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

Comparing libecb/ecb.h (file contents):
Revision 1.157 by root, Fri Feb 20 17:17:26 2015 UTC vs.
Revision 1.182 by root, Mon Jan 20 13:41:18 2020 UTC

… …
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#ifdef _WIN32
48 typedef signed char int8_t; 48 typedef signed char int8_t;
49 typedef unsigned char uint8_t; 49 typedef unsigned char uint8_t;
50 typedef signed char int_fast8_t;
51 typedef unsigned char uint_fast8_t;
50 typedef signed short int16_t; 52 typedef signed short int16_t;
51 typedef unsigned short uint16_t; 53 typedef unsigned short uint16_t;
54 typedef signed int int_fast16_t;
55 typedef unsigned int uint_fast16_t;
52 typedef signed int int32_t; 56 typedef signed int int32_t;
53 typedef unsigned int uint32_t; 57 typedef unsigned int uint32_t;
58 typedef signed int int_fast32_t;
59 typedef unsigned int uint_fast32_t;
54 #if __GNUC__ 60 #if __GNUC__
55 typedef signed long long int64_t; 61 typedef signed long long int64_t;
56 typedef unsigned long long uint64_t; 62 typedef unsigned long long uint64_t;
57 #else /* _MSC_VER || __BORLANDC__ */ 63 #else /* _MSC_VER || __BORLANDC__ */
58 typedef signed __int64 int64_t; 64 typedef signed __int64 int64_t;
59 typedef unsigned __int64 uint64_t; 65 typedef unsigned __int64 uint64_t;
60 #endif 66 #endif
67 typedef int64_t int_fast64_t;
68 typedef uint64_t uint_fast64_t;
61 #ifdef _WIN64 69 #ifdef _WIN64
62 #define ECB_PTRSIZE 8 70 #define ECB_PTRSIZE 8
63 typedef uint64_t uintptr_t; 71 typedef uint64_t uintptr_t;
64 typedef int64_t intptr_t; 72 typedef int64_t intptr_t;
65 #else 73 #else
… …
67 typedef uint32_t uintptr_t; 75 typedef uint32_t uintptr_t;
68 typedef int32_t intptr_t; 76 typedef int32_t intptr_t;
69 #endif 77 #endif
70#else 78#else
71 #include <inttypes.h> 79 #include <inttypes.h>
72 #if UINTMAX_MAX > 0xffffffffU 80 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
73 #define ECB_PTRSIZE 8 81 #define ECB_PTRSIZE 8
74 #else 82 #else
75 #define ECB_PTRSIZE 4 83 #define ECB_PTRSIZE 4
76 #endif 84 #endif
77#endif 85#endif
78 86
87#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
88#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
89
90#ifndef ECB_OPTIMIZE_SIZE
91 #if __OPTIMIZE_SIZE__
92 #define ECB_OPTIMIZE_SIZE 1
93 #else
94 #define ECB_OPTIMIZE_SIZE 0
95 #endif
96#endif
97
79/* work around x32 idiocy by defining proper macros */ 98/* work around x32 idiocy by defining proper macros */
80#if __amd64 || __x86_64 || _M_AMD64 || _M_X64 99#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
81 #if _ILP32 100 #if _ILP32
82 #define ECB_AMD64_X32 1 101 #define ECB_AMD64_X32 1
83 #else 102 #else
84 #define ECB_AMD64 1 103 #define ECB_AMD64 1
85 #endif 104 #endif
… …
112 #define ECB_CLANG_EXTENSION(x) 0 131 #define ECB_CLANG_EXTENSION(x) 0
113#endif 132#endif
114 133
115#define ECB_CPP (__cplusplus+0) 134#define ECB_CPP (__cplusplus+0)
116#define ECB_CPP11 (__cplusplus >= 201103L) 135#define ECB_CPP11 (__cplusplus >= 201103L)
136#define ECB_CPP14 (__cplusplus >= 201402L)
137#define ECB_CPP17 (__cplusplus >= 201703L)
117 138
118#if ECB_CPP 139#if ECB_CPP
119 #define ECB_C 0 140 #define ECB_C 0
120 #define ECB_STDC_VERSION 0 141 #define ECB_STDC_VERSION 0
121#else 142#else
… …
123 #define ECB_STDC_VERSION __STDC_VERSION__ 144 #define ECB_STDC_VERSION __STDC_VERSION__
124#endif 145#endif
125 146
126#define ECB_C99 (ECB_STDC_VERSION >= 199901L) 147#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
127#define ECB_C11 (ECB_STDC_VERSION >= 201112L) 148#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
149#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
128 150
129#if ECB_CPP 151#if ECB_CPP
130 #define ECB_EXTERN_C extern "C" 152 #define ECB_EXTERN_C extern "C"
131 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 153 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
132 #define ECB_EXTERN_C_END } 154 #define ECB_EXTERN_C_END }
… …
147 169
148#if ECB_NO_SMP 170#if ECB_NO_SMP
149 #define ECB_MEMORY_FENCE do { } while (0) 171 #define ECB_MEMORY_FENCE do { } while (0)
150#endif 172#endif
151 173
174/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
175#if __xlC__ && ECB_CPP
176 #include <builtins.h>
177#endif
178
179#if 1400 <= _MSC_VER
180 #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
181#endif
182
152#ifndef ECB_MEMORY_FENCE 183#ifndef ECB_MEMORY_FENCE
153 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 184 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
185 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
154 #if __i386 || __i386__ 186 #if __i386 || __i386__
155 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 187 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
156 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 188 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
157 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 189 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
158 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 190 #elif ECB_GCC_AMD64
159 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 191 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
160 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 192 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
161 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 193 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
162 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 194 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
163 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 195 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
196 #elif defined __ARM_ARCH_2__ \
197 || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
198 || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
199 || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
200 || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
201 || defined __ARM_ARCH_5TEJ__
202 /* 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__ \ 203 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
165 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 204 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
205 || defined __ARM_ARCH_6T2__
166 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 206 #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__ \ 207 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
168 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 208 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
169 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 209 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
170 #elif __aarch64__ 210 #elif __aarch64__
171 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") 211 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
172 #elif (__sparc || __sparc__) && !__sparcv8 212 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
173 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") 213 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
174 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 214 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
175 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 215 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
176 #elif defined __s390__ || defined __s390x__ 216 #elif defined __s390__ || defined __s390x__
177 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
… …
200 #if ECB_GCC_VERSION(4,7) 240 #if ECB_GCC_VERSION(4,7)
201 /* see comment below (stdatomic.h) about the C11 memory model. */ 241 /* see comment below (stdatomic.h) about the C11 memory model. */
202 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 242 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
203 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) 243 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
204 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) 244 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
245 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
205 246
206 #elif ECB_CLANG_EXTENSION(c_atomic) 247 #elif ECB_CLANG_EXTENSION(c_atomic)
207 /* see comment below (stdatomic.h) about the C11 memory model. */ 248 /* see comment below (stdatomic.h) about the C11 memory model. */
208 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 249 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
209 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) 250 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
210 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) 251 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
252 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
211 253
212 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 254 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
213 #define ECB_MEMORY_FENCE __sync_synchronize () 255 #define ECB_MEMORY_FENCE __sync_synchronize ()
214 #elif _MSC_VER >= 1500 /* VC++ 2008 */ 256 #elif _MSC_VER >= 1500 /* VC++ 2008 */
215 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ 257 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
… …
225 #elif defined _WIN32 267 #elif defined _WIN32
226 #include <WinNT.h> 268 #include <WinNT.h>
227 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 269 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
228 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 270 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
229 #include <mbarrier.h> 271 #include <mbarrier.h>
230 #define ECB_MEMORY_FENCE __machine_rw_barrier () 272 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
231 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 273 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
232 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () 274 #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
275 #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
233 #elif __xlC__ 276 #elif __xlC__
234 #define ECB_MEMORY_FENCE __sync () 277 #define ECB_MEMORY_FENCE __sync ()
235 #endif 278 #endif
236#endif 279#endif
237 280
238#ifndef ECB_MEMORY_FENCE 281#ifndef ECB_MEMORY_FENCE
239 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 282 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
240 /* we assume that these memory fences work on all variables/all memory accesses, */ 283 /* we assume that these memory fences work on all variables/all memory accesses, */
241 /* not just C11 atomics and atomic accesses */ 284 /* not just C11 atomics and atomic accesses */
242 #include <stdatomic.h> 285 #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) 286 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
287 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
288 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
252 #endif 289 #endif
253#endif 290#endif
254 291
255#ifndef ECB_MEMORY_FENCE 292#ifndef ECB_MEMORY_FENCE
256 #if !ECB_AVOID_PTHREADS 293 #if !ECB_AVOID_PTHREADS
… …
276 313
277#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 314#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
278 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 315 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
279#endif 316#endif
280 317
318#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
319 #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
320#endif
321
281/*****************************************************************************/ 322/*****************************************************************************/
282 323
283#if ECB_CPP 324#if ECB_CPP
284 #define ecb_inline static inline 325 #define ecb_inline static inline
285#elif ECB_GCC_VERSION(2,5) 326#elif ECB_GCC_VERSION(2,5)
… …
349 #define ecb_deprecated __declspec (deprecated) 390 #define ecb_deprecated __declspec (deprecated)
350#else 391#else
351 #define ecb_deprecated ecb_attribute ((__deprecated__)) 392 #define ecb_deprecated ecb_attribute ((__deprecated__))
352#endif 393#endif
353 394
354#if __MSC_VER >= 1500 395#if _MSC_VER >= 1500
355 #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) 396 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
356#elif ECB_GCC_VERSION(4,5) 397#elif ECB_GCC_VERSION(4,5)
357 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) 398 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
358#else 399#else
359 #define ecb_deprecated_message(msg) ecb_deprecated 400 #define ecb_deprecated_message(msg) ecb_deprecated
… …
368#define ecb_unused ecb_attribute ((__unused__)) 409#define ecb_unused ecb_attribute ((__unused__))
369#define ecb_const ecb_attribute ((__const__)) 410#define ecb_const ecb_attribute ((__const__))
370#define ecb_pure ecb_attribute ((__pure__)) 411#define ecb_pure ecb_attribute ((__pure__))
371 412
372#if ECB_C11 || __IBMC_NORETURN 413#if ECB_C11 || __IBMC_NORETURN
373 /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */ 414 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
374 #define ecb_noreturn _Noreturn 415 #define ecb_noreturn _Noreturn
375#elif ECB_CPP11 416#elif ECB_CPP11
376 #define ecb_noreturn [[noreturn]] 417 #define ecb_noreturn [[noreturn]]
377#elif _MSC_VER >= 1200 418#elif _MSC_VER >= 1200
378 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ 419 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
… …
415#else 456#else
416 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); 457 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
417 ecb_function_ ecb_const int 458 ecb_function_ ecb_const int
418 ecb_ctz32 (uint32_t x) 459 ecb_ctz32 (uint32_t x)
419 { 460 {
461#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
462 unsigned long r;
463 _BitScanForward (&r, x);
464 return (int)r;
465#else
420 int r = 0; 466 int r = 0;
421 467
422 x &= ~x + 1; /* this isolates the lowest bit */ 468 x &= ~x + 1; /* this isolates the lowest bit */
423 469
424#if ECB_branchless_on_i386 470#if ECB_branchless_on_i386
… …
434 if (x & 0xff00ff00) r += 8; 480 if (x & 0xff00ff00) r += 8;
435 if (x & 0xffff0000) r += 16; 481 if (x & 0xffff0000) r += 16;
436#endif 482#endif
437 483
438 return r; 484 return r;
485#endif
439 } 486 }
440 487
441 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); 488 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
442 ecb_function_ ecb_const int 489 ecb_function_ ecb_const int
443 ecb_ctz64 (uint64_t x) 490 ecb_ctz64 (uint64_t x)
444 { 491 {
492#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
493 unsigned long r;
494 _BitScanForward64 (&r, x);
495 return (int)r;
496#else
445 int shift = x & 0xffffffffU ? 0 : 32; 497 int shift = x & 0xffffffff ? 0 : 32;
446 return ecb_ctz32 (x >> shift) + shift; 498 return ecb_ctz32 (x >> shift) + shift;
499#endif
447 } 500 }
448 501
449 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); 502 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
450 ecb_function_ ecb_const int 503 ecb_function_ ecb_const int
451 ecb_popcount32 (uint32_t x) 504 ecb_popcount32 (uint32_t x)
… …
459 } 512 }
460 513
461 ecb_function_ ecb_const int ecb_ld32 (uint32_t x); 514 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
462 ecb_function_ ecb_const int ecb_ld32 (uint32_t x) 515 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
463 { 516 {
517#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
518 unsigned long r;
519 _BitScanReverse (&r, x);
520 return (int)r;
521#else
464 int r = 0; 522 int r = 0;
465 523
466 if (x >> 16) { x >>= 16; r += 16; } 524 if (x >> 16) { x >>= 16; r += 16; }
467 if (x >> 8) { x >>= 8; r += 8; } 525 if (x >> 8) { x >>= 8; r += 8; }
468 if (x >> 4) { x >>= 4; r += 4; } 526 if (x >> 4) { x >>= 4; r += 4; }
469 if (x >> 2) { x >>= 2; r += 2; } 527 if (x >> 2) { x >>= 2; r += 2; }
470 if (x >> 1) { r += 1; } 528 if (x >> 1) { r += 1; }
471 529
472 return r; 530 return r;
531#endif
473 } 532 }
474 533
475 ecb_function_ ecb_const int ecb_ld64 (uint64_t x); 534 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
476 ecb_function_ ecb_const int ecb_ld64 (uint64_t x) 535 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
477 { 536 {
537#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
538 unsigned long r;
539 _BitScanReverse64 (&r, x);
540 return (int)r;
541#else
478 int r = 0; 542 int r = 0;
479 543
480 if (x >> 32) { x >>= 32; r += 32; } 544 if (x >> 32) { x >>= 32; r += 32; }
481 545
482 return r + ecb_ld32 (x); 546 return r + ecb_ld32 (x);
547#endif
483 } 548 }
484#endif 549#endif
485 550
486ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); 551ecb_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)); } 552ecb_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); } 608ecb_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); } 609ecb_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); } 610ecb_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); } 611ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
547 612
613#if ECB_CPP
614
615inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
616inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
617inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
618inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
619
620inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
621inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
622inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
623inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
624
625inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
626inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
627inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
628inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
629
630inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
631inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
632inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
633inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
634
635inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
636inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
637inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
638
639inline uint8_t ecb_rotl (uint8_t v) { return ecb_rotl8 (v); }
640inline uint16_t ecb_rotl (uint16_t v) { return ecb_rotl16 (v); }
641inline uint32_t ecb_rotl (uint32_t v) { return ecb_rotl32 (v); }
642inline uint64_t ecb_rotl (uint64_t v) { return ecb_rotl64 (v); }
643
644inline uint8_t ecb_rotr (uint8_t v) { return ecb_rotr8 (v); }
645inline uint16_t ecb_rotr (uint16_t v) { return ecb_rotr16 (v); }
646inline uint32_t ecb_rotr (uint32_t v) { return ecb_rotr32 (v); }
647inline uint64_t ecb_rotr (uint64_t v) { return ecb_rotr64 (v); }
648
649#endif
650
548#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) 651#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
652 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
653 #define ecb_bswap16(x) __builtin_bswap16 (x)
654 #else
549 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 655 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
656 #endif
550 #define ecb_bswap32(x) __builtin_bswap32 (x) 657 #define ecb_bswap32(x) __builtin_bswap32 (x)
551 #define ecb_bswap64(x) __builtin_bswap64 (x) 658 #define ecb_bswap64(x) __builtin_bswap64 (x)
659#elif _MSC_VER
660 #include <stdlib.h>
661 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
662 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
663 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
552#else 664#else
553 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); 665 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
554 ecb_function_ ecb_const uint16_t 666 ecb_function_ ecb_const uint16_t
555 ecb_bswap16 (uint16_t x) 667 ecb_bswap16 (uint16_t x)
556 { 668 {
… …
581#endif 693#endif
582 694
583/* try to tell the compiler that some condition is definitely true */ 695/* try to tell the compiler that some condition is definitely true */
584#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 696#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
585 697
586ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); 698ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
587ecb_inline ecb_const unsigned char 699ecb_inline ecb_const uint32_t
588ecb_byteorder_helper (void) 700ecb_byteorder_helper (void)
589{ 701{
590 /* the union code still generates code under pressure in gcc, */ 702 /* the union code still generates code under pressure in gcc, */
591 /* but less than using pointers, and always seems to */ 703 /* but less than using pointers, and always seems to */
592 /* successfully return a constant. */ 704 /* successfully return a constant. */
593 /* the reason why we have this horrible preprocessor mess */ 705 /* the reason why we have this horrible preprocessor mess */
594 /* is to avoid it in all cases, at least on common architectures */ 706 /* is to avoid it in all cases, at least on common architectures */
595 /* or when using a recent enough gcc version (>= 4.6) */ 707 /* or when using a recent enough gcc version (>= 4.6) */
596#if ((__i386 || __i386__) && !__VOS__) || _M_X86 || __amd64 || __amd64__ || _M_X64
597 return 0x44;
598#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 708#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
709 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
710 #define ECB_LITTLE_ENDIAN 1
599 return 0x44; 711 return 0x44332211;
600#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 712#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
713 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
714 #define ECB_BIG_ENDIAN 1
601 return 0x11; 715 return 0x11223344;
602#else 716#else
603 union 717 union
604 { 718 {
719 uint8_t c[4];
605 uint32_t i; 720 uint32_t u;
606 uint8_t c;
607 } u = { 0x11223344 }; 721 } u = { 0x11, 0x22, 0x33, 0x44 };
608 return u.c; 722 return u.u;
609#endif 723#endif
610} 724}
611 725
612ecb_inline ecb_const ecb_bool ecb_big_endian (void); 726ecb_inline ecb_const ecb_bool ecb_big_endian (void);
613ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 727ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
614ecb_inline ecb_const ecb_bool ecb_little_endian (void); 728ecb_inline ecb_const ecb_bool ecb_little_endian (void);
615ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 729ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
730
731/*****************************************************************************/
732/* unaligned load/store */
733
734ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
735ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
736ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
737
738ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
739ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
740ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
741
742ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
743ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
744ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
745
746ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
747ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
748ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
749
750ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
751ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
752ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
753
754ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
755ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
756ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
757
758ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
759ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
760ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
761
762ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
763ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
764ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
765
766ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
767ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
768ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
769
770ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
771ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
772ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
773
774#ifdef ECB_CPP
775
776inline uint8_t ecb_bswap (uint8_t v) { return v; }
777inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
778inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
779inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
780
781template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
782template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
783template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
784template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
785template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
786template<typename T> inline T ecb_peek_u (const void *ptr) { T v; std::memcpy (&v, ptr, sizeof (v)); return v; }
787template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
788template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
789
790template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
791template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
792template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
793template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
794template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
795template<typename T> inline void ecb_poke_u (void *ptr, T v) { std::memcpy (ptr, &v, sizeof (v)); }
796template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
797template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
798
799#endif
800
801/*****************************************************************************/
616 802
617#if ECB_GCC_VERSION(3,0) || ECB_C99 803#if ECB_GCC_VERSION(3,0) || ECB_C99
618 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 804 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
619#else 805#else
620 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 806 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
… …
645 } 831 }
646#else 832#else
647 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 833 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
648#endif 834#endif
649 835
836/*****************************************************************************/
837
838ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
839ecb_function_ ecb_const uint32_t
840ecb_binary16_to_binary32 (uint32_t x)
841{
842 unsigned int s = (x & 0x8000) << (31 - 15);
843 int e = (x >> 10) & 0x001f;
844 unsigned int m = x & 0x03ff;
845
846 if (ecb_expect_false (e == 31))
847 /* infinity or NaN */
848 e = 255 - (127 - 15);
849 else if (ecb_expect_false (!e))
850 {
851 if (ecb_expect_true (!m))
852 /* zero, handled by code below by forcing e to 0 */
853 e = 0 - (127 - 15);
854 else
855 {
856 /* subnormal, renormalise */
857 unsigned int s = 10 - ecb_ld32 (m);
858
859 m = (m << s) & 0x3ff; /* mask implicit bit */
860 e -= s - 1;
861 }
862 }
863
864 /* e and m now are normalised, or zero, (or inf or nan) */
865 e += 127 - 15;
866
867 return s | (e << 23) | (m << (23 - 10));
868}
869
870ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
871ecb_function_ ecb_const uint16_t
872ecb_binary32_to_binary16 (uint32_t x)
873{
874 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
875 unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
876 unsigned int m = x & 0x007fffff;
877
878 x &= 0x7fffffff;
879
880 /* if it's within range of binary16 normals, use fast path */
881 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
882 {
883 /* mantissa round-to-even */
884 m += 0x00000fff + ((m >> (23 - 10)) & 1);
885
886 /* handle overflow */
887 if (ecb_expect_false (m >= 0x00800000))
888 {
889 m >>= 1;
890 e += 1;
891 }
892
893 return s | (e << 10) | (m >> (23 - 10));
894 }
895
896 /* handle large numbers and infinity */
897 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
898 return s | 0x7c00;
899
900 /* handle zero, subnormals and small numbers */
901 if (ecb_expect_true (x < 0x38800000))
902 {
903 /* zero */
904 if (ecb_expect_true (!x))
905 return s;
906
907 /* handle subnormals */
908
909 /* too small, will be zero */
910 if (e < (14 - 24)) /* might not be sharp, but is good enough */
911 return s;
912
913 m |= 0x00800000; /* make implicit bit explicit */
914
915 /* very tricky - we need to round to the nearest e (+10) bit value */
916 {
917 unsigned int bits = 14 - e;
918 unsigned int half = (1 << (bits - 1)) - 1;
919 unsigned int even = (m >> bits) & 1;
920
921 /* if this overflows, we will end up with a normalised number */
922 m = (m + half + even) >> bits;
923 }
924
925 return s | m;
926 }
927
928 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
929 m >>= 13;
930
931 return s | 0x7c00 | m | !m;
932}
933
650/*******************************************************************************/ 934/*******************************************************************************/
651/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 935/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
652 936
653/* basically, everything uses "ieee pure-endian" floating point numbers */ 937/* basically, everything uses "ieee pure-endian" floating point numbers */
654/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 938/* the only noteworthy exception is ancient armle, which uses order 43218765 */
655#if 0 \ 939#if 0 \
656 || __i386 || __i386__ \ 940 || __i386 || __i386__ \
657 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 941 || ECB_GCC_AMD64 \
658 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 942 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
659 || defined __s390__ || defined __s390x__ \ 943 || defined __s390__ || defined __s390x__ \
660 || defined __mips__ \ 944 || defined __mips__ \
661 || defined __alpha__ \ 945 || defined __alpha__ \
662 || defined __hppa__ \ 946 || defined __hppa__ \
663 || defined __ia64__ \ 947 || defined __ia64__ \
664 || defined __m68k__ \ 948 || defined __m68k__ \
665 || defined __m88k__ \ 949 || defined __m88k__ \
666 || defined __sh__ \ 950 || defined __sh__ \
667 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ 951 || 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__)) \ 952 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
669 || defined __aarch64__ 953 || defined __aarch64__
670 #define ECB_STDFP 1 954 #define ECB_STDFP 1
671 #include <string.h> /* for memcpy */ 955 #include <string.h> /* for memcpy */
672#else 956#else
… …
690 #define ECB_NAN ECB_INFINITY 974 #define ECB_NAN ECB_INFINITY
691 #endif 975 #endif
692 976
693 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L 977 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
694 #define ecb_ldexpf(x,e) ldexpf ((x), (e)) 978 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
979 #define ecb_frexpf(x,e) frexpf ((x), (e))
695 #else 980 #else
696 #define ecb_ldexpf(x,e) (float) ldexp ((x), (e)) 981 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
982 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
697 #endif 983 #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 }
715 984
716 /* convert a float to ieee single/binary32 */ 985 /* convert a float to ieee single/binary32 */
717 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); 986 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
718 ecb_function_ ecb_const uint32_t 987 ecb_function_ ecb_const uint32_t
719 ecb_float_to_binary32 (float x) 988 ecb_float_to_binary32 (float x)
… …
730 if (x == 0e0f ) return 0x00000000U; 999 if (x == 0e0f ) return 0x00000000U;
731 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 1000 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
732 if (x < -3.40282346638528860e+38f) return 0xff800000U; 1001 if (x < -3.40282346638528860e+38f) return 0xff800000U;
733 if (x != x ) return 0x7fbfffffU; 1002 if (x != x ) return 0x7fbfffffU;
734 1003
735 m = frexpf (x, &e) * 0x1000000U; 1004 m = ecb_frexpf (x, &e) * 0x1000000U;
736 1005
737 r = m & 0x80000000U; 1006 r = m & 0x80000000U;
738 1007
739 if (r) 1008 if (r)
740 m = -m; 1009 m = -m;
… …
851 #endif 1120 #endif
852 1121
853 return r; 1122 return r;
854 } 1123 }
855 1124
856#endif 1125 /* convert a float to ieee half/binary16 */
1126 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1127 ecb_function_ ecb_const uint16_t
1128 ecb_float_to_binary16 (float x)
1129 {
1130 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1131 }
857 1132
858#endif 1133 /* convert an ieee half/binary16 to float */
1134 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1135 ecb_function_ ecb_const float
1136 ecb_binary16_to_float (uint16_t x)
1137 {
1138 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1139 }
859 1140
1141#endif
1142
1143#endif
1144

Diff Legend

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