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

Comparing libecb/ecb.h (file contents):
Revision 1.124 by root, Tue Dec 3 15:35:14 2013 UTC vs.
Revision 1.190 by root, Mon Jun 21 21:36:13 2021 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2013 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de>
5 * Copyright (©) 2011 Emanuele Giaquinta 5 * Copyright (©) 2011 Emanuele Giaquinta
6 * All rights reserved. 6 * All rights reserved.
7 * 7 *
8 * Redistribution and use in source and binary forms, with or without modifica- 8 * Redistribution and use in source and binary forms, with or without modifica-
9 * tion, are permitted provided that the following conditions are met: 9 * tion, are permitted provided that the following conditions are met:
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 0x00010003 45#define ECB_VERSION 0x00010009
35 46
36#ifdef _WIN32 47#include <string.h> /* for memcpy */
48
49#if defined (_WIN32) && !defined (__MINGW32__)
37 typedef signed char int8_t; 50 typedef signed char int8_t;
38 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;
39 typedef signed short int16_t; 54 typedef signed short int16_t;
40 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;
41 typedef signed int int32_t; 58 typedef signed int int32_t;
42 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;
43 #if __GNUC__ 62 #if __GNUC__
44 typedef signed long long int64_t; 63 typedef signed long long int64_t;
45 typedef unsigned long long uint64_t; 64 typedef unsigned long long uint64_t;
46 #else /* _MSC_VER || __BORLANDC__ */ 65 #else /* _MSC_VER || __BORLANDC__ */
47 typedef signed __int64 int64_t; 66 typedef signed __int64 int64_t;
48 typedef unsigned __int64 uint64_t; 67 typedef unsigned __int64 uint64_t;
49 #endif 68 #endif
69 typedef int64_t int_fast64_t;
70 typedef uint64_t uint_fast64_t;
50 #ifdef _WIN64 71 #ifdef _WIN64
51 #define ECB_PTRSIZE 8 72 #define ECB_PTRSIZE 8
52 typedef uint64_t uintptr_t; 73 typedef uint64_t uintptr_t;
53 typedef int64_t intptr_t; 74 typedef int64_t intptr_t;
54 #else 75 #else
56 typedef uint32_t uintptr_t; 77 typedef uint32_t uintptr_t;
57 typedef int32_t intptr_t; 78 typedef int32_t intptr_t;
58 #endif 79 #endif
59#else 80#else
60 #include <inttypes.h> 81 #include <inttypes.h>
61 #if UINTMAX_MAX > 0xffffffffU 82 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
62 #define ECB_PTRSIZE 8 83 #define ECB_PTRSIZE 8
63 #else 84 #else
64 #define ECB_PTRSIZE 4 85 #define ECB_PTRSIZE 4
65 #endif 86 #endif
66#endif 87#endif
67 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
68/* work around x32 idiocy by defining proper macros */ 100/* work around x32 idiocy by defining proper macros */
69#if __x86_64 || _M_AMD64 101#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
70 #if _ILP32 102 #if _ILP32
71 #define ECB_AMD64_X32 1 103 #define ECB_AMD64_X32 1
72 #else 104 #else
73 #define ECB_AMD64 1 105 #define ECB_AMD64 1
74 #endif 106 #endif
107#endif
108
109#if ECB_PTRSIZE >= 8 || ECB_AMD64_X32
110 #define ECB_64BIT_NATIVE 1
111#else
112 #define ECB_64BIT_NATIVE 0
75#endif 113#endif
76 114
77/* many compilers define _GNUC_ to some versions but then only implement 115/* many compilers define _GNUC_ to some versions but then only implement
78 * what their idiot authors think are the "more important" extensions, 116 * what their idiot authors think are the "more important" extensions,
79 * causing enormous grief in return for some better fake benchmark numbers. 117 * causing enormous grief in return for some better fake benchmark numbers.
80 * or so. 118 * or so.
81 * we try to detect these and simply assume they are not gcc - if they have 119 * we try to detect these and simply assume they are not gcc - if they have
82 * an issue with that they should have done it right in the first place. 120 * an issue with that they should have done it right in the first place.
83 */ 121 */
84#ifndef ECB_GCC_VERSION
85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ 122#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
86 #define ECB_GCC_VERSION(major,minor) 0 123 #define ECB_GCC_VERSION(major,minor) 0
87 #else 124#else
88 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 125 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
89 #endif 126#endif
90#endif
91 127
92#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ 128#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
93#define ECB_C99 (__STDC_VERSION__ >= 199901L) 129
94#define ECB_C11 (__STDC_VERSION__ >= 201112L) 130#if __clang__ && defined __has_builtin
131 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
132#else
133 #define ECB_CLANG_BUILTIN(x) 0
134#endif
135
136#if __clang__ && defined __has_extension
137 #define ECB_CLANG_EXTENSION(x) __has_extension (x)
138#else
139 #define ECB_CLANG_EXTENSION(x) 0
140#endif
141
95#define ECB_CPP (__cplusplus+0) 142#define ECB_CPP (__cplusplus+0)
96#define ECB_CPP11 (__cplusplus >= 201103L) 143#define ECB_CPP11 (__cplusplus >= 201103L)
144#define ECB_CPP14 (__cplusplus >= 201402L)
145#define ECB_CPP17 (__cplusplus >= 201703L)
146
147#if ECB_CPP
148 #define ECB_C 0
149 #define ECB_STDC_VERSION 0
150#else
151 #define ECB_C 1
152 #define ECB_STDC_VERSION __STDC_VERSION__
153#endif
154
155#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
156#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
157#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
97 158
98#if ECB_CPP 159#if ECB_CPP
99 #define ECB_EXTERN_C extern "C" 160 #define ECB_EXTERN_C extern "C"
100 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 161 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
101 #define ECB_EXTERN_C_END } 162 #define ECB_EXTERN_C_END }
116 177
117#if ECB_NO_SMP 178#if ECB_NO_SMP
118 #define ECB_MEMORY_FENCE do { } while (0) 179 #define ECB_MEMORY_FENCE do { } while (0)
119#endif 180#endif
120 181
182/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
183#if __xlC__ && ECB_CPP
184 #include <builtins.h>
185#endif
186
187#if 1400 <= _MSC_VER
188 #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189#endif
190
121#ifndef ECB_MEMORY_FENCE 191#ifndef ECB_MEMORY_FENCE
122 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 192 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
193 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
123 #if __i386 || __i386__ 194 #if __i386 || __i386__
124 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 195 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
125 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 196 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
126 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 197 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
127 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 198 #elif ECB_GCC_AMD64
128 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 199 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
129 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 200 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
130 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 201 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
131 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 202 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 203 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
204 #elif defined __ARM_ARCH_2__ \
205 || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
206 || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
207 || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
208 || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
209 || defined __ARM_ARCH_5TEJ__
210 /* should not need any, unless running old code on newer cpu - arm doesn't support that */
133 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 211 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
134 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 212 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213 || defined __ARM_ARCH_6T2__
135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 214 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
136 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 215 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
137 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 216 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
138 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
218 #elif __aarch64__
219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
139 #elif (__sparc || __sparc__) && !__sparcv8 220 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") 221 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
141 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 222 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 223 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
143 #elif defined __s390__ || defined __s390x__ 224 #elif defined __s390__ || defined __s390x__
144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 225 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
165 246
166#ifndef ECB_MEMORY_FENCE 247#ifndef ECB_MEMORY_FENCE
167 #if ECB_GCC_VERSION(4,7) 248 #if ECB_GCC_VERSION(4,7)
168 /* see comment below (stdatomic.h) about the C11 memory model. */ 249 /* see comment below (stdatomic.h) about the C11 memory model. */
169 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 250 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
251 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
252 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
253 #undef ECB_MEMORY_FENCE_RELAXED
254 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
170 255
171 /* The __has_feature syntax from clang is so misdesigned that we cannot use it 256 #elif ECB_CLANG_EXTENSION(c_atomic)
172 * without risking compile time errors with other compilers. We *could*
173 * define our own ecb_clang_has_feature, but I just can't be bothered to work
174 * around this shit time and again.
175 * #elif defined __clang && __has_feature (cxx_atomic)
176 * // see comment below (stdatomic.h) about the C11 memory model. 257 /* see comment below (stdatomic.h) about the C11 memory model. */
177 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 258 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
178 */ 259 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
260 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
261 #undef ECB_MEMORY_FENCE_RELAXED
262 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
179 263
180 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 264 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
181 #define ECB_MEMORY_FENCE __sync_synchronize () 265 #define ECB_MEMORY_FENCE __sync_synchronize ()
266 #elif _MSC_VER >= 1500 /* VC++ 2008 */
267 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
268 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
269 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
270 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
271 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
182 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 272 #elif _MSC_VER >= 1400 /* VC++ 2005 */
183 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 273 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
184 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 274 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
185 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 275 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
186 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 276 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
187 #elif defined _WIN32 277 #elif defined _WIN32
188 #include <WinNT.h> 278 #include <WinNT.h>
189 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 279 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
190 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 280 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
191 #include <mbarrier.h> 281 #include <mbarrier.h>
192 #define ECB_MEMORY_FENCE __machine_rw_barrier () 282 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
193 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 283 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
194 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () 284 #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
285 #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
195 #elif __xlC__ 286 #elif __xlC__
196 #define ECB_MEMORY_FENCE __sync () 287 #define ECB_MEMORY_FENCE __sync ()
197 #endif 288 #endif
198#endif 289#endif
199 290
200#ifndef ECB_MEMORY_FENCE 291#ifndef ECB_MEMORY_FENCE
201 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 292 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
202 /* we assume that these memory fences work on all variables/all memory accesses, */ 293 /* we assume that these memory fences work on all variables/all memory accesses, */
203 /* not just C11 atomics and atomic accesses */ 294 /* not just C11 atomics and atomic accesses */
204 #include <stdatomic.h> 295 #include <stdatomic.h>
205 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
206 /* any fence other than seq_cst, which isn't very efficient for us. */
207 /* Why that is, we don't know - either the C11 memory model is quite useless */
208 /* for most usages, or gcc and clang have a bug */
209 /* I *currently* lean towards the latter, and inefficiently implement */
210 /* all three of ecb's fences as a seq_cst fence */
211 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) 296 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
297 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
298 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
212 #endif 299 #endif
213#endif 300#endif
214 301
215#ifndef ECB_MEMORY_FENCE 302#ifndef ECB_MEMORY_FENCE
216 #if !ECB_AVOID_PTHREADS 303 #if !ECB_AVOID_PTHREADS
236 323
237#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 324#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
238 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 325 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
239#endif 326#endif
240 327
328#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
329 #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
330#endif
331
241/*****************************************************************************/ 332/*****************************************************************************/
242 333
243#if __cplusplus 334#if ECB_CPP
244 #define ecb_inline static inline 335 #define ecb_inline static inline
245#elif ECB_GCC_VERSION(2,5) 336#elif ECB_GCC_VERSION(2,5)
246 #define ecb_inline static __inline__ 337 #define ecb_inline static __inline__
247#elif ECB_C99 338#elif ECB_C99
248 #define ecb_inline static inline 339 #define ecb_inline static inline
262 353
263#define ECB_CONCAT_(a, b) a ## b 354#define ECB_CONCAT_(a, b) a ## b
264#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
265#define ECB_STRINGIFY_(a) # a 356#define ECB_STRINGIFY_(a) # a
266#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 357#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
267 359
268#define ecb_function_ ecb_inline 360#define ecb_function_ ecb_inline
269 361
270#if ECB_GCC_VERSION(3,1) 362#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
271 #define ecb_attribute(attrlist) __attribute__(attrlist) 363 #define ecb_attribute(attrlist) __attribute__ (attrlist)
364#else
365 #define ecb_attribute(attrlist)
366#endif
367
368#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
272 #define ecb_is_constant(expr) __builtin_constant_p (expr) 369 #define ecb_is_constant(expr) __builtin_constant_p (expr)
370#else
371 /* possible C11 impl for integral types
372 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
373 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
374
375 #define ecb_is_constant(expr) 0
376#endif
377
378#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
273 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 379 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
380#else
381 #define ecb_expect(expr,value) (expr)
382#endif
383
384#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
274 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 385 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
275#else 386#else
276 #define ecb_attribute(attrlist)
277 #define ecb_is_constant(expr) 0
278 #define ecb_expect(expr,value) (expr)
279 #define ecb_prefetch(addr,rw,locality) 387 #define ecb_prefetch(addr,rw,locality)
280#endif 388#endif
281 389
282/* no emulation for ecb_decltype */ 390/* no emulation for ecb_decltype */
283#if ECB_GCC_VERSION(4,5) 391#if ECB_CPP11
392 // older implementations might have problems with decltype(x)::type, work around it
393 template<class T> struct ecb_decltype_t { typedef T type; };
284 #define ecb_decltype(x) __decltype(x) 394 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
285#elif ECB_GCC_VERSION(3,0) 395#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
286 #define ecb_decltype(x) __typeof(x) 396 #define ecb_decltype(x) __typeof__ (x)
287#endif 397#endif
288 398
399#if _MSC_VER >= 1300
400 #define ecb_deprecated __declspec (deprecated)
401#else
402 #define ecb_deprecated ecb_attribute ((__deprecated__))
403#endif
404
405#if _MSC_VER >= 1500
406 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
407#elif ECB_GCC_VERSION(4,5)
408 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
409#else
410 #define ecb_deprecated_message(msg) ecb_deprecated
411#endif
412
413#if _MSC_VER >= 1400
414 #define ecb_noinline __declspec (noinline)
415#else
289#define ecb_noinline ecb_attribute ((__noinline__)) 416 #define ecb_noinline ecb_attribute ((__noinline__))
417#endif
418
290#define ecb_unused ecb_attribute ((__unused__)) 419#define ecb_unused ecb_attribute ((__unused__))
291#define ecb_const ecb_attribute ((__const__)) 420#define ecb_const ecb_attribute ((__const__))
292#define ecb_pure ecb_attribute ((__pure__)) 421#define ecb_pure ecb_attribute ((__pure__))
293 422
294#if ECB_C11 423#if ECB_C11 || __IBMC_NORETURN
424 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
295 #define ecb_noreturn _Noreturn 425 #define ecb_noreturn _Noreturn
426#elif ECB_CPP11
427 #define ecb_noreturn [[noreturn]]
428#elif _MSC_VER >= 1200
429 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
430 #define ecb_noreturn __declspec (noreturn)
296#else 431#else
297 #define ecb_noreturn ecb_attribute ((__noreturn__)) 432 #define ecb_noreturn ecb_attribute ((__noreturn__))
298#endif 433#endif
299 434
300#if ECB_GCC_VERSION(4,3) 435#if ECB_GCC_VERSION(4,3)
315/* for compatibility to the rest of the world */ 450/* for compatibility to the rest of the world */
316#define ecb_likely(expr) ecb_expect_true (expr) 451#define ecb_likely(expr) ecb_expect_true (expr)
317#define ecb_unlikely(expr) ecb_expect_false (expr) 452#define ecb_unlikely(expr) ecb_expect_false (expr)
318 453
319/* count trailing zero bits and count # of one bits */ 454/* count trailing zero bits and count # of one bits */
320#if ECB_GCC_VERSION(3,4) 455#if ECB_GCC_VERSION(3,4) \
456 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
457 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
458 && ECB_CLANG_BUILTIN(__builtin_popcount))
321 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ 459 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
322 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 460 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
323 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 461 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
324 #define ecb_ctz32(x) __builtin_ctz (x) 462 #define ecb_ctz32(x) __builtin_ctz (x)
325 #define ecb_ctz64(x) __builtin_ctzll (x) 463 #define ecb_ctz64(x) __builtin_ctzll (x)
326 #define ecb_popcount32(x) __builtin_popcount (x) 464 #define ecb_popcount32(x) __builtin_popcount (x)
327 /* no popcountll */ 465 /* no popcountll */
328#else 466#else
329 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 467 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
330 ecb_function_ int 468 ecb_function_ ecb_const int
331 ecb_ctz32 (uint32_t x) 469 ecb_ctz32 (uint32_t x)
332 { 470 {
471#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
472 unsigned long r;
473 _BitScanForward (&r, x);
474 return (int)r;
475#else
333 int r = 0; 476 int r = 0;
334 477
335 x &= ~x + 1; /* this isolates the lowest bit */ 478 x &= ~x + 1; /* this isolates the lowest bit */
336 479
337#if ECB_branchless_on_i386 480#if ECB_branchless_on_i386
347 if (x & 0xff00ff00) r += 8; 490 if (x & 0xff00ff00) r += 8;
348 if (x & 0xffff0000) r += 16; 491 if (x & 0xffff0000) r += 16;
349#endif 492#endif
350 493
351 return r; 494 return r;
495#endif
352 } 496 }
353 497
354 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 498 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
355 ecb_function_ int 499 ecb_function_ ecb_const int
356 ecb_ctz64 (uint64_t x) 500 ecb_ctz64 (uint64_t x)
357 { 501 {
502#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
503 unsigned long r;
504 _BitScanForward64 (&r, x);
505 return (int)r;
506#else
358 int shift = x & 0xffffffffU ? 0 : 32; 507 int shift = x & 0xffffffff ? 0 : 32;
359 return ecb_ctz32 (x >> shift) + shift; 508 return ecb_ctz32 (x >> shift) + shift;
509#endif
360 } 510 }
361 511
362 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 512 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
363 ecb_function_ int 513 ecb_function_ ecb_const int
364 ecb_popcount32 (uint32_t x) 514 ecb_popcount32 (uint32_t x)
365 { 515 {
366 x -= (x >> 1) & 0x55555555; 516 x -= (x >> 1) & 0x55555555;
367 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 517 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
368 x = ((x >> 4) + x) & 0x0f0f0f0f; 518 x = ((x >> 4) + x) & 0x0f0f0f0f;
369 x *= 0x01010101; 519 x *= 0x01010101;
370 520
371 return x >> 24; 521 return x >> 24;
372 } 522 }
373 523
374 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 524 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
375 ecb_function_ int ecb_ld32 (uint32_t x) 525 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
376 { 526 {
527#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
528 unsigned long r;
529 _BitScanReverse (&r, x);
530 return (int)r;
531#else
377 int r = 0; 532 int r = 0;
378 533
379 if (x >> 16) { x >>= 16; r += 16; } 534 if (x >> 16) { x >>= 16; r += 16; }
380 if (x >> 8) { x >>= 8; r += 8; } 535 if (x >> 8) { x >>= 8; r += 8; }
381 if (x >> 4) { x >>= 4; r += 4; } 536 if (x >> 4) { x >>= 4; r += 4; }
382 if (x >> 2) { x >>= 2; r += 2; } 537 if (x >> 2) { x >>= 2; r += 2; }
383 if (x >> 1) { r += 1; } 538 if (x >> 1) { r += 1; }
384 539
385 return r; 540 return r;
541#endif
386 } 542 }
387 543
388 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 544 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
389 ecb_function_ int ecb_ld64 (uint64_t x) 545 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
390 { 546 {
547#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
548 unsigned long r;
549 _BitScanReverse64 (&r, x);
550 return (int)r;
551#else
391 int r = 0; 552 int r = 0;
392 553
393 if (x >> 32) { x >>= 32; r += 32; } 554 if (x >> 32) { x >>= 32; r += 32; }
394 555
395 return r + ecb_ld32 (x); 556 return r + ecb_ld32 (x);
396 }
397#endif 557#endif
558 }
559#endif
398 560
399ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 561ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
400ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 562ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
401ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 563ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
402ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 564ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
403 565
404ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 566ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
405ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 567ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
406{ 568{
407 return ( (x * 0x0802U & 0x22110U) 569 return ( (x * 0x0802U & 0x22110U)
408 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 570 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
409} 571}
410 572
411ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 573ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
412ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 574ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
413{ 575{
414 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 576 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
415 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 577 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
416 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 578 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
417 x = ( x >> 8 ) | ( x << 8); 579 x = ( x >> 8 ) | ( x << 8);
418 580
419 return x; 581 return x;
420} 582}
421 583
422ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 584ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
423ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 585ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
424{ 586{
425 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 587 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
426 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 588 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
427 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 589 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
428 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 590 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
431 return x; 593 return x;
432} 594}
433 595
434/* popcount64 is only available on 64 bit cpus as gcc builtin */ 596/* popcount64 is only available on 64 bit cpus as gcc builtin */
435/* so for this version we are lazy */ 597/* so for this version we are lazy */
436ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 598ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
437ecb_function_ int 599ecb_function_ ecb_const int
438ecb_popcount64 (uint64_t x) 600ecb_popcount64 (uint64_t x)
439{ 601{
440 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 602 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
441} 603}
442 604
443ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 605ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
444ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 606ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
445ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 607ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
446ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 608ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
447ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 609ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
448ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 610ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
449ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 611ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
450ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 612ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
451 613
452ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 614ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
453ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 615ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
454ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 616ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
455ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 617ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
456ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 618ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
457ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 619ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
458ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 620ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
459ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 621ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
460 622
461#if ECB_GCC_VERSION(4,3) 623#if ECB_CPP
624
625inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
626inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
627inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
628inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
629
630inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
631inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
632inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
633inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
634
635inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
636inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
637inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
638inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
639
640inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
641inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
642inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
643inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
644
645inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
646inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
647inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
648
649inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
650inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
651inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
652inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
653
654inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
655inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
656inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
657inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
658
659#endif
660
661#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
662 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
663 #define ecb_bswap16(x) __builtin_bswap16 (x)
664 #else
462 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 665 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
666 #endif
463 #define ecb_bswap32(x) __builtin_bswap32 (x) 667 #define ecb_bswap32(x) __builtin_bswap32 (x)
464 #define ecb_bswap64(x) __builtin_bswap64 (x) 668 #define ecb_bswap64(x) __builtin_bswap64 (x)
669#elif _MSC_VER
670 #include <stdlib.h>
671 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
672 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
673 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
465#else 674#else
466 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 675 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
467 ecb_function_ uint16_t 676 ecb_function_ ecb_const uint16_t
468 ecb_bswap16 (uint16_t x) 677 ecb_bswap16 (uint16_t x)
469 { 678 {
470 return ecb_rotl16 (x, 8); 679 return ecb_rotl16 (x, 8);
471 } 680 }
472 681
473 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 682 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
474 ecb_function_ uint32_t 683 ecb_function_ ecb_const uint32_t
475 ecb_bswap32 (uint32_t x) 684 ecb_bswap32 (uint32_t x)
476 { 685 {
477 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 686 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
478 } 687 }
479 688
480 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 689 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
481 ecb_function_ uint64_t 690 ecb_function_ ecb_const uint64_t
482 ecb_bswap64 (uint64_t x) 691 ecb_bswap64 (uint64_t x)
483 { 692 {
484 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 693 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
485 } 694 }
486#endif 695#endif
487 696
488#if ECB_GCC_VERSION(4,5) 697#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
489 #define ecb_unreachable() __builtin_unreachable () 698 #define ecb_unreachable() __builtin_unreachable ()
490#else 699#else
491 /* this seems to work fine, but gcc always emits a warning for it :/ */ 700 /* this seems to work fine, but gcc always emits a warning for it :/ */
492 ecb_inline void ecb_unreachable (void) ecb_noreturn; 701 ecb_inline ecb_noreturn void ecb_unreachable (void);
493 ecb_inline void ecb_unreachable (void) { } 702 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
494#endif 703#endif
495 704
496/* try to tell the compiler that some condition is definitely true */ 705/* try to tell the compiler that some condition is definitely true */
497#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 706#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
498 707
499ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 708ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
500ecb_inline unsigned char 709ecb_inline ecb_const uint32_t
501ecb_byteorder_helper (void) 710ecb_byteorder_helper (void)
502{ 711{
503 /* the union code still generates code under pressure in gcc, */ 712 /* the union code still generates code under pressure in gcc, */
504 /* but less than using pointers, and always seems to */ 713 /* but less than using pointers, and always seems to */
505 /* successfully return a constant. */ 714 /* successfully return a constant. */
506 /* the reason why we have this horrible preprocessor mess */ 715 /* the reason why we have this horrible preprocessor mess */
507 /* is to avoid it in all cases, at least on common architectures */ 716 /* is to avoid it in all cases, at least on common architectures */
508 /* or when using a recent enough gcc version (>= 4.6) */ 717 /* or when using a recent enough gcc version (>= 4.6) */
509#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
510 return 0x44;
511#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 718#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
719 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
720 #define ECB_LITTLE_ENDIAN 1
512 return 0x44; 721 return 0x44332211;
513#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 722#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
723 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
724 #define ECB_BIG_ENDIAN 1
514 return 0x11; 725 return 0x11223344;
515#else 726#else
516 union 727 union
517 { 728 {
729 uint8_t c[4];
518 uint32_t i; 730 uint32_t u;
519 uint8_t c;
520 } u = { 0x11223344 }; 731 } u = { 0x11, 0x22, 0x33, 0x44 };
521 return u.c; 732 return u.u;
522#endif 733#endif
523} 734}
524 735
525ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 736ecb_inline ecb_const ecb_bool ecb_big_endian (void);
526ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 737ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
527ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 738ecb_inline ecb_const ecb_bool ecb_little_endian (void);
528ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 739ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
740
741/*****************************************************************************/
742/* unaligned load/store */
743
744ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
745ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
746ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
747
748ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
749ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
750ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
751
752ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
754ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
755
756ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
757ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
758ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
759
760ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
761ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
762ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
763
764ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
765ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
766ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
767
768ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
769ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
770ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
771
772ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
773ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
774ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
775
776ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
777ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
778ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
779
780ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
781ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
782ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
783
784#if ECB_CPP
785
786inline uint8_t ecb_bswap (uint8_t v) { return v; }
787inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
788inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
789inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
790
791template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
792template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
793template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
794template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
795template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
796template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
797template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
798template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
799
800template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
801template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
802template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
803template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
804template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
805template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
806template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
807template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
808
809#endif
810
811/*****************************************************************************/
812/* division */
529 813
530#if ECB_GCC_VERSION(3,0) || ECB_C99 814#if ECB_GCC_VERSION(3,0) || ECB_C99
815 /* C99 tightened the definition of %, so we can use a more efficient version */
531 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 816 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
532#else 817#else
533 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 818 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
534#endif 819#endif
535 820
536#if __cplusplus 821#if ECB_CPP
537 template<typename T> 822 template<typename T>
538 static inline T ecb_div_rd (T val, T div) 823 static inline T ecb_div_rd (T val, T div)
539 { 824 {
540 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 825 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
541 } 826 }
546 } 831 }
547#else 832#else
548 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) 833 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
549 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) 834 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
550#endif 835#endif
836
837/*****************************************************************************/
838/* array length */
551 839
552#if ecb_cplusplus_does_not_suck 840#if ecb_cplusplus_does_not_suck
553 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 841 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
554 template<typename T, int N> 842 template<typename T, int N>
555 static inline int ecb_array_length (const T (&arr)[N]) 843 static inline int ecb_array_length (const T (&arr)[N])
558 } 846 }
559#else 847#else
560 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 848 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
561#endif 849#endif
562 850
851/*****************************************************************************/
852/* IEEE 754-2008 half float conversions */
853
854ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
855ecb_function_ ecb_const uint32_t
856ecb_binary16_to_binary32 (uint32_t x)
857{
858 unsigned int s = (x & 0x8000) << (31 - 15);
859 int e = (x >> 10) & 0x001f;
860 unsigned int m = x & 0x03ff;
861
862 if (ecb_expect_false (e == 31))
863 /* infinity or NaN */
864 e = 255 - (127 - 15);
865 else if (ecb_expect_false (!e))
866 {
867 if (ecb_expect_true (!m))
868 /* zero, handled by code below by forcing e to 0 */
869 e = 0 - (127 - 15);
870 else
871 {
872 /* subnormal, renormalise */
873 unsigned int s = 10 - ecb_ld32 (m);
874
875 m = (m << s) & 0x3ff; /* mask implicit bit */
876 e -= s - 1;
877 }
878 }
879
880 /* e and m now are normalised, or zero, (or inf or nan) */
881 e += 127 - 15;
882
883 return s | (e << 23) | (m << (23 - 10));
884}
885
886ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
887ecb_function_ ecb_const uint16_t
888ecb_binary32_to_binary16 (uint32_t x)
889{
890 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
891 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
892 unsigned int m = x & 0x007fffff;
893
894 x &= 0x7fffffff;
895
896 /* if it's within range of binary16 normals, use fast path */
897 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
898 {
899 /* mantissa round-to-even */
900 m += 0x00000fff + ((m >> (23 - 10)) & 1);
901
902 /* handle overflow */
903 if (ecb_expect_false (m >= 0x00800000))
904 {
905 m >>= 1;
906 e += 1;
907 }
908
909 return s | (e << 10) | (m >> (23 - 10));
910 }
911
912 /* handle large numbers and infinity */
913 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
914 return s | 0x7c00;
915
916 /* handle zero, subnormals and small numbers */
917 if (ecb_expect_true (x < 0x38800000))
918 {
919 /* zero */
920 if (ecb_expect_true (!x))
921 return s;
922
923 /* handle subnormals */
924
925 /* too small, will be zero */
926 if (e < (14 - 24)) /* might not be sharp, but is good enough */
927 return s;
928
929 m |= 0x00800000; /* make implicit bit explicit */
930
931 /* very tricky - we need to round to the nearest e (+10) bit value */
932 {
933 unsigned int bits = 14 - e;
934 unsigned int half = (1 << (bits - 1)) - 1;
935 unsigned int even = (m >> bits) & 1;
936
937 /* if this overflows, we will end up with a normalised number */
938 m = (m + half + even) >> bits;
939 }
940
941 return s | m;
942 }
943
944 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
945 m >>= 13;
946
947 return s | 0x7c00 | m | !m;
948}
949
563/*******************************************************************************/ 950/*******************************************************************************/
564/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 951/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
565 952
566/* basically, everything uses "ieee pure-endian" floating point numbers */ 953/* basically, everything uses "ieee pure-endian" floating point numbers */
567/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 954/* the only noteworthy exception is ancient armle, which uses order 43218765 */
568#if 0 \ 955#if 0 \
569 || __i386 || __i386__ \ 956 || __i386 || __i386__ \
570 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 957 || ECB_GCC_AMD64 \
571 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 958 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
572 || defined __arm__ && defined __ARM_EABI__ \
573 || defined __s390__ || defined __s390x__ \ 959 || defined __s390__ || defined __s390x__ \
574 || defined __mips__ \ 960 || defined __mips__ \
575 || defined __alpha__ \ 961 || defined __alpha__ \
576 || defined __hppa__ \ 962 || defined __hppa__ \
577 || defined __ia64__ \ 963 || defined __ia64__ \
578 || defined __m68k__ \ 964 || defined __m68k__ \
579 || defined __m88k__ \ 965 || defined __m88k__ \
580 || defined __sh__ \ 966 || defined __sh__ \
581 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 967 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
968 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
969 || defined __aarch64__
582 #define ECB_STDFP 1 970 #define ECB_STDFP 1
583 #include <string.h> /* for memcpy */
584#else 971#else
585 #define ECB_STDFP 0 972 #define ECB_STDFP 0
586#endif 973#endif
587 974
588#ifndef ECB_NO_LIBM 975#ifndef ECB_NO_LIBM
600 #define ECB_NAN NAN 987 #define ECB_NAN NAN
601 #else 988 #else
602 #define ECB_NAN ECB_INFINITY 989 #define ECB_NAN ECB_INFINITY
603 #endif 990 #endif
604 991
605 /* converts an ieee half/binary16 to a float */ 992 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
606 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; 993 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
607 ecb_function_ float 994 #define ecb_frexpf(x,e) frexpf ((x), (e))
608 ecb_binary16_to_float (uint16_t x) 995 #else
609 { 996 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
610 int e = (x >> 10) & 0x1f; 997 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
611 int m = x & 0x3ff; 998 #endif
612 float r;
613
614 if (!e ) r = ldexpf (m , -24);
615 else if (e != 31) r = ldexpf (m + 0x400, e - 25);
616 else if (m ) r = ECB_NAN;
617 else r = ECB_INFINITY;
618
619 return x & 0x8000 ? -r : r;
620 }
621 999
622 /* convert a float to ieee single/binary32 */ 1000 /* convert a float to ieee single/binary32 */
623 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 1001 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
624 ecb_function_ uint32_t 1002 ecb_function_ ecb_const uint32_t
625 ecb_float_to_binary32 (float x) 1003 ecb_float_to_binary32 (float x)
626 { 1004 {
627 uint32_t r; 1005 uint32_t r;
628 1006
629 #if ECB_STDFP 1007 #if ECB_STDFP
636 if (x == 0e0f ) return 0x00000000U; 1014 if (x == 0e0f ) return 0x00000000U;
637 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 1015 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
638 if (x < -3.40282346638528860e+38f) return 0xff800000U; 1016 if (x < -3.40282346638528860e+38f) return 0xff800000U;
639 if (x != x ) return 0x7fbfffffU; 1017 if (x != x ) return 0x7fbfffffU;
640 1018
641 m = frexpf (x, &e) * 0x1000000U; 1019 m = ecb_frexpf (x, &e) * 0x1000000U;
642 1020
643 r = m & 0x80000000U; 1021 r = m & 0x80000000U;
644 1022
645 if (r) 1023 if (r)
646 m = -m; 1024 m = -m;
658 1036
659 return r; 1037 return r;
660 } 1038 }
661 1039
662 /* converts an ieee single/binary32 to a float */ 1040 /* converts an ieee single/binary32 to a float */
663 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 1041 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
664 ecb_function_ float 1042 ecb_function_ ecb_const float
665 ecb_binary32_to_float (uint32_t x) 1043 ecb_binary32_to_float (uint32_t x)
666 { 1044 {
667 float r; 1045 float r;
668 1046
669 #if ECB_STDFP 1047 #if ECB_STDFP
679 x |= 0x800000U; 1057 x |= 0x800000U;
680 else 1058 else
681 e = 1; 1059 e = 1;
682 1060
683 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ 1061 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
684 r = ldexpf (x * (0.5f / 0x800000U), e - 126); 1062 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
685 1063
686 r = neg ? -r : r; 1064 r = neg ? -r : r;
687 #endif 1065 #endif
688 1066
689 return r; 1067 return r;
690 } 1068 }
691 1069
692 /* convert a double to ieee double/binary64 */ 1070 /* convert a double to ieee double/binary64 */
693 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 1071 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
694 ecb_function_ uint64_t 1072 ecb_function_ ecb_const uint64_t
695 ecb_double_to_binary64 (double x) 1073 ecb_double_to_binary64 (double x)
696 { 1074 {
697 uint64_t r; 1075 uint64_t r;
698 1076
699 #if ECB_STDFP 1077 #if ECB_STDFP
728 1106
729 return r; 1107 return r;
730 } 1108 }
731 1109
732 /* converts an ieee double/binary64 to a double */ 1110 /* converts an ieee double/binary64 to a double */
733 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 1111 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
734 ecb_function_ double 1112 ecb_function_ ecb_const double
735 ecb_binary64_to_double (uint64_t x) 1113 ecb_binary64_to_double (uint64_t x)
736 { 1114 {
737 double r; 1115 double r;
738 1116
739 #if ECB_STDFP 1117 #if ECB_STDFP
757 #endif 1135 #endif
758 1136
759 return r; 1137 return r;
760 } 1138 }
761 1139
762#endif 1140 /* convert a float to ieee half/binary16 */
1141 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1142 ecb_function_ ecb_const uint16_t
1143 ecb_float_to_binary16 (float x)
1144 {
1145 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1146 }
763 1147
764#endif 1148 /* convert an ieee half/binary16 to float */
1149 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1150 ecb_function_ ecb_const float
1151 ecb_binary16_to_float (uint16_t x)
1152 {
1153 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1154 }
765 1155
1156#endif
1157
1158#endif
1159

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines