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

Comparing libecb/ecb.h (file contents):
Revision 1.109 by root, Fri Jun 29 14:05:50 2012 UTC vs.
Revision 1.169 by root, Sat Nov 21 20:56:38 2015 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2015 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:
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- 25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED 26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27 * OF THE POSSIBILITY OF SUCH DAMAGE. 27 * OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * Alternatively, the contents of this file may be used under the terms of
30 * the GNU General Public License ("GPL") version 2 or any later version,
31 * in which case the provisions of the GPL are applicable instead of
32 * the above. If you wish to allow the use of your version of this file
33 * only under the terms of the GPL and not to allow others to use your
34 * version of this file under the BSD license, indicate your decision
35 * by deleting the provisions above and replace them with the notice
36 * and other provisions required by the GPL. If you do not delete the
37 * provisions above, a recipient may use your version of this file under
38 * either the BSD or the GPL.
28 */ 39 */
29 40
30#ifndef ECB_H 41#ifndef ECB_H
31#define ECB_H 42#define ECB_H
32 43
33/* 16 bits major, 16 bits minor */ 44/* 16 bits major, 16 bits minor */
34#define ECB_VERSION 0x00010001 45#define ECB_VERSION 0x00010005
35 46
36#ifdef _WIN32 47#ifdef _WIN32
37 typedef signed char int8_t; 48 typedef signed char int8_t;
38 typedef unsigned char uint8_t; 49 typedef unsigned char uint8_t;
39 typedef signed short int16_t; 50 typedef signed short int16_t;
63 #else 74 #else
64 #define ECB_PTRSIZE 4 75 #define ECB_PTRSIZE 4
65 #endif 76 #endif
66#endif 77#endif
67 78
79#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
80#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
81
82/* work around x32 idiocy by defining proper macros */
83#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
84 #if _ILP32
85 #define ECB_AMD64_X32 1
86 #else
87 #define ECB_AMD64 1
88 #endif
89#endif
90
68/* many compilers define _GNUC_ to some versions but then only implement 91/* many compilers define _GNUC_ to some versions but then only implement
69 * what their idiot authors think are the "more important" extensions, 92 * what their idiot authors think are the "more important" extensions,
70 * causing enormous grief in return for some better fake benchmark numbers. 93 * causing enormous grief in return for some better fake benchmark numbers.
71 * or so. 94 * or so.
72 * we try to detect these and simply assume they are not gcc - if they have 95 * we try to detect these and simply assume they are not gcc - if they have
73 * an issue with that they should have done it right in the first place. 96 * an issue with that they should have done it right in the first place.
74 */ 97 */
75#ifndef ECB_GCC_VERSION
76 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ 98#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
77 #define ECB_GCC_VERSION(major,minor) 0 99 #define ECB_GCC_VERSION(major,minor) 0
78 #else 100#else
79 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 101 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
80 #endif 102#endif
81#endif
82 103
83#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ 104#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
84#define ECB_C99 (__STDC_VERSION__ >= 199901L) 105
85#define ECB_C11 (__STDC_VERSION__ >= 201112L) 106#if __clang__ && defined __has_builtin
107 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
108#else
109 #define ECB_CLANG_BUILTIN(x) 0
110#endif
111
112#if __clang__ && defined __has_extension
113 #define ECB_CLANG_EXTENSION(x) __has_extension (x)
114#else
115 #define ECB_CLANG_EXTENSION(x) 0
116#endif
117
86#define ECB_CPP (__cplusplus+0) 118#define ECB_CPP (__cplusplus+0)
87#define ECB_CPP11 (__cplusplus >= 201103L) 119#define ECB_CPP11 (__cplusplus >= 201103L)
120
121#if ECB_CPP
122 #define ECB_C 0
123 #define ECB_STDC_VERSION 0
124#else
125 #define ECB_C 1
126 #define ECB_STDC_VERSION __STDC_VERSION__
127#endif
128
129#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
130#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
88 131
89#if ECB_CPP 132#if ECB_CPP
90 #define ECB_EXTERN_C extern "C" 133 #define ECB_EXTERN_C extern "C"
91 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 134 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
92 #define ECB_EXTERN_C_END } 135 #define ECB_EXTERN_C_END }
105 #define ECB_NO_SMP 1 148 #define ECB_NO_SMP 1
106#endif 149#endif
107 150
108#if ECB_NO_SMP 151#if ECB_NO_SMP
109 #define ECB_MEMORY_FENCE do { } while (0) 152 #define ECB_MEMORY_FENCE do { } while (0)
153#endif
154
155/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
156#if __xlC__ && ECB_CPP
157 #include <builtins.h>
110#endif 158#endif
111 159
112#ifndef ECB_MEMORY_FENCE 160#ifndef ECB_MEMORY_FENCE
113 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 161 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
114 #if __i386 || __i386__ 162 #if __i386 || __i386__
115 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 163 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
116 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 164 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
117 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 165 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
118 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 166 #elif ECB_GCC_AMD64
119 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 167 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
120 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 168 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
121 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 169 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
122 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 170 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
123 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 171 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
125 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 173 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
126 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 174 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
127 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 175 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
128 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 176 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
129 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 177 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
130 #elif __sparc || __sparc__ 178 #elif __aarch64__
179 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
180 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
131 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") 181 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
132 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 182 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
133 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 183 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
134 #elif defined __s390__ || defined __s390x__ 184 #elif defined __s390__ || defined __s390x__
135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 185 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
136 #elif defined __mips__ 186 #elif defined __mips__
187 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
188 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
137 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 189 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
138 #elif defined __alpha__ 190 #elif defined __alpha__
139 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") 191 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
140 #elif defined __hppa__ 192 #elif defined __hppa__
141 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") 193 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 194 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
143 #elif defined __ia64__ 195 #elif defined __ia64__
144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") 196 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
197 #elif defined __m68k__
198 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
199 #elif defined __m88k__
200 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
201 #elif defined __sh__
202 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
145 #endif 203 #endif
146 #endif 204 #endif
147#endif 205#endif
148 206
149#ifndef ECB_MEMORY_FENCE 207#ifndef ECB_MEMORY_FENCE
150 #if ECB_GCC_VERSION(4,7) 208 #if ECB_GCC_VERSION(4,7)
151 /* see comment below (stdatomic.h) about the C11 memory model. */ 209 /* see comment below (stdatomic.h) about the C11 memory model. */
152 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 210 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
153 /*#elif defined __clang && __has_feature (cxx_atomic)*/ 211 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
212 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
213
214 #elif ECB_CLANG_EXTENSION(c_atomic)
154 /* see comment below (stdatomic.h) about the C11 memory model. */ 215 /* see comment below (stdatomic.h) about the C11 memory model. */
155 /*#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)*/ 216 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
217 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
218 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
219
156 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 220 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
157 #define ECB_MEMORY_FENCE __sync_synchronize () 221 #define ECB_MEMORY_FENCE __sync_synchronize ()
222 #elif _MSC_VER >= 1500 /* VC++ 2008 */
223 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
224 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
225 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
226 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
227 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
158 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 228 #elif _MSC_VER >= 1400 /* VC++ 2005 */
159 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 229 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
160 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 230 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
161 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 231 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
162 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 232 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
182 /* any fence other than seq_cst, which isn't very efficient for us. */ 252 /* any fence other than seq_cst, which isn't very efficient for us. */
183 /* Why that is, we don't know - either the C11 memory model is quite useless */ 253 /* Why that is, we don't know - either the C11 memory model is quite useless */
184 /* for most usages, or gcc and clang have a bug */ 254 /* for most usages, or gcc and clang have a bug */
185 /* I *currently* lean towards the latter, and inefficiently implement */ 255 /* I *currently* lean towards the latter, and inefficiently implement */
186 /* all three of ecb's fences as a seq_cst fence */ 256 /* all three of ecb's fences as a seq_cst fence */
257 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
258 /* for all __atomic_thread_fence's except seq_cst */
187 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) 259 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
188 #endif 260 #endif
189#endif 261#endif
190 262
191#ifndef ECB_MEMORY_FENCE 263#ifndef ECB_MEMORY_FENCE
214 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 286 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
215#endif 287#endif
216 288
217/*****************************************************************************/ 289/*****************************************************************************/
218 290
219#if __cplusplus 291#if ECB_CPP
220 #define ecb_inline static inline 292 #define ecb_inline static inline
221#elif ECB_GCC_VERSION(2,5) 293#elif ECB_GCC_VERSION(2,5)
222 #define ecb_inline static __inline__ 294 #define ecb_inline static __inline__
223#elif ECB_C99 295#elif ECB_C99
224 #define ecb_inline static inline 296 #define ecb_inline static inline
238 310
239#define ECB_CONCAT_(a, b) a ## b 311#define ECB_CONCAT_(a, b) a ## b
240#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 312#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
241#define ECB_STRINGIFY_(a) # a 313#define ECB_STRINGIFY_(a) # a
242#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 314#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
315#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
243 316
244#define ecb_function_ ecb_inline 317#define ecb_function_ ecb_inline
245 318
246#if ECB_GCC_VERSION(3,1) 319#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
247 #define ecb_attribute(attrlist) __attribute__(attrlist) 320 #define ecb_attribute(attrlist) __attribute__ (attrlist)
321#else
322 #define ecb_attribute(attrlist)
323#endif
324
325#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
248 #define ecb_is_constant(expr) __builtin_constant_p (expr) 326 #define ecb_is_constant(expr) __builtin_constant_p (expr)
327#else
328 /* possible C11 impl for integral types
329 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
330 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
331
332 #define ecb_is_constant(expr) 0
333#endif
334
335#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
249 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 336 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
337#else
338 #define ecb_expect(expr,value) (expr)
339#endif
340
341#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
250 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 342 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
251#else 343#else
252 #define ecb_attribute(attrlist)
253 #define ecb_is_constant(expr) 0
254 #define ecb_expect(expr,value) (expr)
255 #define ecb_prefetch(addr,rw,locality) 344 #define ecb_prefetch(addr,rw,locality)
256#endif 345#endif
257 346
258/* no emulation for ecb_decltype */ 347/* no emulation for ecb_decltype */
259#if ECB_GCC_VERSION(4,5) 348#if ECB_CPP11
349 // older implementations might have problems with decltype(x)::type, work around it
350 template<class T> struct ecb_decltype_t { typedef T type; };
260 #define ecb_decltype(x) __decltype(x) 351 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
261#elif ECB_GCC_VERSION(3,0) 352#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
262 #define ecb_decltype(x) __typeof(x) 353 #define ecb_decltype(x) __typeof__ (x)
263#endif 354#endif
264 355
356#if _MSC_VER >= 1300
357 #define ecb_deprecated __declspec (deprecated)
358#else
359 #define ecb_deprecated ecb_attribute ((__deprecated__))
360#endif
361
362#if _MSC_VER >= 1500
363 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
364#elif ECB_GCC_VERSION(4,5)
365 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
366#else
367 #define ecb_deprecated_message(msg) ecb_deprecated
368#endif
369
370#if _MSC_VER >= 1400
371 #define ecb_noinline __declspec (noinline)
372#else
265#define ecb_noinline ecb_attribute ((__noinline__)) 373 #define ecb_noinline ecb_attribute ((__noinline__))
374#endif
375
266#define ecb_unused ecb_attribute ((__unused__)) 376#define ecb_unused ecb_attribute ((__unused__))
267#define ecb_const ecb_attribute ((__const__)) 377#define ecb_const ecb_attribute ((__const__))
268#define ecb_pure ecb_attribute ((__pure__)) 378#define ecb_pure ecb_attribute ((__pure__))
269 379
270#if ECB_C11 380#if ECB_C11 || __IBMC_NORETURN
381 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
271 #define ecb_noreturn _Noreturn 382 #define ecb_noreturn _Noreturn
383#elif ECB_CPP11
384 #define ecb_noreturn [[noreturn]]
385#elif _MSC_VER >= 1200
386 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
387 #define ecb_noreturn __declspec (noreturn)
272#else 388#else
273 #define ecb_noreturn ecb_attribute ((__noreturn__)) 389 #define ecb_noreturn ecb_attribute ((__noreturn__))
274#endif 390#endif
275 391
276#if ECB_GCC_VERSION(4,3) 392#if ECB_GCC_VERSION(4,3)
291/* for compatibility to the rest of the world */ 407/* for compatibility to the rest of the world */
292#define ecb_likely(expr) ecb_expect_true (expr) 408#define ecb_likely(expr) ecb_expect_true (expr)
293#define ecb_unlikely(expr) ecb_expect_false (expr) 409#define ecb_unlikely(expr) ecb_expect_false (expr)
294 410
295/* count trailing zero bits and count # of one bits */ 411/* count trailing zero bits and count # of one bits */
296#if ECB_GCC_VERSION(3,4) 412#if ECB_GCC_VERSION(3,4) \
413 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
414 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
415 && ECB_CLANG_BUILTIN(__builtin_popcount))
297 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ 416 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
298 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 417 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
299 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 418 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
300 #define ecb_ctz32(x) __builtin_ctz (x) 419 #define ecb_ctz32(x) __builtin_ctz (x)
301 #define ecb_ctz64(x) __builtin_ctzll (x) 420 #define ecb_ctz64(x) __builtin_ctzll (x)
302 #define ecb_popcount32(x) __builtin_popcount (x) 421 #define ecb_popcount32(x) __builtin_popcount (x)
303 /* no popcountll */ 422 /* no popcountll */
304#else 423#else
305 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 424 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
306 ecb_function_ int 425 ecb_function_ ecb_const int
307 ecb_ctz32 (uint32_t x) 426 ecb_ctz32 (uint32_t x)
308 { 427 {
309 int r = 0; 428 int r = 0;
310 429
311 x &= ~x + 1; /* this isolates the lowest bit */ 430 x &= ~x + 1; /* this isolates the lowest bit */
325#endif 444#endif
326 445
327 return r; 446 return r;
328 } 447 }
329 448
330 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 449 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
331 ecb_function_ int 450 ecb_function_ ecb_const int
332 ecb_ctz64 (uint64_t x) 451 ecb_ctz64 (uint64_t x)
333 { 452 {
334 int shift = x & 0xffffffffU ? 0 : 32; 453 int shift = x & 0xffffffff ? 0 : 32;
335 return ecb_ctz32 (x >> shift) + shift; 454 return ecb_ctz32 (x >> shift) + shift;
336 } 455 }
337 456
338 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 457 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
339 ecb_function_ int 458 ecb_function_ ecb_const int
340 ecb_popcount32 (uint32_t x) 459 ecb_popcount32 (uint32_t x)
341 { 460 {
342 x -= (x >> 1) & 0x55555555; 461 x -= (x >> 1) & 0x55555555;
343 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 462 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
344 x = ((x >> 4) + x) & 0x0f0f0f0f; 463 x = ((x >> 4) + x) & 0x0f0f0f0f;
345 x *= 0x01010101; 464 x *= 0x01010101;
346 465
347 return x >> 24; 466 return x >> 24;
348 } 467 }
349 468
350 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 469 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
351 ecb_function_ int ecb_ld32 (uint32_t x) 470 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
352 { 471 {
353 int r = 0; 472 int r = 0;
354 473
355 if (x >> 16) { x >>= 16; r += 16; } 474 if (x >> 16) { x >>= 16; r += 16; }
356 if (x >> 8) { x >>= 8; r += 8; } 475 if (x >> 8) { x >>= 8; r += 8; }
359 if (x >> 1) { r += 1; } 478 if (x >> 1) { r += 1; }
360 479
361 return r; 480 return r;
362 } 481 }
363 482
364 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 483 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
365 ecb_function_ int ecb_ld64 (uint64_t x) 484 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
366 { 485 {
367 int r = 0; 486 int r = 0;
368 487
369 if (x >> 32) { x >>= 32; r += 32; } 488 if (x >> 32) { x >>= 32; r += 32; }
370 489
371 return r + ecb_ld32 (x); 490 return r + ecb_ld32 (x);
372 } 491 }
373#endif 492#endif
374 493
375ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 494ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
376ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 495ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
377ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 496ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
378ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 497ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
379 498
380ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 499ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
381ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 500ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
382{ 501{
383 return ( (x * 0x0802U & 0x22110U) 502 return ( (x * 0x0802U & 0x22110U)
384 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 503 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
385} 504}
386 505
387ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 506ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
388ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 507ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
389{ 508{
390 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 509 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
391 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 510 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
392 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 511 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
393 x = ( x >> 8 ) | ( x << 8); 512 x = ( x >> 8 ) | ( x << 8);
394 513
395 return x; 514 return x;
396} 515}
397 516
398ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 517ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
399ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 518ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
400{ 519{
401 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 520 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
402 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 521 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
403 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 522 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
404 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 523 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
407 return x; 526 return x;
408} 527}
409 528
410/* popcount64 is only available on 64 bit cpus as gcc builtin */ 529/* popcount64 is only available on 64 bit cpus as gcc builtin */
411/* so for this version we are lazy */ 530/* so for this version we are lazy */
412ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 531ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
413ecb_function_ int 532ecb_function_ ecb_const int
414ecb_popcount64 (uint64_t x) 533ecb_popcount64 (uint64_t x)
415{ 534{
416 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 535 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
417} 536}
418 537
419ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 538ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
420ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 539ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
421ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 540ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
422ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 541ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
423ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 542ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
424ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 543ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
425ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 544ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
426ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 545ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
427 546
428ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 547ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
429ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 548ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
430ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 549ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
431ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 550ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
432ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 551ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
433ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 552ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
434ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 553ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
435ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 554ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
436 555
437#if ECB_GCC_VERSION(4,3) 556#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
557 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
558 #define ecb_bswap16(x) __builtin_bswap16 (x)
559 #else
438 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 560 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
561 #endif
439 #define ecb_bswap32(x) __builtin_bswap32 (x) 562 #define ecb_bswap32(x) __builtin_bswap32 (x)
440 #define ecb_bswap64(x) __builtin_bswap64 (x) 563 #define ecb_bswap64(x) __builtin_bswap64 (x)
564#elif _MSC_VER
565 #include <stdlib.h>
566 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
567 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
568 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
441#else 569#else
442 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 570 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
443 ecb_function_ uint16_t 571 ecb_function_ ecb_const uint16_t
444 ecb_bswap16 (uint16_t x) 572 ecb_bswap16 (uint16_t x)
445 { 573 {
446 return ecb_rotl16 (x, 8); 574 return ecb_rotl16 (x, 8);
447 } 575 }
448 576
449 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 577 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
450 ecb_function_ uint32_t 578 ecb_function_ ecb_const uint32_t
451 ecb_bswap32 (uint32_t x) 579 ecb_bswap32 (uint32_t x)
452 { 580 {
453 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 581 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
454 } 582 }
455 583
456 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 584 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
457 ecb_function_ uint64_t 585 ecb_function_ ecb_const uint64_t
458 ecb_bswap64 (uint64_t x) 586 ecb_bswap64 (uint64_t x)
459 { 587 {
460 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 588 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
461 } 589 }
462#endif 590#endif
463 591
464#if ECB_GCC_VERSION(4,5) 592#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
465 #define ecb_unreachable() __builtin_unreachable () 593 #define ecb_unreachable() __builtin_unreachable ()
466#else 594#else
467 /* this seems to work fine, but gcc always emits a warning for it :/ */ 595 /* this seems to work fine, but gcc always emits a warning for it :/ */
468 ecb_inline void ecb_unreachable (void) ecb_noreturn; 596 ecb_inline ecb_noreturn void ecb_unreachable (void);
469 ecb_inline void ecb_unreachable (void) { } 597 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
470#endif 598#endif
471 599
472/* try to tell the compiler that some condition is definitely true */ 600/* try to tell the compiler that some condition is definitely true */
473#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 601#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
474 602
475ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 603ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);
476ecb_inline unsigned char 604ecb_inline ecb_const unsigned char
477ecb_byteorder_helper (void) 605ecb_byteorder_helper (void)
478{ 606{
479 /* the union code still generates code under pressure in gcc, */ 607 /* the union code still generates code under pressure in gcc, */
480 /* but less than using pointers, and always seem to */ 608 /* but less than using pointers, and always seems to */
481 /* successfully return a constant. */ 609 /* successfully return a constant. */
482 /* the reason why we have this horrible preprocessor mess */ 610 /* the reason why we have this horrible preprocessor mess */
483 /* is to avoid it in all cases, at least on common architectures */ 611 /* is to avoid it in all cases, at least on common architectures */
484 /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ 612 /* or when using a recent enough gcc version (>= 4.6) */
485#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 613#if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64
486 return 0x44; 614 return 0x44;
487#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 615#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
488 return 0x44; 616 return 0x44;
489#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 617#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
490 retrurn 0x11; 618 return 0x11;
491#else 619#else
492 union 620 union
493 { 621 {
494 uint32_t i; 622 uint32_t i;
495 uint8_t c; 623 uint8_t c;
496 } u = { 0x11223344 }; 624 } u = { 0x11223344 };
497 return u.c; 625 return u.c;
498#endif 626#endif
499} 627}
500 628
501ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 629ecb_inline ecb_const ecb_bool ecb_big_endian (void);
502ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 630ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
503ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 631ecb_inline ecb_const ecb_bool ecb_little_endian (void);
504ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 632ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
505 633
506#if ECB_GCC_VERSION(3,0) || ECB_C99 634#if ECB_GCC_VERSION(3,0) || ECB_C99
507 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 635 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
508#else 636#else
509 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 637 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
510#endif 638#endif
511 639
512#if __cplusplus 640#if ECB_CPP
513 template<typename T> 641 template<typename T>
514 static inline T ecb_div_rd (T val, T div) 642 static inline T ecb_div_rd (T val, T div)
515 { 643 {
516 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 644 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
517 } 645 }
534 } 662 }
535#else 663#else
536 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 664 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
537#endif 665#endif
538 666
667ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint16_t x);
668ecb_function_ ecb_const uint32_t
669ecb_binary16_to_binary32 (uint16_t x)
670{
671 unsigned int s = (x & 0x8000) << (31 - 15);
672 int e = (x >> 10) & 0x001f;
673 unsigned int m = x & 0x03ff;
674
675 if (ecb_expect_false (e == 31))
676 /* infinity or NaN */
677 e = 255 - (127 - 15);
678 else if (ecb_expect_false (!e))
679 {
680 if (ecb_expect_true (!m))
681 /* zero, handled by code below by forcing e to 0 */
682 e = 0 - (127 - 15);
683 else
684 {
685 /* subnormal, renormalise */
686 unsigned int s = 10 - ecb_ld32 (m);
687
688 m = (m << s) & 0x3ff; /* mask implicit bit */
689 e -= s - 1;
690 }
691 }
692
693 /* e and m now are normalised, or zero, (or inf or nan) */
694 e += 127 - 15;
695
696 return s | (e << 23) | (m << (23 - 10));
697}
698
699ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
700ecb_function_ ecb_const uint16_t
701ecb_binary32_to_binary16 (uint32_t x)
702{
703 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
704 unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
705 unsigned int m = x & 0x007fffff;
706
707 x &= 0x7fffffff;
708
709 /* if it's within range of binary16 normals, use fast path */
710 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
711 {
712 /* mantissa round-to-even */
713 m += 0x00000fff + ((m >> (23 - 10)) & 1);
714
715 /* handle overflow */
716 if (ecb_expect_false (m >= 0x00800000))
717 {
718 m >>= 1;
719 e += 1;
720 }
721
722 return s | (e << 10) | (m >> (23 - 10));
723 }
724
725 /* handle large numbers and infinity */
726 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
727 return s | 0x7c00;
728
729 /* handle zero, subnormals and small numbers */
730 if (ecb_expect_true (x < 0x38800000))
731 {
732 /* zero */
733 if (ecb_expect_true (!x))
734 return s;
735
736 /* handle subnormals */
737
738 /* too small, will be zero */
739 if (e < (14 - 24)) /* might not be sharp, but is good enough */
740 return s;
741
742 m |= 0x00800000; /* make implicit bit explicit */
743
744 /* very tricky - we need to round to the nearest e (+10) bit value */
745 {
746 unsigned int bits = 14 - e;
747 unsigned int half = (1 << (bits - 1)) - 1;
748 unsigned int even = (m >> bits) & 1;
749
750 /* if this overflows, we will end up with a normalised number */
751 m = (m + half + even) >> bits;
752 }
753
754 return s | m;
755 }
756
757 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
758 m >>= 13;
759
760 return s | 0x7c00 | m | !m;
761}
762
539/*******************************************************************************/ 763/*******************************************************************************/
540/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 764/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
541 765
542/* basically, everything uses "ieee pure-endian" floating point numbers */ 766/* basically, everything uses "ieee pure-endian" floating point numbers */
543/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 767/* the only noteworthy exception is ancient armle, which uses order 43218765 */
544#if 0 \ 768#if 0 \
545 || __i386 || __i386__ \ 769 || __i386 || __i386__ \
546 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 770 || ECB_GCC_AMD64 \
547 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 771 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
548 || defined __arm__ && defined __ARM_EABI__ \
549 || defined __s390__ || defined __s390x__ \ 772 || defined __s390__ || defined __s390x__ \
550 || defined __mips__ \ 773 || defined __mips__ \
551 || defined __alpha__ \ 774 || defined __alpha__ \
552 || defined __hppa__ \ 775 || defined __hppa__ \
553 || defined __ia64__ \ 776 || defined __ia64__ \
777 || defined __m68k__ \
778 || defined __m88k__ \
779 || defined __sh__ \
554 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 780 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
781 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
782 || defined __aarch64__
555 #define ECB_STDFP 1 783 #define ECB_STDFP 1
556 #include <string.h> /* for memcpy */ 784 #include <string.h> /* for memcpy */
557#else 785#else
558 #define ECB_STDFP 0 786 #define ECB_STDFP 0
559 #include <math.h> /* for frexp*, ldexp* */
560#endif 787#endif
561 788
562#ifndef ECB_NO_LIBM 789#ifndef ECB_NO_LIBM
563 790
791 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
792
793 /* only the oldest of old doesn't have this one. solaris. */
794 #ifdef INFINITY
795 #define ECB_INFINITY INFINITY
796 #else
797 #define ECB_INFINITY HUGE_VAL
798 #endif
799
800 #ifdef NAN
801 #define ECB_NAN NAN
802 #else
803 #define ECB_NAN ECB_INFINITY
804 #endif
805
806 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
807 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
808 #define ecb_frexpf(x,e) frexpf ((x), (e))
809 #else
810 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
811 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
812 #endif
813
564 /* convert a float to ieee single/binary32 */ 814 /* convert a float to ieee single/binary32 */
565 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 815 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
566 ecb_function_ uint32_t 816 ecb_function_ ecb_const uint32_t
567 ecb_float_to_binary32 (float x) 817 ecb_float_to_binary32 (float x)
568 { 818 {
569 uint32_t r; 819 uint32_t r;
570 820
571 #if ECB_STDFP 821 #if ECB_STDFP
578 if (x == 0e0f ) return 0x00000000U; 828 if (x == 0e0f ) return 0x00000000U;
579 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 829 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
580 if (x < -3.40282346638528860e+38f) return 0xff800000U; 830 if (x < -3.40282346638528860e+38f) return 0xff800000U;
581 if (x != x ) return 0x7fbfffffU; 831 if (x != x ) return 0x7fbfffffU;
582 832
583 m = frexpf (x, &e) * 0x1000000U; 833 m = ecb_frexpf (x, &e) * 0x1000000U;
584 834
585 r = m & 0x80000000U; 835 r = m & 0x80000000U;
586 836
587 if (r) 837 if (r)
588 m = -m; 838 m = -m;
600 850
601 return r; 851 return r;
602 } 852 }
603 853
604 /* converts an ieee single/binary32 to a float */ 854 /* converts an ieee single/binary32 to a float */
605 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 855 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
606 ecb_function_ float 856 ecb_function_ ecb_const float
607 ecb_binary32_to_float (uint32_t x) 857 ecb_binary32_to_float (uint32_t x)
608 { 858 {
609 float r; 859 float r;
610 860
611 #if ECB_STDFP 861 #if ECB_STDFP
621 x |= 0x800000U; 871 x |= 0x800000U;
622 else 872 else
623 e = 1; 873 e = 1;
624 874
625 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ 875 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
626 r = ldexpf (x * (0.5f / 0x800000U), e - 126); 876 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
627 877
628 r = neg ? -r : r; 878 r = neg ? -r : r;
629 #endif 879 #endif
630 880
631 return r; 881 return r;
632 } 882 }
633 883
634 /* convert a double to ieee double/binary64 */ 884 /* convert a double to ieee double/binary64 */
635 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 885 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
636 ecb_function_ uint64_t 886 ecb_function_ ecb_const uint64_t
637 ecb_double_to_binary64 (double x) 887 ecb_double_to_binary64 (double x)
638 { 888 {
639 uint64_t r; 889 uint64_t r;
640 890
641 #if ECB_STDFP 891 #if ECB_STDFP
670 920
671 return r; 921 return r;
672 } 922 }
673 923
674 /* converts an ieee double/binary64 to a double */ 924 /* converts an ieee double/binary64 to a double */
675 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 925 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
676 ecb_function_ double 926 ecb_function_ ecb_const double
677 ecb_binary64_to_double (uint64_t x) 927 ecb_binary64_to_double (uint64_t x)
678 { 928 {
679 double r; 929 double r;
680 930
681 #if ECB_STDFP 931 #if ECB_STDFP
699 #endif 949 #endif
700 950
701 return r; 951 return r;
702 } 952 }
703 953
704#endif 954 /* convert a float to ieee half/binary16 */
955 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
956 ecb_function_ ecb_const uint16_t
957 ecb_float_to_binary16 (float x)
958 {
959 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
960 }
705 961
706#endif 962 /* convert an ieee half/binary16 to float */
963 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
964 ecb_function_ ecb_const float
965 ecb_binary16_to_float (uint16_t x)
966 {
967 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
968 }
707 969
970#endif
971
972#endif
973

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines