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

Comparing libeio/ecb.h (file contents):
Revision 1.24 by root, Sat Jun 13 00:06:13 2015 UTC vs.
Revision 1.31 by root, Thu Sep 3 17:20:45 2020 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2015 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 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
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 ECB_CPP 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)
… …
349 #define ecb_deprecated __declspec (deprecated) 392 #define ecb_deprecated __declspec (deprecated)
350#else 393#else
351 #define ecb_deprecated ecb_attribute ((__deprecated__)) 394 #define ecb_deprecated ecb_attribute ((__deprecated__))
352#endif 395#endif
353 396
354#if __MSC_VER >= 1500 397#if _MSC_VER >= 1500
355 #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) 398 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
356#elif ECB_GCC_VERSION(4,5) 399#elif ECB_GCC_VERSION(4,5)
357 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) 400 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
358#else 401#else
359 #define ecb_deprecated_message(msg) ecb_deprecated 402 #define ecb_deprecated_message(msg) ecb_deprecated
… …
367 410
368#define ecb_unused ecb_attribute ((__unused__)) 411#define ecb_unused ecb_attribute ((__unused__))
369#define ecb_const ecb_attribute ((__const__)) 412#define ecb_const ecb_attribute ((__const__))
370#define ecb_pure ecb_attribute ((__pure__)) 413#define ecb_pure ecb_attribute ((__pure__))
371 414
372/* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
373#if ECB_C11 || __IBMC_NORETURN 415#if ECB_C11 || __IBMC_NORETURN
374 /* 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 */
375 #define ecb_noreturn _Noreturn 417 #define ecb_noreturn _Noreturn
376#elif ECB_CPP11 418#elif ECB_CPP11
377 #define ecb_noreturn [[noreturn]] 419 #define ecb_noreturn [[noreturn]]
378#elif _MSC_VER >= 1200 420#elif _MSC_VER >= 1200
421 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
379 #define ecb_noreturn __declspec (noreturn) 422 #define ecb_noreturn __declspec (noreturn)
380#else 423#else
381 #define ecb_noreturn ecb_attribute ((__noreturn__)) 424 #define ecb_noreturn ecb_attribute ((__noreturn__))
382#endif 425#endif
383 426
… …
415#else 458#else
416 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); 459 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
417 ecb_function_ ecb_const int 460 ecb_function_ ecb_const int
418 ecb_ctz32 (uint32_t x) 461 ecb_ctz32 (uint32_t x)
419 { 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
420 int r = 0; 468 int r = 0;
421 469
422 x &= ~x + 1; /* this isolates the lowest bit */ 470 x &= ~x + 1; /* this isolates the lowest bit */
423 471
424#if ECB_branchless_on_i386 472#if ECB_branchless_on_i386
… …
434 if (x & 0xff00ff00) r += 8; 482 if (x & 0xff00ff00) r += 8;
435 if (x & 0xffff0000) r += 16; 483 if (x & 0xffff0000) r += 16;
436#endif 484#endif
437 485
438 return r; 486 return r;
487#endif
439 } 488 }
440 489
441 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); 490 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
442 ecb_function_ ecb_const int 491 ecb_function_ ecb_const int
443 ecb_ctz64 (uint64_t x) 492 ecb_ctz64 (uint64_t x)
444 { 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
445 int shift = x & 0xffffffffU ? 0 : 32; 499 int shift = x & 0xffffffff ? 0 : 32;
446 return ecb_ctz32 (x >> shift) + shift; 500 return ecb_ctz32 (x >> shift) + shift;
501#endif
447 } 502 }
448 503
449 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); 504 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
450 ecb_function_ ecb_const int 505 ecb_function_ ecb_const int
451 ecb_popcount32 (uint32_t x) 506 ecb_popcount32 (uint32_t x)
… …
459 } 514 }
460 515
461 ecb_function_ ecb_const int ecb_ld32 (uint32_t x); 516 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
462 ecb_function_ ecb_const int ecb_ld32 (uint32_t x) 517 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
463 { 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
464 int r = 0; 524 int r = 0;
465 525
466 if (x >> 16) { x >>= 16; r += 16; } 526 if (x >> 16) { x >>= 16; r += 16; }
467 if (x >> 8) { x >>= 8; r += 8; } 527 if (x >> 8) { x >>= 8; r += 8; }
468 if (x >> 4) { x >>= 4; r += 4; } 528 if (x >> 4) { x >>= 4; r += 4; }
469 if (x >> 2) { x >>= 2; r += 2; } 529 if (x >> 2) { x >>= 2; r += 2; }
470 if (x >> 1) { r += 1; } 530 if (x >> 1) { r += 1; }
471 531
472 return r; 532 return r;
533#endif
473 } 534 }
474 535
475 ecb_function_ ecb_const int ecb_ld64 (uint64_t x); 536 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
476 ecb_function_ ecb_const int ecb_ld64 (uint64_t x) 537 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
477 { 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
478 int r = 0; 544 int r = 0;
479 545
480 if (x >> 32) { x >>= 32; r += 32; } 546 if (x >> 32) { x >>= 32; r += 32; }
481 547
482 return r + ecb_ld32 (x); 548 return r + ecb_ld32 (x);
549#endif
483 } 550 }
484#endif 551#endif
485 552
486ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); 553ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
487ecb_function_ ecb_const 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)); }
… …
543ecb_inline ecb_const 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); }
544ecb_inline ecb_const 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); }
545ecb_inline ecb_const 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); }
546ecb_inline ecb_const 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); }
547 614
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
652
548#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
549 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 657 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
658 #endif
550 #define ecb_bswap32(x) __builtin_bswap32 (x) 659 #define ecb_bswap32(x) __builtin_bswap32 (x)
551 #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)))
552#else 666#else
553 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); 667 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
554 ecb_function_ ecb_const uint16_t 668 ecb_function_ ecb_const uint16_t
555 ecb_bswap16 (uint16_t x) 669 ecb_bswap16 (uint16_t x)
556 { 670 {
… …
581#endif 695#endif
582 696
583/* try to tell the compiler that some condition is definitely true */ 697/* try to tell the compiler that some condition is definitely true */
584#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 698#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
585 699
586ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); 700ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
587ecb_inline ecb_const unsigned char 701ecb_inline ecb_const uint32_t
588ecb_byteorder_helper (void) 702ecb_byteorder_helper (void)
589{ 703{
590 /* the union code still generates code under pressure in gcc, */ 704 /* the union code still generates code under pressure in gcc, */
591 /* but less than using pointers, and always seems to */ 705 /* but less than using pointers, and always seems to */
592 /* successfully return a constant. */ 706 /* successfully return a constant. */
593 /* the reason why we have this horrible preprocessor mess */ 707 /* the reason why we have this horrible preprocessor mess */
594 /* 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 */
595 /* or when using a recent enough gcc version (>= 4.6) */ 709 /* or when using a recent enough gcc version (>= 4.6) */
596#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
597 return 0x44;
598#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
599 return 0x44; 713 return 0x44332211;
600#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
601 return 0x11; 717 return 0x11223344;
602#else 718#else
603 union 719 union
604 { 720 {
721 uint8_t c[4];
605 uint32_t i; 722 uint32_t u;
606 uint8_t c;
607 } u = { 0x11223344 }; 723 } u = { 0x11, 0x22, 0x33, 0x44 };
608 return u.c; 724 return u.u;
609#endif 725#endif
610} 726}
611 727
612ecb_inline ecb_const ecb_bool ecb_big_endian (void); 728ecb_inline ecb_const ecb_bool ecb_big_endian (void);
613ecb_inline ecb_const 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; }
614ecb_inline ecb_const ecb_bool ecb_little_endian (void); 730ecb_inline ecb_const ecb_bool ecb_little_endian (void);
615ecb_inline ecb_const 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/*****************************************************************************/
804/* division */
616 805
617#if ECB_GCC_VERSION(3,0) || ECB_C99 806#if ECB_GCC_VERSION(3,0) || ECB_C99
807 /* C99 tightened the definition of %, so we can use a more efficient version */
618 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 808 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
619#else 809#else
620 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 810 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
621#endif 811#endif
622 812
… …
633 } 823 }
634#else 824#else
635 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) 825 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
636 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) 826 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
637#endif 827#endif
828
829/*****************************************************************************/
830/* array length */
638 831
639#if ecb_cplusplus_does_not_suck 832#if ecb_cplusplus_does_not_suck
640 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 833 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
641 template<typename T, int N> 834 template<typename T, int N>
642 static inline int ecb_array_length (const T (&arr)[N]) 835 static inline int ecb_array_length (const T (&arr)[N])
… …
645 } 838 }
646#else 839#else
647 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 840 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
648#endif 841#endif
649 842
843/*****************************************************************************/
844/* IEEE 754-2008 half float conversions */
845
846ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
847ecb_function_ ecb_const uint32_t
848ecb_binary16_to_binary32 (uint32_t x)
849{
850 unsigned int s = (x & 0x8000) << (31 - 15);
851 int e = (x >> 10) & 0x001f;
852 unsigned int m = x & 0x03ff;
853
854 if (ecb_expect_false (e == 31))
855 /* infinity or NaN */
856 e = 255 - (127 - 15);
857 else if (ecb_expect_false (!e))
858 {
859 if (ecb_expect_true (!m))
860 /* zero, handled by code below by forcing e to 0 */
861 e = 0 - (127 - 15);
862 else
863 {
864 /* subnormal, renormalise */
865 unsigned int s = 10 - ecb_ld32 (m);
866
867 m = (m << s) & 0x3ff; /* mask implicit bit */
868 e -= s - 1;
869 }
870 }
871
872 /* e and m now are normalised, or zero, (or inf or nan) */
873 e += 127 - 15;
874
875 return s | (e << 23) | (m << (23 - 10));
876}
877
878ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
879ecb_function_ ecb_const uint16_t
880ecb_binary32_to_binary16 (uint32_t x)
881{
882 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
883 unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
884 unsigned int m = x & 0x007fffff;
885
886 x &= 0x7fffffff;
887
888 /* if it's within range of binary16 normals, use fast path */
889 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
890 {
891 /* mantissa round-to-even */
892 m += 0x00000fff + ((m >> (23 - 10)) & 1);
893
894 /* handle overflow */
895 if (ecb_expect_false (m >= 0x00800000))
896 {
897 m >>= 1;
898 e += 1;
899 }
900
901 return s | (e << 10) | (m >> (23 - 10));
902 }
903
904 /* handle large numbers and infinity */
905 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
906 return s | 0x7c00;
907
908 /* handle zero, subnormals and small numbers */
909 if (ecb_expect_true (x < 0x38800000))
910 {
911 /* zero */
912 if (ecb_expect_true (!x))
913 return s;
914
915 /* handle subnormals */
916
917 /* too small, will be zero */
918 if (e < (14 - 24)) /* might not be sharp, but is good enough */
919 return s;
920
921 m |= 0x00800000; /* make implicit bit explicit */
922
923 /* very tricky - we need to round to the nearest e (+10) bit value */
924 {
925 unsigned int bits = 14 - e;
926 unsigned int half = (1 << (bits - 1)) - 1;
927 unsigned int even = (m >> bits) & 1;
928
929 /* if this overflows, we will end up with a normalised number */
930 m = (m + half + even) >> bits;
931 }
932
933 return s | m;
934 }
935
936 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
937 m >>= 13;
938
939 return s | 0x7c00 | m | !m;
940}
941
650/*******************************************************************************/ 942/*******************************************************************************/
651/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 943/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
652 944
653/* basically, everything uses "ieee pure-endian" floating point numbers */ 945/* basically, everything uses "ieee pure-endian" floating point numbers */
654/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 946/* the only noteworthy exception is ancient armle, which uses order 43218765 */
655#if 0 \ 947#if 0 \
656 || __i386 || __i386__ \ 948 || __i386 || __i386__ \
657 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 949 || ECB_GCC_AMD64 \
658 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 950 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
659 || defined __s390__ || defined __s390x__ \ 951 || defined __s390__ || defined __s390x__ \
660 || defined __mips__ \ 952 || defined __mips__ \
661 || defined __alpha__ \ 953 || defined __alpha__ \
662 || defined __hppa__ \ 954 || defined __hppa__ \
663 || defined __ia64__ \ 955 || defined __ia64__ \
664 || defined __m68k__ \ 956 || defined __m68k__ \
665 || defined __m88k__ \ 957 || defined __m88k__ \
666 || defined __sh__ \ 958 || defined __sh__ \
667 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ 959 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
668 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ 960 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
669 || defined __aarch64__ 961 || defined __aarch64__
670 #define ECB_STDFP 1 962 #define ECB_STDFP 1
671 #include <string.h> /* for memcpy */
672#else 963#else
673 #define ECB_STDFP 0 964 #define ECB_STDFP 0
674#endif 965#endif
675 966
676#ifndef ECB_NO_LIBM 967#ifndef ECB_NO_LIBM
… …
690 #define ECB_NAN ECB_INFINITY 981 #define ECB_NAN ECB_INFINITY
691 #endif 982 #endif
692 983
693 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L 984 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
694 #define ecb_ldexpf(x,e) ldexpf ((x), (e)) 985 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
986 #define ecb_frexpf(x,e) frexpf ((x), (e))
695 #else 987 #else
696 #define ecb_ldexpf(x,e) (float) ldexp ((x), (e)) 988 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
989 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
697 #endif 990 #endif
698
699 /* converts an ieee half/binary16 to a float */
700 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
701 ecb_function_ ecb_const float
702 ecb_binary16_to_float (uint16_t x)
703 {
704 int e = (x >> 10) & 0x1f;
705 int m = x & 0x3ff;
706 float r;
707
708 if (!e ) r = ecb_ldexpf (m , -24);
709 else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
710 else if (m ) r = ECB_NAN;
711 else r = ECB_INFINITY;
712
713 return x & 0x8000 ? -r : r;
714 }
715 991
716 /* convert a float to ieee single/binary32 */ 992 /* convert a float to ieee single/binary32 */
717 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); 993 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
718 ecb_function_ ecb_const uint32_t 994 ecb_function_ ecb_const uint32_t
719 ecb_float_to_binary32 (float x) 995 ecb_float_to_binary32 (float x)
… …
730 if (x == 0e0f ) return 0x00000000U; 1006 if (x == 0e0f ) return 0x00000000U;
731 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 1007 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
732 if (x < -3.40282346638528860e+38f) return 0xff800000U; 1008 if (x < -3.40282346638528860e+38f) return 0xff800000U;
733 if (x != x ) return 0x7fbfffffU; 1009 if (x != x ) return 0x7fbfffffU;
734 1010
735 m = frexpf (x, &e) * 0x1000000U; 1011 m = ecb_frexpf (x, &e) * 0x1000000U;
736 1012
737 r = m & 0x80000000U; 1013 r = m & 0x80000000U;
738 1014
739 if (r) 1015 if (r)
740 m = -m; 1016 m = -m;
… …
851 #endif 1127 #endif
852 1128
853 return r; 1129 return r;
854 } 1130 }
855 1131
856#endif 1132 /* convert a float to ieee half/binary16 */
1133 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1134 ecb_function_ ecb_const uint16_t
1135 ecb_float_to_binary16 (float x)
1136 {
1137 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1138 }
857 1139
858#endif 1140 /* convert an ieee half/binary16 to float */
1141 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1142 ecb_function_ ecb_const float
1143 ecb_binary16_to_float (uint16_t x)
1144 {
1145 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1146 }
859 1147
1148#endif
1149
1150#endif
1151

Diff Legend

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