ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
(Generate patch)

Comparing libecb/ecb.h (file contents):
Revision 1.148 by root, Fri Oct 24 04:50:13 2014 UTC vs.
Revision 1.186 by root, Tue Jan 21 23:51:21 2020 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2014 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2015,2018-2020 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 0x00010008
46
47#include <string.h> /* for memcpy */
46 48
47#ifdef _WIN32 49#ifdef _WIN32
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
89#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
90#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
91
92#ifndef ECB_OPTIMIZE_SIZE
93 #if __OPTIMIZE_SIZE__
94 #define ECB_OPTIMIZE_SIZE 1
95 #else
96 #define ECB_OPTIMIZE_SIZE 0
97 #endif
98#endif
99
79/* work around x32 idiocy by defining proper macros */ 100/* work around x32 idiocy by defining proper macros */
80#if __amd64 || __x86_64 || _M_AMD64 || _M_X64 101#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
81 #if _ILP32 102 #if _ILP32
82 #define ECB_AMD64_X32 1 103 #define ECB_AMD64_X32 1
83 #else 104 #else
84 #define ECB_AMD64 1 105 #define ECB_AMD64 1
85 #endif 106 #endif
… …
112 #define ECB_CLANG_EXTENSION(x) 0 133 #define ECB_CLANG_EXTENSION(x) 0
113#endif 134#endif
114 135
115#define ECB_CPP (__cplusplus+0) 136#define ECB_CPP (__cplusplus+0)
116#define ECB_CPP11 (__cplusplus >= 201103L) 137#define ECB_CPP11 (__cplusplus >= 201103L)
138#define ECB_CPP14 (__cplusplus >= 201402L)
139#define ECB_CPP17 (__cplusplus >= 201703L)
117 140
118#if ECB_CPP 141#if ECB_CPP
119 #define ECB_C 0 142 #define ECB_C 0
120 #define ECB_STDC_VERSION 0 143 #define ECB_STDC_VERSION 0
121#else 144#else
… …
123 #define ECB_STDC_VERSION __STDC_VERSION__ 146 #define ECB_STDC_VERSION __STDC_VERSION__
124#endif 147#endif
125 148
126#define ECB_C99 (ECB_STDC_VERSION >= 199901L) 149#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
127#define ECB_C11 (ECB_STDC_VERSION >= 201112L) 150#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
151#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
128 152
129#if ECB_CPP 153#if ECB_CPP
130 #define ECB_EXTERN_C extern "C" 154 #define ECB_EXTERN_C extern "C"
131 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 155 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
132 #define ECB_EXTERN_C_END } 156 #define ECB_EXTERN_C_END }
… …
147 171
148#if ECB_NO_SMP 172#if ECB_NO_SMP
149 #define ECB_MEMORY_FENCE do { } while (0) 173 #define ECB_MEMORY_FENCE do { } while (0)
150#endif 174#endif
151 175
176/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
177#if __xlC__ && ECB_CPP
178 #include <builtins.h>
179#endif
180
181#if 1400 <= _MSC_VER
182 #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
183#endif
184
152#ifndef ECB_MEMORY_FENCE 185#ifndef ECB_MEMORY_FENCE
153 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 186 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
187 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
154 #if __i386 || __i386__ 188 #if __i386 || __i386__
155 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 189 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
156 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 190 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
157 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 191 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
158 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 192 #elif ECB_GCC_AMD64
159 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 193 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
160 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 194 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
161 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 195 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
162 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 196 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
163 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 197 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
198 #elif defined __ARM_ARCH_2__ \
199 || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
200 || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
201 || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
202 || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
203 || defined __ARM_ARCH_5TEJ__
204 /* should not need any, unless running old code on newer cpu - arm doesn't support that */
164 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 205 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
165 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 206 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
207 || defined __ARM_ARCH_6T2__
166 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 208 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
167 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 209 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
168 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 210 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
169 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 211 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
170 #elif __aarch64__ 212 #elif __aarch64__
171 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") 213 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
172 #elif (__sparc || __sparc__) && !__sparcv8 214 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
173 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") 215 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
174 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 216 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
175 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 217 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
176 #elif defined __s390__ || defined __s390x__ 218 #elif defined __s390__ || defined __s390x__
177 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
… …
200 #if ECB_GCC_VERSION(4,7) 242 #if ECB_GCC_VERSION(4,7)
201 /* see comment below (stdatomic.h) about the C11 memory model. */ 243 /* see comment below (stdatomic.h) about the C11 memory model. */
202 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 244 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
203 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) 245 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
204 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) 246 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
247 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
205 248
206 #elif ECB_CLANG_EXTENSION(c_atomic) 249 #elif ECB_CLANG_EXTENSION(c_atomic)
207 /* see comment below (stdatomic.h) about the C11 memory model. */ 250 /* see comment below (stdatomic.h) about the C11 memory model. */
208 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 251 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
209 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) 252 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
210 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) 253 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
254 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
211 255
212 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 256 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
213 #define ECB_MEMORY_FENCE __sync_synchronize () 257 #define ECB_MEMORY_FENCE __sync_synchronize ()
214 #elif _MSC_VER >= 1500 /* VC++ 2008 */ 258 #elif _MSC_VER >= 1500 /* VC++ 2008 */
215 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ 259 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
… …
225 #elif defined _WIN32 269 #elif defined _WIN32
226 #include <WinNT.h> 270 #include <WinNT.h>
227 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 271 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
228 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 272 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
229 #include <mbarrier.h> 273 #include <mbarrier.h>
230 #define ECB_MEMORY_FENCE __machine_rw_barrier () 274 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
231 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 275 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
232 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () 276 #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
277 #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
233 #elif __xlC__ 278 #elif __xlC__
234 #define ECB_MEMORY_FENCE __sync () 279 #define ECB_MEMORY_FENCE __sync ()
235 #endif 280 #endif
236#endif 281#endif
237 282
238#ifndef ECB_MEMORY_FENCE 283#ifndef ECB_MEMORY_FENCE
239 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 284 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
240 /* we assume that these memory fences work on all variables/all memory accesses, */ 285 /* we assume that these memory fences work on all variables/all memory accesses, */
241 /* not just C11 atomics and atomic accesses */ 286 /* not just C11 atomics and atomic accesses */
242 #include <stdatomic.h> 287 #include <stdatomic.h>
243 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
244 /* any fence other than seq_cst, which isn't very efficient for us. */
245 /* Why that is, we don't know - either the C11 memory model is quite useless */
246 /* for most usages, or gcc and clang have a bug */
247 /* I *currently* lean towards the latter, and inefficiently implement */
248 /* all three of ecb's fences as a seq_cst fence */
249 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
250 /* for all __atomic_thread_fence's except seq_cst */
251 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) 288 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
289 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
290 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
252 #endif 291 #endif
253#endif 292#endif
254 293
255#ifndef ECB_MEMORY_FENCE 294#ifndef ECB_MEMORY_FENCE
256 #if !ECB_AVOID_PTHREADS 295 #if !ECB_AVOID_PTHREADS
… …
276 315
277#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 316#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
278 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 317 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
279#endif 318#endif
280 319
320#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
321 #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
322#endif
323
281/*****************************************************************************/ 324/*****************************************************************************/
282 325
283#if __cplusplus 326#if ECB_CPP
284 #define ecb_inline static inline 327 #define ecb_inline static inline
285#elif ECB_GCC_VERSION(2,5) 328#elif ECB_GCC_VERSION(2,5)
286 #define ecb_inline static __inline__ 329 #define ecb_inline static __inline__
287#elif ECB_C99 330#elif ECB_C99
288 #define ecb_inline static inline 331 #define ecb_inline static inline
… …
302 345
303#define ECB_CONCAT_(a, b) a ## b 346#define ECB_CONCAT_(a, b) a ## b
304#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 347#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
305#define ECB_STRINGIFY_(a) # a 348#define ECB_STRINGIFY_(a) # a
306#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 349#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
350#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
307 351
308#define ecb_function_ ecb_inline 352#define ecb_function_ ecb_inline
309 353
310#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) 354#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
311 #define ecb_attribute(attrlist) __attribute__ (attrlist) 355 #define ecb_attribute(attrlist) __attribute__ (attrlist)
… …
343#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) 387#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
344 #define ecb_decltype(x) __typeof__ (x) 388 #define ecb_decltype(x) __typeof__ (x)
345#endif 389#endif
346 390
347#if _MSC_VER >= 1300 391#if _MSC_VER >= 1300
348 #define ecb_deprecated __declspec(deprecated) 392 #define ecb_deprecated __declspec (deprecated)
349#else 393#else
350 #define ecb_deprecated ecb_attribute ((__deprecated__)) 394 #define ecb_deprecated ecb_attribute ((__deprecated__))
351#endif 395#endif
352 396
397#if _MSC_VER >= 1500
398 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
399#elif ECB_GCC_VERSION(4,5)
400 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
401#else
402 #define ecb_deprecated_message(msg) ecb_deprecated
403#endif
404
405#if _MSC_VER >= 1400
406 #define ecb_noinline __declspec (noinline)
407#else
353#define ecb_noinline ecb_attribute ((__noinline__)) 408 #define ecb_noinline ecb_attribute ((__noinline__))
409#endif
410
354#define ecb_unused ecb_attribute ((__unused__)) 411#define ecb_unused ecb_attribute ((__unused__))
355#define ecb_const ecb_attribute ((__const__)) 412#define ecb_const ecb_attribute ((__const__))
356#define ecb_pure ecb_attribute ((__pure__)) 413#define ecb_pure ecb_attribute ((__pure__))
357 414
358/* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
359#if ECB_C11 || __IBMC_NORETURN 415#if ECB_C11 || __IBMC_NORETURN
360 /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */ 416 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
361 #define ecb_noreturn _Noreturn 417 #define ecb_noreturn _Noreturn
418#elif ECB_CPP11
419 #define ecb_noreturn [[noreturn]]
420#elif _MSC_VER >= 1200
421 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
422 #define ecb_noreturn __declspec (noreturn)
362#else 423#else
363 #define ecb_noreturn ecb_attribute ((__noreturn__)) 424 #define ecb_noreturn ecb_attribute ((__noreturn__))
364#endif 425#endif
365 426
366#if ECB_GCC_VERSION(4,3) 427#if ECB_GCC_VERSION(4,3)
… …
393 #define ecb_ctz32(x) __builtin_ctz (x) 454 #define ecb_ctz32(x) __builtin_ctz (x)
394 #define ecb_ctz64(x) __builtin_ctzll (x) 455 #define ecb_ctz64(x) __builtin_ctzll (x)
395 #define ecb_popcount32(x) __builtin_popcount (x) 456 #define ecb_popcount32(x) __builtin_popcount (x)
396 /* no popcountll */ 457 /* no popcountll */
397#else 458#else
398 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 459 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
399 ecb_function_ int 460 ecb_function_ ecb_const int
400 ecb_ctz32 (uint32_t x) 461 ecb_ctz32 (uint32_t x)
401 { 462 {
463#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
464 unsigned long r;
465 _BitScanForward (&r, x);
466 return (int)r;
467#else
402 int r = 0; 468 int r = 0;
403 469
404 x &= ~x + 1; /* this isolates the lowest bit */ 470 x &= ~x + 1; /* this isolates the lowest bit */
405 471
406#if ECB_branchless_on_i386 472#if ECB_branchless_on_i386
… …
416 if (x & 0xff00ff00) r += 8; 482 if (x & 0xff00ff00) r += 8;
417 if (x & 0xffff0000) r += 16; 483 if (x & 0xffff0000) r += 16;
418#endif 484#endif
419 485
420 return r; 486 return r;
487#endif
421 } 488 }
422 489
423 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 490 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
424 ecb_function_ int 491 ecb_function_ ecb_const int
425 ecb_ctz64 (uint64_t x) 492 ecb_ctz64 (uint64_t x)
426 { 493 {
494#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
495 unsigned long r;
496 _BitScanForward64 (&r, x);
497 return (int)r;
498#else
427 int shift = x & 0xffffffffU ? 0 : 32; 499 int shift = x & 0xffffffff ? 0 : 32;
428 return ecb_ctz32 (x >> shift) + shift; 500 return ecb_ctz32 (x >> shift) + shift;
501#endif
429 } 502 }
430 503
431 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 504 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
432 ecb_function_ int 505 ecb_function_ ecb_const int
433 ecb_popcount32 (uint32_t x) 506 ecb_popcount32 (uint32_t x)
434 { 507 {
435 x -= (x >> 1) & 0x55555555; 508 x -= (x >> 1) & 0x55555555;
436 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 509 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
437 x = ((x >> 4) + x) & 0x0f0f0f0f; 510 x = ((x >> 4) + x) & 0x0f0f0f0f;
438 x *= 0x01010101; 511 x *= 0x01010101;
439 512
440 return x >> 24; 513 return x >> 24;
441 } 514 }
442 515
443 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 516 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
444 ecb_function_ int ecb_ld32 (uint32_t x) 517 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
445 { 518 {
519#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
520 unsigned long r;
521 _BitScanReverse (&r, x);
522 return (int)r;
523#else
446 int r = 0; 524 int r = 0;
447 525
448 if (x >> 16) { x >>= 16; r += 16; } 526 if (x >> 16) { x >>= 16; r += 16; }
449 if (x >> 8) { x >>= 8; r += 8; } 527 if (x >> 8) { x >>= 8; r += 8; }
450 if (x >> 4) { x >>= 4; r += 4; } 528 if (x >> 4) { x >>= 4; r += 4; }
451 if (x >> 2) { x >>= 2; r += 2; } 529 if (x >> 2) { x >>= 2; r += 2; }
452 if (x >> 1) { r += 1; } 530 if (x >> 1) { r += 1; }
453 531
454 return r; 532 return r;
533#endif
455 } 534 }
456 535
457 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 536 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
458 ecb_function_ int ecb_ld64 (uint64_t x) 537 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
459 { 538 {
539#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
540 unsigned long r;
541 _BitScanReverse64 (&r, x);
542 return (int)r;
543#else
460 int r = 0; 544 int r = 0;
461 545
462 if (x >> 32) { x >>= 32; r += 32; } 546 if (x >> 32) { x >>= 32; r += 32; }
463 547
464 return r + ecb_ld32 (x); 548 return r + ecb_ld32 (x);
465 }
466#endif 549#endif
550 }
551#endif
467 552
468ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 553ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
469ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 554ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
470ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 555ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
471ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 556ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
472 557
473ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 558ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
474ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 559ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
475{ 560{
476 return ( (x * 0x0802U & 0x22110U) 561 return ( (x * 0x0802U & 0x22110U)
477 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 562 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
478} 563}
479 564
480ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 565ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
481ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 566ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
482{ 567{
483 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 568 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
484 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 569 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
485 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 570 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
486 x = ( x >> 8 ) | ( x << 8); 571 x = ( x >> 8 ) | ( x << 8);
487 572
488 return x; 573 return x;
489} 574}
490 575
491ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 576ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
492ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 577ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
493{ 578{
494 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 579 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
495 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 580 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
496 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 581 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
497 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 582 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
… …
500 return x; 585 return x;
501} 586}
502 587
503/* popcount64 is only available on 64 bit cpus as gcc builtin */ 588/* popcount64 is only available on 64 bit cpus as gcc builtin */
504/* so for this version we are lazy */ 589/* so for this version we are lazy */
505ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 590ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
506ecb_function_ int 591ecb_function_ ecb_const int
507ecb_popcount64 (uint64_t x) 592ecb_popcount64 (uint64_t x)
508{ 593{
509 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 594 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
510} 595}
511 596
512ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 597ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
513ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 598ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
514ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 599ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
515ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 600ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
516ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 601ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
517ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 602ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
518ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 603ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
519ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 604ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
520 605
521ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 606ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
522ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 607ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
523ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 608ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
524ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 609ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
525ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 610ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
526ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 611ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
527ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 612ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
528ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 613ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
614
615#if ECB_CPP
616
617inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
618inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
619inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
620inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
621
622inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
623inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
624inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
625inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
626
627inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
628inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
629inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
630inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
631
632inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
633inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
634inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
635inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
636
637inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
638inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
639inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
640
641inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
642inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
643inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
644inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
645
646inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
647inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
648inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
649inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
650
651#endif
529 652
530#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) 653#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
654 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
655 #define ecb_bswap16(x) __builtin_bswap16 (x)
656 #else
531 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 657 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
658 #endif
532 #define ecb_bswap32(x) __builtin_bswap32 (x) 659 #define ecb_bswap32(x) __builtin_bswap32 (x)
533 #define ecb_bswap64(x) __builtin_bswap64 (x) 660 #define ecb_bswap64(x) __builtin_bswap64 (x)
661#elif _MSC_VER
662 #include <stdlib.h>
663 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
664 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
665 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
534#else 666#else
535 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 667 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
536 ecb_function_ uint16_t 668 ecb_function_ ecb_const uint16_t
537 ecb_bswap16 (uint16_t x) 669 ecb_bswap16 (uint16_t x)
538 { 670 {
539 return ecb_rotl16 (x, 8); 671 return ecb_rotl16 (x, 8);
540 } 672 }
541 673
542 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 674 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
543 ecb_function_ uint32_t 675 ecb_function_ ecb_const uint32_t
544 ecb_bswap32 (uint32_t x) 676 ecb_bswap32 (uint32_t x)
545 { 677 {
546 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 678 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
547 } 679 }
548 680
549 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 681 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
550 ecb_function_ uint64_t 682 ecb_function_ ecb_const uint64_t
551 ecb_bswap64 (uint64_t x) 683 ecb_bswap64 (uint64_t x)
552 { 684 {
553 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 685 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
554 } 686 }
555#endif 687#endif
556 688
557#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) 689#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
558 #define ecb_unreachable() __builtin_unreachable () 690 #define ecb_unreachable() __builtin_unreachable ()
559#else 691#else
560 /* this seems to work fine, but gcc always emits a warning for it :/ */ 692 /* this seems to work fine, but gcc always emits a warning for it :/ */
561 ecb_inline void ecb_unreachable (void) ecb_noreturn; 693 ecb_inline ecb_noreturn void ecb_unreachable (void);
562 ecb_inline void ecb_unreachable (void) { } 694 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
563#endif 695#endif
564 696
565/* try to tell the compiler that some condition is definitely true */ 697/* try to tell the compiler that some condition is definitely true */
566#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 698#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
567 699
568ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 700ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
569ecb_inline unsigned char 701ecb_inline ecb_const uint32_t
570ecb_byteorder_helper (void) 702ecb_byteorder_helper (void)
571{ 703{
572 /* the union code still generates code under pressure in gcc, */ 704 /* the union code still generates code under pressure in gcc, */
573 /* but less than using pointers, and always seems to */ 705 /* but less than using pointers, and always seems to */
574 /* successfully return a constant. */ 706 /* successfully return a constant. */
575 /* the reason why we have this horrible preprocessor mess */ 707 /* the reason why we have this horrible preprocessor mess */
576 /* is to avoid it in all cases, at least on common architectures */ 708 /* is to avoid it in all cases, at least on common architectures */
577 /* or when using a recent enough gcc version (>= 4.6) */ 709 /* or when using a recent enough gcc version (>= 4.6) */
578#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
579 return 0x44;
580#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 710#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
711 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
712 #define ECB_LITTLE_ENDIAN 1
581 return 0x44; 713 return 0x44332211;
582#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 714#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
715 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
716 #define ECB_BIG_ENDIAN 1
583 return 0x11; 717 return 0x11223344;
584#else 718#else
585 union 719 union
586 { 720 {
721 uint8_t c[4];
587 uint32_t i; 722 uint32_t u;
588 uint8_t c;
589 } u = { 0x11223344 }; 723 } u = { 0x11, 0x22, 0x33, 0x44 };
590 return u.c; 724 return u.u;
591#endif 725#endif
592} 726}
593 727
594ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 728ecb_inline ecb_const ecb_bool ecb_big_endian (void);
595ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 729ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
596ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 730ecb_inline ecb_const ecb_bool ecb_little_endian (void);
597ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 731ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
732
733/*****************************************************************************/
734/* unaligned load/store */
735
736ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
737ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
738ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
739
740ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
741ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
742ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
743
744ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
745ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
746ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
747
748ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
749ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
750ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
751
752ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
753ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
754ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
755
756ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
757ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
758ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
759
760ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
761ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
762ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
763
764ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
765ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
766ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
767
768ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
769ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
770ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
771
772ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
773ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
774ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
775
776#if ECB_CPP
777
778inline uint8_t ecb_bswap (uint8_t v) { return v; }
779inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
780inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
781inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
782
783template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
784template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
785template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
786template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
787template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
788template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
789template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
790template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
791
792template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
793template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
794template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
795template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
796template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
797template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
798template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
799template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
800
801#endif
802
803/*****************************************************************************/
598 804
599#if ECB_GCC_VERSION(3,0) || ECB_C99 805#if ECB_GCC_VERSION(3,0) || ECB_C99
600 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 806 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
601#else 807#else
602 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 808 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
603#endif 809#endif
604 810
605#if __cplusplus 811#if ECB_CPP
606 template<typename T> 812 template<typename T>
607 static inline T ecb_div_rd (T val, T div) 813 static inline T ecb_div_rd (T val, T div)
608 { 814 {
609 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 815 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
610 } 816 }
… …
627 } 833 }
628#else 834#else
629 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 835 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
630#endif 836#endif
631 837
838/*****************************************************************************/
839
840ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
841ecb_function_ ecb_const uint32_t
842ecb_binary16_to_binary32 (uint32_t x)
843{
844 unsigned int s = (x & 0x8000) << (31 - 15);
845 int e = (x >> 10) & 0x001f;
846 unsigned int m = x & 0x03ff;
847
848 if (ecb_expect_false (e == 31))
849 /* infinity or NaN */
850 e = 255 - (127 - 15);
851 else if (ecb_expect_false (!e))
852 {
853 if (ecb_expect_true (!m))
854 /* zero, handled by code below by forcing e to 0 */
855 e = 0 - (127 - 15);
856 else
857 {
858 /* subnormal, renormalise */
859 unsigned int s = 10 - ecb_ld32 (m);
860
861 m = (m << s) & 0x3ff; /* mask implicit bit */
862 e -= s - 1;
863 }
864 }
865
866 /* e and m now are normalised, or zero, (or inf or nan) */
867 e += 127 - 15;
868
869 return s | (e << 23) | (m << (23 - 10));
870}
871
872ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
873ecb_function_ ecb_const uint16_t
874ecb_binary32_to_binary16 (uint32_t x)
875{
876 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
877 unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
878 unsigned int m = x & 0x007fffff;
879
880 x &= 0x7fffffff;
881
882 /* if it's within range of binary16 normals, use fast path */
883 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
884 {
885 /* mantissa round-to-even */
886 m += 0x00000fff + ((m >> (23 - 10)) & 1);
887
888 /* handle overflow */
889 if (ecb_expect_false (m >= 0x00800000))
890 {
891 m >>= 1;
892 e += 1;
893 }
894
895 return s | (e << 10) | (m >> (23 - 10));
896 }
897
898 /* handle large numbers and infinity */
899 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
900 return s | 0x7c00;
901
902 /* handle zero, subnormals and small numbers */
903 if (ecb_expect_true (x < 0x38800000))
904 {
905 /* zero */
906 if (ecb_expect_true (!x))
907 return s;
908
909 /* handle subnormals */
910
911 /* too small, will be zero */
912 if (e < (14 - 24)) /* might not be sharp, but is good enough */
913 return s;
914
915 m |= 0x00800000; /* make implicit bit explicit */
916
917 /* very tricky - we need to round to the nearest e (+10) bit value */
918 {
919 unsigned int bits = 14 - e;
920 unsigned int half = (1 << (bits - 1)) - 1;
921 unsigned int even = (m >> bits) & 1;
922
923 /* if this overflows, we will end up with a normalised number */
924 m = (m + half + even) >> bits;
925 }
926
927 return s | m;
928 }
929
930 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
931 m >>= 13;
932
933 return s | 0x7c00 | m | !m;
934}
935
632/*******************************************************************************/ 936/*******************************************************************************/
633/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 937/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
634 938
635/* basically, everything uses "ieee pure-endian" floating point numbers */ 939/* basically, everything uses "ieee pure-endian" floating point numbers */
636/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 940/* the only noteworthy exception is ancient armle, which uses order 43218765 */
637#if 0 \ 941#if 0 \
638 || __i386 || __i386__ \ 942 || __i386 || __i386__ \
639 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 943 || ECB_GCC_AMD64 \
640 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 944 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
641 || defined __s390__ || defined __s390x__ \ 945 || defined __s390__ || defined __s390x__ \
642 || defined __mips__ \ 946 || defined __mips__ \
643 || defined __alpha__ \ 947 || defined __alpha__ \
644 || defined __hppa__ \ 948 || defined __hppa__ \
645 || defined __ia64__ \ 949 || defined __ia64__ \
646 || defined __m68k__ \ 950 || defined __m68k__ \
647 || defined __m88k__ \ 951 || defined __m88k__ \
648 || defined __sh__ \ 952 || defined __sh__ \
649 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ 953 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
650 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ 954 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
651 || defined __aarch64__ 955 || defined __aarch64__
652 #define ECB_STDFP 1 956 #define ECB_STDFP 1
653 #include <string.h> /* for memcpy */
654#else 957#else
655 #define ECB_STDFP 0 958 #define ECB_STDFP 0
656#endif 959#endif
657 960
658#ifndef ECB_NO_LIBM 961#ifndef ECB_NO_LIBM
… …
672 #define ECB_NAN ECB_INFINITY 975 #define ECB_NAN ECB_INFINITY
673 #endif 976 #endif
674 977
675 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L 978 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
676 #define ecb_ldexpf(x,e) ldexpf ((x), (e)) 979 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
980 #define ecb_frexpf(x,e) frexpf ((x), (e))
677 #else 981 #else
678 #define ecb_ldexpf(x,e) (float) ldexp ((double)(x), (e)) 982 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
983 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
679 #endif 984 #endif
680
681 /* converts an ieee half/binary16 to a float */
682 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
683 ecb_function_ float
684 ecb_binary16_to_float (uint16_t x)
685 {
686 int e = (x >> 10) & 0x1f;
687 int m = x & 0x3ff;
688 float r;
689
690 if (!e ) r = ecb_ldexpf (m , -24);
691 else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
692 else if (m ) r = ECB_NAN;
693 else r = ECB_INFINITY;
694
695 return x & 0x8000 ? -r : r;
696 }
697 985
698 /* convert a float to ieee single/binary32 */ 986 /* convert a float to ieee single/binary32 */
699 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 987 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
700 ecb_function_ uint32_t 988 ecb_function_ ecb_const uint32_t
701 ecb_float_to_binary32 (float x) 989 ecb_float_to_binary32 (float x)
702 { 990 {
703 uint32_t r; 991 uint32_t r;
704 992
705 #if ECB_STDFP 993 #if ECB_STDFP
… …
712 if (x == 0e0f ) return 0x00000000U; 1000 if (x == 0e0f ) return 0x00000000U;
713 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 1001 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
714 if (x < -3.40282346638528860e+38f) return 0xff800000U; 1002 if (x < -3.40282346638528860e+38f) return 0xff800000U;
715 if (x != x ) return 0x7fbfffffU; 1003 if (x != x ) return 0x7fbfffffU;
716 1004
717 m = frexpf (x, &e) * 0x1000000U; 1005 m = ecb_frexpf (x, &e) * 0x1000000U;
718 1006
719 r = m & 0x80000000U; 1007 r = m & 0x80000000U;
720 1008
721 if (r) 1009 if (r)
722 m = -m; 1010 m = -m;
… …
734 1022
735 return r; 1023 return r;
736 } 1024 }
737 1025
738 /* converts an ieee single/binary32 to a float */ 1026 /* converts an ieee single/binary32 to a float */
739 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 1027 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
740 ecb_function_ float 1028 ecb_function_ ecb_const float
741 ecb_binary32_to_float (uint32_t x) 1029 ecb_binary32_to_float (uint32_t x)
742 { 1030 {
743 float r; 1031 float r;
744 1032
745 #if ECB_STDFP 1033 #if ECB_STDFP
… …
764 1052
765 return r; 1053 return r;
766 } 1054 }
767 1055
768 /* convert a double to ieee double/binary64 */ 1056 /* convert a double to ieee double/binary64 */
769 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 1057 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
770 ecb_function_ uint64_t 1058 ecb_function_ ecb_const uint64_t
771 ecb_double_to_binary64 (double x) 1059 ecb_double_to_binary64 (double x)
772 { 1060 {
773 uint64_t r; 1061 uint64_t r;
774 1062
775 #if ECB_STDFP 1063 #if ECB_STDFP
… …
804 1092
805 return r; 1093 return r;
806 } 1094 }
807 1095
808 /* converts an ieee double/binary64 to a double */ 1096 /* converts an ieee double/binary64 to a double */
809 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 1097 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
810 ecb_function_ double 1098 ecb_function_ ecb_const double
811 ecb_binary64_to_double (uint64_t x) 1099 ecb_binary64_to_double (uint64_t x)
812 { 1100 {
813 double r; 1101 double r;
814 1102
815 #if ECB_STDFP 1103 #if ECB_STDFP
… …
833 #endif 1121 #endif
834 1122
835 return r; 1123 return r;
836 } 1124 }
837 1125
838#endif 1126 /* convert a float to ieee half/binary16 */
1127 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1128 ecb_function_ ecb_const uint16_t
1129 ecb_float_to_binary16 (float x)
1130 {
1131 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1132 }
839 1133
840#endif 1134 /* convert an ieee half/binary16 to float */
1135 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1136 ecb_function_ ecb_const float
1137 ecb_binary16_to_float (uint16_t x)
1138 {
1139 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1140 }
841 1141
1142#endif
1143
1144#endif
1145

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines