481 | } | 613 | } |

482 | 614 | ||

483 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); | 615 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |

484 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) | 616 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |

485 | { | 617 | { |

618 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) | ||

619 | unsigned long r; | ||

620 | _BitScanReverse64 (&r, x); | ||

621 | return (int)r; | ||

622 | #else | ||

486 | int r = 0; | 623 | int r = 0; |

487 | 624 | ||

488 | if (x >> 32) { x >>= 32; r += 32; } | 625 | if (x >> 32) { x >>= 32; r += 32; } |

489 | 626 | ||

490 | return r + ecb_ld32 (x); | 627 | return r + ecb_ld32 (x); |

628 | #endif | ||

491 | } | 629 | } |

492 | #endif | 630 | #endif |

493 | 631 | ||

494 | 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); |

495 | 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_pot32 (uint32_t x) { return !(x & (x - 1)); } |

… | … | ||

524 | x = ( x >> 16 ) | ( x << 16); | 662 | x = ( x >> 16 ) | ( x << 16); |

525 | 663 | ||

526 | return x; | 664 | return x; |

527 | } | 665 | } |

528 | 666 | ||

529 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ | ||

530 | /* so for this version we are lazy */ | ||

531 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); | 667 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |

532 | ecb_function_ ecb_const int | 668 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x) |

533 | ecb_popcount64 (uint64_t x) | ||

534 | { | 669 | { |

670 | /* popcount64 is only available on 64 bit cpus as gcc builtin. */ | ||

671 | /* also, gcc/clang make this surprisingly difficult to use */ | ||

672 | #if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl)) | ||

673 | return __builtin_popcountl (x); | ||

674 | #else | ||

535 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); | 675 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |

676 | #endif | ||

536 | } | 677 | } |

537 | 678 | ||

538 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); | 679 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |

539 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); | 680 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |

540 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); | 681 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |

… | … | ||

542 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); | 683 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |

543 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); | 684 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |

544 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); | 685 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |

545 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); | 686 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |

546 | 687 | ||

547 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } | 688 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); } |

548 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } | 689 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); } |

549 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } | 690 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); } |

550 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } | 691 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); } |

551 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } | 692 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); } |

552 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } | 693 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); } |

553 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } | 694 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); } |

554 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } | 695 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); } |

696 | |||

697 | #if ECB_CPP | ||

698 | |||

699 | inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } | ||

700 | inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } | ||

701 | inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); } | ||

702 | inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); } | ||

703 | |||

704 | inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); } | ||

705 | inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); } | ||

706 | inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); } | ||

707 | inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); } | ||

708 | |||

709 | inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); } | ||

710 | inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); } | ||

711 | inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); } | ||

712 | inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); } | ||

713 | |||

714 | inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); } | ||

715 | inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); } | ||

716 | inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); } | ||

717 | inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); } | ||

718 | |||

719 | inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); } | ||

720 | inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); } | ||

721 | inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); } | ||

722 | |||

723 | inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); } | ||

724 | inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); } | ||

725 | inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); } | ||

726 | inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); } | ||

727 | |||

728 | inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); } | ||

729 | inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); } | ||

730 | inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); } | ||

731 | inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); } | ||

732 | |||

733 | #endif | ||

555 | 734 | ||

556 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) | 735 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |

557 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) | 736 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |

558 | #define ecb_bswap16(x) __builtin_bswap16 (x) | 737 | #define ecb_bswap16(x) __builtin_bswap16 (x) |

559 | #else | 738 | #else |

… | … | ||

566 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) | 745 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |

567 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) | 746 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |

568 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) | 747 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |

569 | #else | 748 | #else |

570 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); | 749 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |

571 | ecb_function_ ecb_const uint16_t | 750 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x) |

572 | ecb_bswap16 (uint16_t x) | ||

573 | { | 751 | { |

574 | return ecb_rotl16 (x, 8); | 752 | return ecb_rotl16 (x, 8); |

575 | } | 753 | } |

576 | 754 | ||

577 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); | 755 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |

578 | ecb_function_ ecb_const uint32_t | 756 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x) |

579 | ecb_bswap32 (uint32_t x) | ||

580 | { | 757 | { |

581 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); | 758 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |

582 | } | 759 | } |

583 | 760 | ||

584 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); | 761 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |

585 | ecb_function_ ecb_const uint64_t | 762 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x) |

586 | ecb_bswap64 (uint64_t x) | ||

587 | { | 763 | { |

588 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); | 764 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |

589 | } | 765 | } |

590 | #endif | 766 | #endif |

591 | 767 | ||

… | … | ||

598 | #endif | 774 | #endif |

599 | 775 | ||

600 | /* try to tell the compiler that some condition is definitely true */ | 776 | /* try to tell the compiler that some condition is definitely true */ |

601 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 | 777 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |

602 | 778 | ||

603 | ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); | 779 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |

604 | ecb_inline ecb_const unsigned char | 780 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void) |

605 | ecb_byteorder_helper (void) | ||

606 | { | 781 | { |

607 | /* the union code still generates code under pressure in gcc, */ | 782 | /* the union code still generates code under pressure in gcc, */ |

608 | /* but less than using pointers, and always seems to */ | 783 | /* but less than using pointers, and always seems to */ |

609 | /* successfully return a constant. */ | 784 | /* successfully return a constant. */ |

610 | /* the reason why we have this horrible preprocessor mess */ | 785 | /* the reason why we have this horrible preprocessor mess */ |

611 | /* is to avoid it in all cases, at least on common architectures */ | 786 | /* is to avoid it in all cases, at least on common architectures */ |

612 | /* or when using a recent enough gcc version (>= 4.6) */ | 787 | /* or when using a recent enough gcc version (>= 4.6) */ |

613 | #if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64 | ||

614 | return 0x44; | ||

615 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | 788 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |

789 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) | ||

790 | #define ECB_LITTLE_ENDIAN 1 | ||

616 | return 0x44; | 791 | return 0x44332211; |

617 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ | 792 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |

793 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) | ||

794 | #define ECB_BIG_ENDIAN 1 | ||

618 | return 0x11; | 795 | return 0x11223344; |

619 | #else | 796 | #else |

620 | union | 797 | union |

621 | { | 798 | { |

799 | uint8_t c[4]; | ||

622 | uint32_t i; | 800 | uint32_t u; |

623 | uint8_t c; | ||

624 | } u = { 0x11223344 }; | 801 | } u = { 0x11, 0x22, 0x33, 0x44 }; |

625 | return u.c; | 802 | return u.u; |

626 | #endif | 803 | #endif |

627 | } | 804 | } |

628 | 805 | ||

629 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); | 806 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |

630 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } | 807 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |

631 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); | 808 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |

632 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } | 809 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |

810 | |||

811 | /*****************************************************************************/ | ||

812 | /* unaligned load/store */ | ||

813 | |||

814 | ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } | ||

815 | ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } | ||

816 | ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } | ||

817 | |||

818 | ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } | ||

819 | ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } | ||

820 | ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } | ||

821 | |||

822 | ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; } | ||

823 | ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; } | ||

824 | ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; } | ||

825 | |||

826 | ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); } | ||

827 | ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); } | ||

828 | ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); } | ||

829 | |||

830 | ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); } | ||

831 | ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); } | ||

832 | ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); } | ||

833 | |||

834 | ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } | ||

835 | ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } | ||

836 | ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } | ||

837 | |||

838 | ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } | ||

839 | ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } | ||

840 | ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } | ||

841 | |||

842 | ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); } | ||

843 | ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); } | ||

844 | ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } | ||

845 | |||

846 | 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)); } | ||

847 | 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)); } | ||

848 | 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)); } | ||

849 | |||

850 | 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)); } | ||

851 | 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)); } | ||

852 | 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)); } | ||

853 | |||

854 | #if ECB_CPP | ||

855 | |||

856 | inline uint8_t ecb_bswap (uint8_t v) { return v; } | ||

857 | inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); } | ||

858 | inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); } | ||

859 | inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); } | ||

860 | |||

861 | template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } | ||

862 | template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } | ||

863 | template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; } | ||

864 | template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); } | ||

865 | template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); } | ||

866 | template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; } | ||

867 | template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); } | ||

868 | template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); } | ||

869 | |||

870 | template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } | ||

871 | template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } | ||

872 | template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; } | ||

873 | template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); } | ||

874 | template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); } | ||

875 | template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } | ||

876 | template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } | ||

877 | template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } | ||

878 | |||

879 | #endif | ||

880 | |||

881 | /*****************************************************************************/ | ||

882 | /* pointer/integer hashing */ | ||

883 | |||

884 | /* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */ | ||

885 | ecb_function_ uint32_t ecb_mix32 (uint32_t v); | ||

886 | ecb_function_ uint32_t ecb_mix32 (uint32_t v) | ||

887 | { | ||

888 | v ^= v >> 16; v *= 0x7feb352dU; | ||

889 | v ^= v >> 15; v *= 0x846ca68bU; | ||

890 | v ^= v >> 16; | ||

891 | return v; | ||

892 | } | ||

893 | |||

894 | ecb_function_ uint32_t ecb_unmix32 (uint32_t v); | ||

895 | ecb_function_ uint32_t ecb_unmix32 (uint32_t v) | ||

896 | { | ||

897 | v ^= v >> 16 ; v *= 0x43021123U; | ||

898 | v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U; | ||

899 | v ^= v >> 16 ; | ||

900 | return v; | ||

901 | } | ||

902 | |||

903 | /* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */ | ||

904 | ecb_function_ uint64_t ecb_mix64 (uint64_t v); | ||

905 | ecb_function_ uint64_t ecb_mix64 (uint64_t v) | ||

906 | { | ||

907 | v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U; | ||

908 | v ^= v >> 27; v *= 0x94d049bb133111ebU; | ||

909 | v ^= v >> 31; | ||

910 | return v; | ||

911 | } | ||

912 | |||

913 | ecb_function_ uint64_t ecb_unmix64 (uint64_t v); | ||

914 | ecb_function_ uint64_t ecb_unmix64 (uint64_t v) | ||

915 | { | ||

916 | v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U; | ||

917 | v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U; | ||

918 | v ^= v >> 30 ^ v >> 60; | ||

919 | return v; | ||

920 | } | ||

921 | |||

922 | ecb_function_ uintptr_t ecb_ptrmix (void *p); | ||

923 | ecb_function_ uintptr_t ecb_ptrmix (void *p) | ||

924 | { | ||

925 | #if ECB_PTRSIZE <= 4 | ||

926 | return ecb_mix32 ((uint32_t)p); | ||

927 | #else | ||

928 | return ecb_mix64 ((uint64_t)p); | ||

929 | #endif | ||

930 | } | ||

931 | |||

932 | ecb_function_ void *ecb_ptrunmix (uintptr_t v); | ||

933 | ecb_function_ void *ecb_ptrunmix (uintptr_t v) | ||

934 | { | ||

935 | #if ECB_PTRSIZE <= 4 | ||

936 | return (void *)ecb_unmix32 (v); | ||

937 | #else | ||

938 | return (void *)ecb_unmix64 (v); | ||

939 | #endif | ||

940 | } | ||

941 | |||

942 | #if ECB_CPP | ||

943 | |||

944 | template<typename T> | ||

945 | inline uintptr_t ecb_ptrmix (T *p) | ||

946 | { | ||

947 | return ecb_ptrmix (static_cast<void *>(p)); | ||

948 | } | ||

949 | |||

950 | template<typename T> | ||

951 | inline T *ecb_ptrunmix (uintptr_t v) | ||

952 | { | ||

953 | return static_cast<T *>(ecb_ptrunmix (v)); | ||

954 | } | ||

955 | |||

956 | #endif | ||

957 | |||

958 | /*****************************************************************************/ | ||

959 | /* gray code */ | ||

960 | |||

961 | ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); } | ||

962 | ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); } | ||

963 | ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); } | ||

964 | ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); } | ||

965 | |||

966 | ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g); | ||

967 | ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g) | ||

968 | { | ||

969 | g ^= g >> 1; | ||

970 | g ^= g >> 2; | ||

971 | g ^= g >> 4; | ||

972 | |||

973 | return g; | ||

974 | } | ||

975 | |||

976 | ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g); | ||

977 | ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g) | ||

978 | { | ||

979 | g ^= g >> 1; | ||

980 | g ^= g >> 2; | ||

981 | g ^= g >> 4; | ||

982 | g ^= g >> 8; | ||

983 | |||

984 | return g; | ||

985 | } | ||

986 | |||

987 | ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g); | ||

988 | ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g) | ||

989 | { | ||

990 | g ^= g >> 1; | ||

991 | g ^= g >> 2; | ||

992 | g ^= g >> 4; | ||

993 | g ^= g >> 8; | ||

994 | g ^= g >> 16; | ||

995 | |||

996 | return g; | ||

997 | } | ||

998 | |||

999 | ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g); | ||

1000 | ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g) | ||

1001 | { | ||

1002 | g ^= g >> 1; | ||

1003 | g ^= g >> 2; | ||

1004 | g ^= g >> 4; | ||

1005 | g ^= g >> 8; | ||

1006 | g ^= g >> 16; | ||

1007 | g ^= g >> 32; | ||

1008 | |||

1009 | return g; | ||

1010 | } | ||

1011 | |||

1012 | #if ECB_CPP | ||

1013 | |||

1014 | ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); } | ||

1015 | ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); } | ||

1016 | ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); } | ||

1017 | ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); } | ||

1018 | |||

1019 | ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); } | ||

1020 | ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); } | ||

1021 | ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); } | ||

1022 | ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); } | ||

1023 | |||

1024 | #endif | ||

1025 | |||

1026 | /*****************************************************************************/ | ||

1027 | /* 2d hilbert curves */ | ||

1028 | |||

1029 | /* algorithm from the book Hacker's Delight, modified to not */ | ||

1030 | /* run into undefined behaviour for n==16 */ | ||

1031 | static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s); | ||

1032 | static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s) | ||

1033 | { | ||

1034 | uint32_t comp, swap, cs, t, sr; | ||

1035 | |||

1036 | /* pad s on the left (unused) bits with 01 (no change groups) */ | ||

1037 | s |= 0x55555555U << n << n; | ||

1038 | /* "s shift right" */ | ||

1039 | sr = (s >> 1) & 0x55555555U; | ||

1040 | /* compute complement and swap info in two-bit groups */ | ||

1041 | cs = ((s & 0x55555555U) + sr) ^ 0x55555555U; | ||

1042 | |||

1043 | /* parallel prefix xor op to propagate both complement | ||

1044 | * and swap info together from left to right (there is | ||

1045 | * no step "cs ^= cs >> 1", so in effect it computes | ||

1046 | * two independent parallel prefix operations on two | ||

1047 | * interleaved sets of sixteen bits). | ||

1048 | */ | ||

1049 | cs ^= cs >> 2; | ||

1050 | cs ^= cs >> 4; | ||

1051 | cs ^= cs >> 8; | ||

1052 | cs ^= cs >> 16; | ||

1053 | |||

1054 | /* separate swap and complement bits */ | ||

1055 | swap = cs & 0x55555555U; | ||

1056 | comp = (cs >> 1) & 0x55555555U; | ||

1057 | |||

1058 | /* calculate coordinates in odd and even bit positions */ | ||

1059 | t = (s & swap) ^ comp; | ||

1060 | s = s ^ sr ^ t ^ (t << 1); | ||

1061 | |||

1062 | /* unpad/clear out any junk on the left */ | ||

1063 | s = s & ((1 << n << n) - 1); | ||

1064 | |||

1065 | /* Now "unshuffle" to separate the x and y bits. */ | ||

1066 | t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1); | ||

1067 | t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2); | ||

1068 | t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4); | ||

1069 | t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8); | ||

1070 | |||

1071 | /* now s contains two 16-bit coordinates */ | ||

1072 | return s; | ||

1073 | } | ||

1074 | |||

1075 | /* 64 bit, a straightforward extension to the 32 bit case */ | ||

1076 | static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s); | ||

1077 | static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s) | ||

1078 | { | ||

1079 | uint64_t comp, swap, cs, t, sr; | ||

1080 | |||

1081 | /* pad s on the left (unused) bits with 01 (no change groups) */ | ||

1082 | s |= 0x5555555555555555U << n << n; | ||

1083 | /* "s shift right" */ | ||

1084 | sr = (s >> 1) & 0x5555555555555555U; | ||

1085 | /* compute complement and swap info in two-bit groups */ | ||

1086 | cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U; | ||

1087 | |||

1088 | /* parallel prefix xor op to propagate both complement | ||

1089 | * and swap info together from left to right (there is | ||

1090 | * no step "cs ^= cs >> 1", so in effect it computes | ||

1091 | * two independent parallel prefix operations on two | ||

1092 | * interleaved sets of thirty-two bits). | ||

1093 | */ | ||

1094 | cs ^= cs >> 2; | ||

1095 | cs ^= cs >> 4; | ||

1096 | cs ^= cs >> 8; | ||

1097 | cs ^= cs >> 16; | ||

1098 | cs ^= cs >> 32; | ||

1099 | |||

1100 | /* separate swap and complement bits */ | ||

1101 | swap = cs & 0x5555555555555555U; | ||

1102 | comp = (cs >> 1) & 0x5555555555555555U; | ||

1103 | |||

1104 | /* calculate coordinates in odd and even bit positions */ | ||

1105 | t = (s & swap) ^ comp; | ||

1106 | s = s ^ sr ^ t ^ (t << 1); | ||

1107 | |||

1108 | /* unpad/clear out any junk on the left */ | ||

1109 | s = s & ((1 << n << n) - 1); | ||

1110 | |||

1111 | /* Now "unshuffle" to separate the x and y bits. */ | ||

1112 | t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1); | ||

1113 | t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2); | ||

1114 | t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4); | ||

1115 | t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8); | ||

1116 | t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16); | ||

1117 | |||

1118 | /* now s contains two 32-bit coordinates */ | ||

1119 | return s; | ||

1120 | } | ||

1121 | |||

1122 | /* algorithm from the book Hacker's Delight, but a similar algorithm*/ | ||

1123 | /* is given in https://doi.org/10.1002/spe.4380160103 */ | ||

1124 | /* this has been slightly improved over the original version */ | ||

1125 | ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy); | ||

1126 | ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy) | ||

1127 | { | ||

1128 | uint32_t row; | ||

1129 | uint32_t state = 0; | ||

1130 | uint32_t s = 0; | ||

1131 | |||

1132 | do | ||

1133 | { | ||

1134 | --n; | ||

1135 | |||

1136 | row = 4 * state | ||

1137 | | (2 & (xy >> n >> 15)) | ||

1138 | | (1 & (xy >> n )); | ||

1139 | |||

1140 | /* these funky constants are lookup tables for two-bit values */ | ||

1141 | s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; | ||

1142 | state = (0x8fe65831U >> 2 * row) & 3; | ||

1143 | } | ||

1144 | while (n > 0); | ||

1145 | |||

1146 | return s; | ||

1147 | } | ||

1148 | |||

1149 | /* 64 bit, essentially the same as 32 bit */ | ||

1150 | ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy); | ||

1151 | ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy) | ||

1152 | { | ||

1153 | uint32_t row; | ||

1154 | uint32_t state = 0; | ||

1155 | uint64_t s = 0; | ||

1156 | |||

1157 | do | ||

1158 | { | ||

1159 | --n; | ||

1160 | |||

1161 | row = 4 * state | ||

1162 | | (2 & (xy >> n >> 31)) | ||

1163 | | (1 & (xy >> n )); | ||

1164 | |||

1165 | /* these funky constants are lookup tables for two-bit values */ | ||

1166 | s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; | ||

1167 | state = (0x8fe65831U >> 2 * row) & 3; | ||

1168 | } | ||

1169 | while (n > 0); | ||

1170 | |||

1171 | return s; | ||

1172 | } | ||

1173 | |||

1174 | /*****************************************************************************/ | ||

1175 | /* division */ | ||

633 | 1176 | ||

634 | #if ECB_GCC_VERSION(3,0) || ECB_C99 | 1177 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |

1178 | /* C99 tightened the definition of %, so we can use a more efficient version */ | ||

635 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) | 1179 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |

636 | #else | 1180 | #else |

637 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) | 1181 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |

638 | #endif | 1182 | #endif |

639 | 1183 | ||

… | … | ||

650 | } | 1194 | } |

651 | #else | 1195 | #else |

652 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) | 1196 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |

653 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) | 1197 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |

654 | #endif | 1198 | #endif |

1199 | |||

1200 | /*****************************************************************************/ | ||

1201 | /* array length */ | ||

655 | 1202 | ||

656 | #if ecb_cplusplus_does_not_suck | 1203 | #if ecb_cplusplus_does_not_suck |

657 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ | 1204 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |

658 | template<typename T, int N> | 1205 | template<typename T, int N> |

659 | static inline int ecb_array_length (const T (&arr)[N]) | 1206 | static inline int ecb_array_length (const T (&arr)[N]) |

… | … | ||

661 | return N; | 1208 | return N; |

662 | } | 1209 | } |

663 | #else | 1210 | #else |

664 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) | 1211 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |

665 | #endif | 1212 | #endif |

1213 | |||

1214 | /*****************************************************************************/ | ||

1215 | /* IEEE 754-2008 half float conversions */ | ||

1216 | |||

1217 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); | ||

1218 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x) | ||

1219 | { | ||

1220 | unsigned int s = (x & 0x8000) << (31 - 15); | ||

1221 | int e = (x >> 10) & 0x001f; | ||

1222 | unsigned int m = x & 0x03ff; | ||

1223 | |||

1224 | if (ecb_expect_false (e == 31)) | ||

1225 | /* infinity or NaN */ | ||

1226 | e = 255 - (127 - 15); | ||

1227 | else if (ecb_expect_false (!e)) | ||

1228 | { | ||

1229 | if (ecb_expect_true (!m)) | ||

1230 | /* zero, handled by code below by forcing e to 0 */ | ||

1231 | e = 0 - (127 - 15); | ||

1232 | else | ||

1233 | { | ||

1234 | /* subnormal, renormalise */ | ||

1235 | unsigned int s = 10 - ecb_ld32 (m); | ||

1236 | |||

1237 | m = (m << s) & 0x3ff; /* mask implicit bit */ | ||

1238 | e -= s - 1; | ||

1239 | } | ||

1240 | } | ||

1241 | |||

1242 | /* e and m now are normalised, or zero, (or inf or nan) */ | ||

1243 | e += 127 - 15; | ||

1244 | |||

1245 | return s | (e << 23) | (m << (23 - 10)); | ||

1246 | } | ||

1247 | |||

1248 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); | ||

1249 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x) | ||

1250 | { | ||

1251 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ | ||

1252 | int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ | ||

1253 | unsigned int m = x & 0x007fffff; | ||

1254 | |||

1255 | x &= 0x7fffffff; | ||

1256 | |||

1257 | /* if it's within range of binary16 normals, use fast path */ | ||

1258 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) | ||

1259 | { | ||

1260 | /* mantissa round-to-even */ | ||

1261 | m += 0x00000fff + ((m >> (23 - 10)) & 1); | ||

1262 | |||

1263 | /* handle overflow */ | ||

1264 | if (ecb_expect_false (m >= 0x00800000)) | ||

1265 | { | ||

1266 | m >>= 1; | ||

1267 | e += 1; | ||

1268 | } | ||

1269 | |||

1270 | return s | (e << 10) | (m >> (23 - 10)); | ||

1271 | } | ||

1272 | |||

1273 | /* handle large numbers and infinity */ | ||

1274 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) | ||

1275 | return s | 0x7c00; | ||

1276 | |||

1277 | /* handle zero, subnormals and small numbers */ | ||

1278 | if (ecb_expect_true (x < 0x38800000)) | ||

1279 | { | ||

1280 | /* zero */ | ||

1281 | if (ecb_expect_true (!x)) | ||

1282 | return s; | ||

1283 | |||

1284 | /* handle subnormals */ | ||

1285 | |||

1286 | /* too small, will be zero */ | ||

1287 | if (e < (14 - 24)) /* might not be sharp, but is good enough */ | ||

1288 | return s; | ||

1289 | |||

1290 | m |= 0x00800000; /* make implicit bit explicit */ | ||

1291 | |||

1292 | /* very tricky - we need to round to the nearest e (+10) bit value */ | ||

1293 | { | ||

1294 | unsigned int bits = 14 - e; | ||

1295 | unsigned int half = (1 << (bits - 1)) - 1; | ||

1296 | unsigned int even = (m >> bits) & 1; | ||

1297 | |||

1298 | /* if this overflows, we will end up with a normalised number */ | ||

1299 | m = (m + half + even) >> bits; | ||

1300 | } | ||

1301 | |||

1302 | return s | m; | ||

1303 | } | ||

1304 | |||

1305 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ | ||

1306 | m >>= 13; | ||

1307 | |||

1308 | return s | 0x7c00 | m | !m; | ||

1309 | } | ||

1310 | |||

1311 | /*******************************************************************************/ | ||

1312 | /* fast integer to ascii */ | ||

1313 | |||

1314 | /* | ||

1315 | * This code is pretty complicated because it is general. The idea behind it, | ||

1316 | * however, is pretty simple: first, the number is multiplied with a scaling | ||

1317 | * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point | ||

1318 | * number with the first digit in the upper bits. | ||

1319 | * Then this digit is converted to text and masked out. The resulting number | ||

1320 | * is then multiplied by 10, by multiplying the fixed point representation | ||

1321 | * by 5 and shifting the (binary) decimal point one to the right, so a 4.28 | ||

1322 | * format becomes 5.27, 6.26 and so on. | ||

1323 | * The rest involves only advancing the pointer if we already generated a | ||

1324 | * non-zero digit, so leading zeroes are overwritten. | ||

1325 | */ | ||

1326 | |||

1327 | /* simply return a mask with "bits" bits set */ | ||

1328 | #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) | ||

1329 | |||

1330 | /* oputput a single digit. maskvalue is 10**digitidx */ | ||

1331 | #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ | ||

1332 | if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ | ||

1333 | { \ | ||

1334 | char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ | ||

1335 | *ptr = digit + '0'; /* output it */ \ | ||

1336 | nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ | ||

1337 | ptr += nz; /* output digit only if non-zero digit seen */ \ | ||

1338 | x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \ | ||

1339 | } | ||

1340 | |||

1341 | /* convert integer to fixed point format and multiply out digits, highest first */ | ||

1342 | /* requires magic constants: max. digits and number of bits after the decimal point */ | ||

1343 | #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \ | ||

1344 | ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ | ||

1345 | { \ | ||

1346 | char nz = lz; /* non-zero digit seen? */ \ | ||

1347 | /* convert to x.bits fixed-point */ \ | ||

1348 | type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \ | ||

1349 | /* output up to 10 digits */ \ | ||

1350 | ecb_i2a_digit (type,bits,digitmask, 1, 0); \ | ||

1351 | ecb_i2a_digit (type,bits,digitmask, 10, 1); \ | ||

1352 | ecb_i2a_digit (type,bits,digitmask, 100, 2); \ | ||

1353 | ecb_i2a_digit (type,bits,digitmask, 1000, 3); \ | ||

1354 | ecb_i2a_digit (type,bits,digitmask, 10000, 4); \ | ||

1355 | ecb_i2a_digit (type,bits,digitmask, 100000, 5); \ | ||

1356 | ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \ | ||

1357 | ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \ | ||

1358 | ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ | ||

1359 | ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ | ||

1360 | return ptr; \ | ||

1361 | } | ||

1362 | |||

1363 | /* predefined versions of the above, for various digits */ | ||

1364 | /* ecb_i2a_xN = almost N digits, limit defined by macro */ | ||

1365 | /* ecb_i2a_N = up to N digits, leading zeroes suppressed */ | ||

1366 | /* ecb_i2a_0N = exactly N digits, including leading zeroes */ | ||

1367 | |||

1368 | /* non-leading-zero versions, limited range */ | ||

1369 | #define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */ | ||

1370 | #define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */ | ||

1371 | ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) | ||

1372 | ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) | ||

1373 | |||

1374 | /* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */ | ||

1375 | ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) | ||

1376 | ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) | ||

1377 | ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) | ||

1378 | ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) | ||

1379 | ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) | ||

1380 | ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) | ||

1381 | ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) | ||

1382 | ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) | ||

1383 | |||

1384 | /* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */ | ||

1385 | ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) | ||

1386 | ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) | ||

1387 | ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) | ||

1388 | ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) | ||

1389 | ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) | ||

1390 | ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1) | ||

1391 | ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1) | ||

1392 | ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1) | ||

1393 | |||

1394 | #define ECB_I2A_I32_DIGITS 11 | ||

1395 | #define ECB_I2A_U32_DIGITS 10 | ||

1396 | #define ECB_I2A_I64_DIGITS 20 | ||

1397 | #define ECB_I2A_U64_DIGITS 21 | ||

1398 | #define ECB_I2A_MAX_DIGITS 21 | ||

1399 | |||

1400 | ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u); | ||

1401 | ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u) | ||

1402 | { | ||

1403 | #if ECB_64BIT_NATIVE | ||

1404 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) | ||

1405 | ptr = ecb_i2a_x10 (ptr, u); | ||

1406 | else /* x10 almost, but not fully, covers 32 bit */ | ||

1407 | { | ||

1408 | uint32_t u1 = u % 1000000000; | ||

1409 | uint32_t u2 = u / 1000000000; | ||

1410 | |||

1411 | *ptr++ = u2 + '0'; | ||

1412 | ptr = ecb_i2a_09 (ptr, u1); | ||

1413 | } | ||

1414 | #else | ||

1415 | if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) | ||

1416 | ecb_i2a_x5 (ptr, u); | ||

1417 | else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000)) | ||

1418 | { | ||

1419 | uint32_t u1 = u % 10000; | ||

1420 | uint32_t u2 = u / 10000; | ||

1421 | |||

1422 | ptr = ecb_i2a_x5 (ptr, u2); | ||

1423 | ptr = ecb_i2a_04 (ptr, u1); | ||

1424 | } | ||

1425 | else | ||

1426 | { | ||

1427 | uint32_t u1 = u % 10000; | ||

1428 | uint32_t ua = u / 10000; | ||

1429 | uint32_t u2 = ua % 10000; | ||

1430 | uint32_t u3 = ua / 10000; | ||

1431 | |||

1432 | ptr = ecb_i2a_2 (ptr, u3); | ||

1433 | ptr = ecb_i2a_04 (ptr, u2); | ||

1434 | ptr = ecb_i2a_04 (ptr, u1); | ||

1435 | } | ||

1436 | #endif | ||

1437 | |||

1438 | return ptr; | ||

1439 | } | ||

1440 | |||

1441 | ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v); | ||

1442 | ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v) | ||

1443 | { | ||

1444 | *ptr = '-'; ptr += v < 0; | ||

1445 | uint32_t u = v < 0 ? -(uint32_t)v : v; | ||

1446 | |||

1447 | #if ECB_64BIT_NATIVE | ||

1448 | ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */ | ||

1449 | #else | ||

1450 | ptr = ecb_i2a_u32 (ptr, u); | ||

1451 | #endif | ||

1452 | |||

1453 | return ptr; | ||

1454 | } | ||

1455 | |||

1456 | ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u); | ||

1457 | ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u) | ||

1458 | { | ||

1459 | #if ECB_64BIT_NATIVE | ||

1460 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) | ||

1461 | ptr = ecb_i2a_x10 (ptr, u); | ||

1462 | else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) | ||

1463 | { | ||

1464 | uint64_t u1 = u % 1000000000; | ||

1465 | uint64_t u2 = u / 1000000000; | ||

1466 | |||

1467 | ptr = ecb_i2a_x10 (ptr, u2); | ||

1468 | ptr = ecb_i2a_09 (ptr, u1); | ||

1469 | } | ||

1470 | else | ||

1471 | { | ||

1472 | uint64_t u1 = u % 1000000000; | ||

1473 | uint64_t ua = u / 1000000000; | ||

1474 | uint64_t u2 = ua % 1000000000; | ||

1475 | uint64_t u3 = ua / 1000000000; | ||

1476 | |||

1477 | ptr = ecb_i2a_2 (ptr, u3); | ||

1478 | ptr = ecb_i2a_09 (ptr, u2); | ||

1479 | ptr = ecb_i2a_09 (ptr, u1); | ||

1480 | } | ||

1481 | #else | ||

1482 | if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) | ||

1483 | ptr = ecb_i2a_x5 (ptr, u); | ||

1484 | else | ||

1485 | { | ||

1486 | uint64_t u1 = u % 10000; | ||

1487 | uint64_t u2 = u / 10000; | ||

1488 | |||

1489 | ptr = ecb_i2a_u64 (ptr, u2); | ||

1490 | ptr = ecb_i2a_04 (ptr, u1); | ||

1491 | } | ||

1492 | #endif | ||

1493 | |||

1494 | return ptr; | ||

1495 | } | ||

1496 | |||

1497 | ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v); | ||

1498 | ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v) | ||

1499 | { | ||

1500 | *ptr = '-'; ptr += v < 0; | ||

1501 | uint64_t u = v < 0 ? -(uint64_t)v : v; | ||

1502 | |||

1503 | #if ECB_64BIT_NATIVE | ||

1504 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) | ||

1505 | ptr = ecb_i2a_x10 (ptr, u); | ||

1506 | else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) | ||

1507 | { | ||

1508 | uint64_t u1 = u % 1000000000; | ||

1509 | uint64_t u2 = u / 1000000000; | ||

1510 | |||

1511 | ptr = ecb_i2a_x10 (ptr, u2); | ||

1512 | ptr = ecb_i2a_09 (ptr, u1); | ||

1513 | } | ||

1514 | else | ||

1515 | { | ||

1516 | uint64_t u1 = u % 1000000000; | ||

1517 | uint64_t ua = u / 1000000000; | ||

1518 | uint64_t u2 = ua % 1000000000; | ||

1519 | uint64_t u3 = ua / 1000000000; | ||

1520 | |||

1521 | /* 2**31 is 19 digits, so the top is exactly one digit */ | ||

1522 | *ptr++ = u3 + '0'; | ||

1523 | ptr = ecb_i2a_09 (ptr, u2); | ||

1524 | ptr = ecb_i2a_09 (ptr, u1); | ||

1525 | } | ||

1526 | #else | ||

1527 | ptr = ecb_i2a_u64 (ptr, u); | ||

1528 | #endif | ||

1529 | |||

1530 | return ptr; | ||

1531 | } | ||

666 | 1532 | ||

667 | /*******************************************************************************/ | 1533 | /*******************************************************************************/ |

668 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ | 1534 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |

669 | 1535 | ||

670 | /* basically, everything uses "ieee pure-endian" floating point numbers */ | 1536 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |

… | … | ||

683 | || defined __sh__ \ | 1549 | || defined __sh__ \ |

684 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ | 1550 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |

685 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ | 1551 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |

686 | || defined __aarch64__ | 1552 | || defined __aarch64__ |

687 | #define ECB_STDFP 1 | 1553 | #define ECB_STDFP 1 |

688 | #include <string.h> /* for memcpy */ | ||

689 | #else | 1554 | #else |

690 | #define ECB_STDFP 0 | 1555 | #define ECB_STDFP 0 |

691 | #endif | 1556 | #endif |

692 | 1557 | ||

693 | #ifndef ECB_NO_LIBM | 1558 | #ifndef ECB_NO_LIBM |

… | … | ||

713 | #else | 1578 | #else |

714 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) | 1579 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |

715 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) | 1580 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |

716 | #endif | 1581 | #endif |

717 | 1582 | ||

718 | /* converts an ieee half/binary16 to a float */ | ||

719 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); | ||

720 | ecb_function_ ecb_const float | ||

721 | ecb_binary16_to_float (uint16_t x) | ||

722 | { | ||

723 | int e = (x >> 10) & 0x1f; | ||

724 | int m = x & 0x3ff; | ||

725 | float r; | ||

726 | |||

727 | if (!e ) r = ecb_ldexpf (m , -24); | ||

728 | else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25); | ||

729 | else if (m ) r = ECB_NAN; | ||

730 | else r = ECB_INFINITY; | ||

731 | |||

732 | return x & 0x8000 ? -r : r; | ||

733 | } | ||

734 | |||

735 | /* convert a float to ieee single/binary32 */ | 1583 | /* convert a float to ieee single/binary32 */ |

736 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); | 1584 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |

737 | ecb_function_ ecb_const uint32_t | 1585 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x) |

738 | ecb_float_to_binary32 (float x) | ||

739 | { | 1586 | { |

740 | uint32_t r; | 1587 | uint32_t r; |

741 | 1588 | ||

742 | #if ECB_STDFP | 1589 | #if ECB_STDFP |

743 | memcpy (&r, &x, 4); | 1590 | memcpy (&r, &x, 4); |

… | … | ||

772 | return r; | 1619 | return r; |

773 | } | 1620 | } |

774 | 1621 | ||

775 | /* converts an ieee single/binary32 to a float */ | 1622 | /* converts an ieee single/binary32 to a float */ |

776 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); | 1623 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |

777 | ecb_function_ ecb_const float | 1624 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x) |

778 | ecb_binary32_to_float (uint32_t x) | ||

779 | { | 1625 | { |

780 | float r; | 1626 | float r; |

781 | 1627 | ||

782 | #if ECB_STDFP | 1628 | #if ECB_STDFP |

783 | memcpy (&r, &x, 4); | 1629 | memcpy (&r, &x, 4); |

… | … | ||

802 | return r; | 1648 | return r; |

803 | } | 1649 | } |

804 | 1650 | ||

805 | /* convert a double to ieee double/binary64 */ | 1651 | /* convert a double to ieee double/binary64 */ |

806 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); | 1652 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |

807 | ecb_function_ ecb_const uint64_t | 1653 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x) |

808 | ecb_double_to_binary64 (double x) | ||

809 | { | 1654 | { |

810 | uint64_t r; | 1655 | uint64_t r; |

811 | 1656 | ||

812 | #if ECB_STDFP | 1657 | #if ECB_STDFP |

813 | memcpy (&r, &x, 8); | 1658 | memcpy (&r, &x, 8); |

… | … | ||

842 | return r; | 1687 | return r; |

843 | } | 1688 | } |

844 | 1689 | ||

845 | /* converts an ieee double/binary64 to a double */ | 1690 | /* converts an ieee double/binary64 to a double */ |

846 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); | 1691 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |

847 | ecb_function_ ecb_const double | 1692 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x) |

848 | ecb_binary64_to_double (uint64_t x) | ||

849 | { | 1693 | { |

850 | double r; | 1694 | double r; |

851 | 1695 | ||

852 | #if ECB_STDFP | 1696 | #if ECB_STDFP |

853 | memcpy (&r, &x, 8); | 1697 | memcpy (&r, &x, 8); |

… | … | ||

870 | #endif | 1714 | #endif |

871 | 1715 | ||

872 | return r; | 1716 | return r; |

873 | } | 1717 | } |

874 | 1718 | ||

875 | #endif | 1719 | /* convert a float to ieee half/binary16 */ |

1720 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); | ||

1721 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x) | ||

1722 | { | ||

1723 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); | ||

1724 | } | ||

876 | 1725 | ||

877 | #endif | 1726 | /* convert an ieee half/binary16 to float */ |

1727 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); | ||

1728 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x) | ||

1729 | { | ||

1730 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); | ||

1731 | } | ||

878 | 1732 | ||

1733 | #endif | ||

1734 | |||

1735 | #endif | ||

1736 |

– |
Removed lines |

+ |
Added lines |

< |
Changed lines |

> |
Changed lines |