ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.190
Committed: Mon Jun 21 21:36:13 2021 UTC (2 years, 11 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.189: +2 -0 lines
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 /*
2 root 1.17 * libecb - http://software.schmorp.de/pkg/libecb
3 root 1.1 *
4 root 1.189 * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de>
5 root 1.7 * Copyright (©) 2011 Emanuele Giaquinta
6 root 1.1 * All rights reserved.
7     *
8     * Redistribution and use in source and binary forms, with or without modifica-
9     * tion, are permitted provided that the following conditions are met:
10     *
11     * 1. Redistributions of source code must retain the above copyright notice,
12     * this list of conditions and the following disclaimer.
13     *
14     * 2. Redistributions in binary form must reproduce the above copyright
15     * notice, this list of conditions and the following disclaimer in the
16     * documentation and/or other materials provided with the distribution.
17     *
18     * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
19     * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
20     * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
21     * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
22     * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
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,
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
27     * OF THE POSSIBILITY OF SUCH DAMAGE.
28 root 1.133 *
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.
39 root 1.1 */
40    
41     #ifndef ECB_H
42     #define ECB_H
43    
44 root 1.87 /* 16 bits major, 16 bits minor */
45 root 1.189 #define ECB_VERSION 0x00010009
46 root 1.87
47 root 1.184 #include <string.h> /* for memcpy */
48    
49 root 1.187 #if defined (_WIN32) && !defined (__MINGW32__)
50 root 1.44 typedef signed char int8_t;
51     typedef unsigned char uint8_t;
52 root 1.180 typedef signed char int_fast8_t;
53     typedef unsigned char uint_fast8_t;
54 root 1.44 typedef signed short int16_t;
55     typedef unsigned short uint16_t;
56 root 1.180 typedef signed int int_fast16_t;
57     typedef unsigned int uint_fast16_t;
58 root 1.44 typedef signed int int32_t;
59     typedef unsigned int uint32_t;
60 root 1.180 typedef signed int int_fast32_t;
61     typedef unsigned int uint_fast32_t;
62 root 1.44 #if __GNUC__
63     typedef signed long long int64_t;
64     typedef unsigned long long uint64_t;
65 root 1.51 #else /* _MSC_VER || __BORLANDC__ */
66 root 1.44 typedef signed __int64 int64_t;
67     typedef unsigned __int64 uint64_t;
68     #endif
69 root 1.180 typedef int64_t int_fast64_t;
70     typedef uint64_t uint_fast64_t;
71 root 1.87 #ifdef _WIN64
72     #define ECB_PTRSIZE 8
73     typedef uint64_t uintptr_t;
74     typedef int64_t intptr_t;
75     #else
76     #define ECB_PTRSIZE 4
77     typedef uint32_t uintptr_t;
78     typedef int32_t intptr_t;
79     #endif
80 root 1.44 #else
81     #include <inttypes.h>
82 root 1.173 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
83 root 1.87 #define ECB_PTRSIZE 8
84     #else
85     #define ECB_PTRSIZE 4
86     #endif
87 root 1.44 #endif
88 root 1.6
89 sf-exg 1.159 #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
90     #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
91    
92 root 1.179 #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    
100 root 1.114 /* work around x32 idiocy by defining proper macros */
101 sf-exg 1.159 #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
102 root 1.119 #if _ILP32
103 root 1.115 #define ECB_AMD64_X32 1
104 root 1.114 #else
105 root 1.115 #define ECB_AMD64 1
106 root 1.114 #endif
107     #endif
108    
109 root 1.189 #if ECB_PTRSIZE >= 8 || ECB_AMD64_X32
110     #define ECB_64BIT_NATIVE 1
111     #else
112     #define ECB_64BIT_NATIVE 0
113     #endif
114    
115 root 1.12 /* many compilers define _GNUC_ to some versions but then only implement
116     * what their idiot authors think are the "more important" extensions,
117 sf-exg 1.59 * causing enormous grief in return for some better fake benchmark numbers.
118 root 1.18 * or so.
119 root 1.12 * we try to detect these and simply assume they are not gcc - if they have
120     * an issue with that they should have done it right in the first place.
121     */
122 root 1.137 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
123     #define ECB_GCC_VERSION(major,minor) 0
124     #else
125     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
126 root 1.12 #endif
127 root 1.1
128 sf-exg 1.138 #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
129    
130 root 1.147 #if __clang__ && defined __has_builtin
131     #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
132 sf-exg 1.138 #else
133     #define ECB_CLANG_BUILTIN(x) 0
134     #endif
135    
136 root 1.147 #if __clang__ && defined __has_extension
137     #define ECB_CLANG_EXTENSION(x) __has_extension (x)
138 sf-exg 1.140 #else
139     #define ECB_CLANG_EXTENSION(x) 0
140     #endif
141    
142 root 1.91 #define ECB_CPP (__cplusplus+0)
143     #define ECB_CPP11 (__cplusplus >= 201103L)
144 root 1.177 #define ECB_CPP14 (__cplusplus >= 201402L)
145     #define ECB_CPP17 (__cplusplus >= 201703L)
146 root 1.90
147 root 1.102 #if ECB_CPP
148 root 1.127 #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 root 1.177 #define ECB_C17 (ECB_STDC_VERSION >= 201710L)
158 root 1.127
159     #if ECB_CPP
160 root 1.102 #define ECB_EXTERN_C extern "C"
161     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
162     #define ECB_EXTERN_C_END }
163     #else
164     #define ECB_EXTERN_C extern
165     #define ECB_EXTERN_C_BEG
166     #define ECB_EXTERN_C_END
167     #endif
168    
169 root 1.52 /*****************************************************************************/
170    
171 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
172     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
173    
174 root 1.79 #if ECB_NO_THREADS
175 root 1.95 #define ECB_NO_SMP 1
176 root 1.79 #endif
177    
178 root 1.93 #if ECB_NO_SMP
179 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
180 root 1.58 #endif
181    
182 sf-exg 1.165 /* 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 root 1.171 #if 1400 <= _MSC_VER
188     #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189     #endif
190    
191 root 1.52 #ifndef ECB_MEMORY_FENCE
192 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
193 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
194 root 1.73 #if __i386 || __i386__
195 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
196 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
197 root 1.176 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
198 sf-exg 1.159 #elif ECB_GCC_AMD64
199 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
200     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
201 root 1.176 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
202 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
203 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
204 root 1.175 #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 */
211 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
212 root 1.175 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213     || defined __ARM_ARCH_6T2__
214 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
215 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
216 root 1.175 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
217 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
218 root 1.129 #elif __aarch64__
219     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
220 root 1.166 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
221 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
222     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
223     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
224 root 1.85 #elif defined __s390__ || defined __s390x__
225 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
226 root 1.85 #elif defined __mips__
227 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
228 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
229     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
230 root 1.86 #elif defined __alpha__
231 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
232     #elif defined __hppa__
233     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
234     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
235     #elif defined __ia64__
236     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
237 root 1.117 #elif defined __m68k__
238     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
239     #elif defined __m88k__
240     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
241     #elif defined __sh__
242     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
243 root 1.52 #endif
244     #endif
245     #endif
246    
247     #ifndef ECB_MEMORY_FENCE
248 root 1.93 #if ECB_GCC_VERSION(4,7)
249 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
250 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
251 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
252     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
253 root 1.190 #undef ECB_MEMORY_FENCE_RELAXED
254 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
255 root 1.110
256 sf-exg 1.140 #elif ECB_CLANG_EXTENSION(c_atomic)
257     /* see comment below (stdatomic.h) about the C11 memory model. */
258     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
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 root 1.190 #undef ECB_MEMORY_FENCE_RELAXED
262 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
263 root 1.110
264 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
265 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
266 root 1.126 #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()
272 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
273     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
274     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
275     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
276     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
277 root 1.85 #elif defined _WIN32
278 root 1.55 #include <WinNT.h>
279 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
280 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
281     #include <mbarrier.h>
282 root 1.178 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
283     #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
284     #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
285     #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
286 root 1.82 #elif __xlC__
287 root 1.83 #define ECB_MEMORY_FENCE __sync ()
288 root 1.52 #endif
289     #endif
290    
291 root 1.53 #ifndef ECB_MEMORY_FENCE
292 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
293     /* we assume that these memory fences work on all variables/all memory accesses, */
294     /* not just C11 atomics and atomic accesses */
295     #include <stdatomic.h>
296     #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
297 root 1.178 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
298     #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
299 root 1.94 #endif
300     #endif
301    
302     #ifndef ECB_MEMORY_FENCE
303 root 1.62 #if !ECB_AVOID_PTHREADS
304     /*
305     * if you get undefined symbol references to pthread_mutex_lock,
306     * or failure to find pthread.h, then you should implement
307     * the ECB_MEMORY_FENCE operations for your cpu/compiler
308     * OR provide pthread.h and link against the posix thread library
309     * of your system.
310     */
311     #include <pthread.h>
312     #define ECB_NEEDS_PTHREADS 1
313     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
314 root 1.52
315 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
316     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
317     #endif
318     #endif
319    
320 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
321 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
322 root 1.62 #endif
323    
324 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
325 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
326     #endif
327    
328 root 1.178 #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    
332 root 1.52 /*****************************************************************************/
333    
334 root 1.149 #if ECB_CPP
335 root 1.46 #define ecb_inline static inline
336 root 1.38 #elif ECB_GCC_VERSION(2,5)
337 root 1.46 #define ecb_inline static __inline__
338 root 1.39 #elif ECB_C99
339 root 1.46 #define ecb_inline static inline
340 root 1.29 #else
341 root 1.46 #define ecb_inline static
342 root 1.38 #endif
343    
344     #if ECB_GCC_VERSION(3,3)
345     #define ecb_restrict __restrict__
346 root 1.39 #elif ECB_C99
347 root 1.38 #define ecb_restrict restrict
348     #else
349     #define ecb_restrict
350 root 1.4 #endif
351    
352 root 1.38 typedef int ecb_bool;
353    
354 root 1.8 #define ECB_CONCAT_(a, b) a ## b
355     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
356     #define ECB_STRINGIFY_(a) # a
357     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358 root 1.155 #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
359 root 1.8
360 root 1.46 #define ecb_function_ ecb_inline
361 root 1.3
362 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
363 root 1.142 #define ecb_attribute(attrlist) __attribute__ (attrlist)
364 root 1.37 #else
365     #define ecb_attribute(attrlist)
366 sf-exg 1.138 #endif
367 root 1.127
368 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
369     #define ecb_is_constant(expr) __builtin_constant_p (expr)
370     #else
371 root 1.127 /* 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 root 1.37 #define ecb_is_constant(expr) 0
376 sf-exg 1.138 #endif
377    
378     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
379     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
380     #else
381 root 1.37 #define ecb_expect(expr,value) (expr)
382 sf-exg 1.138 #endif
383    
384     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
385     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
386     #else
387 root 1.37 #define ecb_prefetch(addr,rw,locality)
388 root 1.1 #endif
389    
390 root 1.2 /* no emulation for ecb_decltype */
391 root 1.143 #if ECB_CPP11
392 root 1.144 // older implementations might have problems with decltype(x)::type, work around it
393 root 1.146 template<class T> struct ecb_decltype_t { typedef T type; };
394     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
395 root 1.143 #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
396     #define ecb_decltype(x) __typeof__ (x)
397 root 1.1 #endif
398    
399 root 1.135 #if _MSC_VER >= 1300
400 root 1.149 #define ecb_deprecated __declspec (deprecated)
401 root 1.135 #else
402     #define ecb_deprecated ecb_attribute ((__deprecated__))
403     #endif
404    
405 sf-exg 1.162 #if _MSC_VER >= 1500
406 root 1.154 #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
416     #define ecb_noinline ecb_attribute ((__noinline__))
417     #endif
418    
419 root 1.24 #define ecb_unused ecb_attribute ((__unused__))
420     #define ecb_const ecb_attribute ((__const__))
421     #define ecb_pure ecb_attribute ((__pure__))
422 root 1.35
423 root 1.145 #if ECB_C11 || __IBMC_NORETURN
424 sf-exg 1.165 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
425 root 1.90 #define ecb_noreturn _Noreturn
426 root 1.153 #elif ECB_CPP11
427     #define ecb_noreturn [[noreturn]]
428     #elif _MSC_VER >= 1200
429 sf-exg 1.156 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
430 root 1.153 #define ecb_noreturn __declspec (noreturn)
431 root 1.90 #else
432     #define ecb_noreturn ecb_attribute ((__noreturn__))
433     #endif
434    
435 root 1.35 #if ECB_GCC_VERSION(4,3)
436 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
437     #define ecb_hot ecb_attribute ((__hot__))
438     #define ecb_cold ecb_attribute ((__cold__))
439 root 1.35 #else
440     #define ecb_artificial
441     #define ecb_hot
442     #define ecb_cold
443     #endif
444 root 1.1
445 root 1.39 /* put around conditional expressions if you are very sure that the */
446     /* expression is mostly true or mostly false. note that these return */
447     /* booleans, not the expression. */
448 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
449     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
450 root 1.36 /* for compatibility to the rest of the world */
451 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
452     #define ecb_unlikely(expr) ecb_expect_false (expr)
453 root 1.1
454 root 1.3 /* count trailing zero bits and count # of one bits */
455 root 1.139 #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))
459 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
460     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
461     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
462 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
463 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
464 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
465 root 1.49 /* no popcountll */
466 root 1.1 #else
467 root 1.151 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
468     ecb_function_ ecb_const int
469 root 1.35 ecb_ctz32 (uint32_t x)
470     {
471 root 1.171 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
472 root 1.172 unsigned long r;
473 root 1.171 _BitScanForward (&r, x);
474     return (int)r;
475     #else
476 root 1.35 int r = 0;
477    
478 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
479 root 1.35
480 root 1.50 #if ECB_branchless_on_i386
481     r += !!(x & 0xaaaaaaaa) << 0;
482     r += !!(x & 0xcccccccc) << 1;
483     r += !!(x & 0xf0f0f0f0) << 2;
484     r += !!(x & 0xff00ff00) << 3;
485     r += !!(x & 0xffff0000) << 4;
486     #else
487 root 1.35 if (x & 0xaaaaaaaa) r += 1;
488     if (x & 0xcccccccc) r += 2;
489     if (x & 0xf0f0f0f0) r += 4;
490     if (x & 0xff00ff00) r += 8;
491     if (x & 0xffff0000) r += 16;
492 root 1.50 #endif
493 root 1.35
494     return r;
495 root 1.171 #endif
496 root 1.35 }
497    
498 root 1.151 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
499     ecb_function_ ecb_const int
500 root 1.49 ecb_ctz64 (uint64_t x)
501     {
502 root 1.171 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
503 root 1.172 unsigned long r;
504 root 1.171 _BitScanForward64 (&r, x);
505     return (int)r;
506     #else
507 root 1.168 int shift = x & 0xffffffff ? 0 : 32;
508 root 1.50 return ecb_ctz32 (x >> shift) + shift;
509 root 1.171 #endif
510 root 1.49 }
511    
512 root 1.151 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
513     ecb_function_ ecb_const int
514 root 1.35 ecb_popcount32 (uint32_t x)
515     {
516     x -= (x >> 1) & 0x55555555;
517     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
518     x = ((x >> 4) + x) & 0x0f0f0f0f;
519     x *= 0x01010101;
520 root 1.1
521 root 1.35 return x >> 24;
522     }
523 root 1.49
524 root 1.151 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
525     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
526 root 1.49 {
527 root 1.171 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
528 root 1.172 unsigned long r;
529 root 1.171 _BitScanReverse (&r, x);
530     return (int)r;
531     #else
532 root 1.50 int r = 0;
533 root 1.49
534 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
535     if (x >> 8) { x >>= 8; r += 8; }
536     if (x >> 4) { x >>= 4; r += 4; }
537     if (x >> 2) { x >>= 2; r += 2; }
538     if (x >> 1) { r += 1; }
539 root 1.49
540     return r;
541 root 1.171 #endif
542 root 1.49 }
543    
544 root 1.151 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
545     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
546 root 1.49 {
547 root 1.171 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
548 root 1.172 unsigned long r;
549 root 1.171 _BitScanReverse64 (&r, x);
550     return (int)r;
551     #else
552 root 1.50 int r = 0;
553 root 1.49
554 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
555 root 1.49
556 root 1.50 return r + ecb_ld32 (x);
557 root 1.171 #endif
558 root 1.49 }
559 root 1.1 #endif
560    
561 root 1.151 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
562     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
563     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
564     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
565 root 1.88
566 root 1.151 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
567     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
568 root 1.70 {
569     return ( (x * 0x0802U & 0x22110U)
570 root 1.151 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
571 root 1.70 }
572    
573 root 1.151 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
574     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
575 root 1.70 {
576     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
577     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
578     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
579     x = ( x >> 8 ) | ( x << 8);
580    
581     return x;
582     }
583    
584 root 1.151 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
585     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
586 root 1.70 {
587     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
588     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
589     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
590     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
591     x = ( x >> 16 ) | ( x << 16);
592    
593     return x;
594     }
595    
596 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
597     /* so for this version we are lazy */
598 root 1.151 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
599     ecb_function_ ecb_const int
600 root 1.49 ecb_popcount64 (uint64_t x)
601     {
602     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
603     }
604    
605 root 1.151 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
606     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
607     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
608     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
609     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
610     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
611     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
612     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
613    
614     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
615     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
616     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
617     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
618     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
619     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
620     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
621     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
622 root 1.50
623 root 1.182 #if ECB_CPP
624    
625     inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
626     inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
627     inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
628     inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
629    
630     inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
631     inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
632     inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
633     inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
634    
635     inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
636     inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
637     inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
638     inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
639    
640     inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
641     inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
642     inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
643     inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
644    
645     inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
646     inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
647     inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
648    
649 root 1.183 inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
650     inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
651     inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
652     inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
653    
654     inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
655     inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
656     inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
657     inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
658 root 1.182
659     #endif
660    
661 sf-exg 1.138 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
662 root 1.164 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
663     #define ecb_bswap16(x) __builtin_bswap16 (x)
664     #else
665 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
666 root 1.164 #endif
667 root 1.49 #define ecb_bswap32(x) __builtin_bswap32 (x)
668     #define ecb_bswap64(x) __builtin_bswap64 (x)
669 root 1.164 #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)))
674 root 1.13 #else
675 root 1.151 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
676     ecb_function_ ecb_const uint16_t
677 root 1.50 ecb_bswap16 (uint16_t x)
678 root 1.49 {
679 root 1.50 return ecb_rotl16 (x, 8);
680 root 1.49 }
681    
682 root 1.151 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
683     ecb_function_ ecb_const uint32_t
684 root 1.35 ecb_bswap32 (uint32_t x)
685     {
686 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
687 root 1.35 }
688    
689 root 1.151 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
690     ecb_function_ ecb_const uint64_t
691 root 1.49 ecb_bswap64 (uint64_t x)
692 root 1.35 {
693 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
694 root 1.35 }
695 root 1.13 #endif
696    
697 sf-exg 1.138 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
698 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
699 root 1.13 #else
700 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
701 root 1.151 ecb_inline ecb_noreturn void ecb_unreachable (void);
702     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
703 root 1.13 #endif
704    
705 root 1.41 /* try to tell the compiler that some condition is definitely true */
706 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
707 root 1.41
708 root 1.174 ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
709     ecb_inline ecb_const uint32_t
710 root 1.23 ecb_byteorder_helper (void)
711 root 1.3 {
712 root 1.98 /* the union code still generates code under pressure in gcc, */
713 sf-exg 1.111 /* but less than using pointers, and always seems to */
714 root 1.98 /* successfully return a constant. */
715     /* the reason why we have this horrible preprocessor mess */
716 root 1.99 /* is to avoid it in all cases, at least on common architectures */
717 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
718 root 1.174 #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
721     return 0x44332211;
722     #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
723     || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
724     #define ECB_BIG_ENDIAN 1
725     return 0x11223344;
726 root 1.98 #else
727     union
728     {
729 root 1.174 uint8_t c[4];
730     uint32_t u;
731     } u = { 0x11, 0x22, 0x33, 0x44 };
732     return u.u;
733 root 1.98 #endif
734 root 1.3 }
735    
736 root 1.151 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
737 root 1.174 ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
738 root 1.151 ecb_inline ecb_const ecb_bool ecb_little_endian (void);
739 root 1.174 ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
740 root 1.3
741 root 1.180 /*****************************************************************************/
742     /* unaligned load/store */
743    
744     ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
745     ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
746     ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
747    
748     ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
749     ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
750     ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
751    
752     ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753     ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
754     ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
755    
756     ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
757     ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
758     ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
759    
760     ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
761     ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
762     ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
763    
764     ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
765     ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
766     ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
767    
768     ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
769     ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
770     ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
771    
772     ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
773     ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
774     ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
775    
776     ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
777     ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
778     ecb_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    
780     ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
781     ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
782     ecb_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 root 1.186 #if ECB_CPP
785 root 1.180
786     inline uint8_t ecb_bswap (uint8_t v) { return v; }
787     inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
788     inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
789     inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
790    
791     template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
792     template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
793     template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
794     template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
795     template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
796 root 1.184 template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
797 root 1.180 template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
798     template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
799    
800     template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
801     template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
802     template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
803     template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
804     template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
805 root 1.184 template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
806 root 1.180 template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
807     template<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 root 1.188 /* division */
813 root 1.180
814 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
815 root 1.188 /* C99 tightened the definition of %, so we can use a more efficient version */
816 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
817 root 1.31 #else
818 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
819 root 1.31 #endif
820 root 1.21
821 root 1.149 #if ECB_CPP
822 sf-exg 1.68 template<typename T>
823     static inline T ecb_div_rd (T val, T div)
824     {
825     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
826     }
827     template<typename T>
828     static inline T ecb_div_ru (T val, T div)
829     {
830     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
831     }
832     #else
833     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
834     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
835     #endif
836 sf-exg 1.67
837 root 1.188 /*****************************************************************************/
838     /* array length */
839    
840 root 1.5 #if ecb_cplusplus_does_not_suck
841 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
842 root 1.35 template<typename T, int N>
843     static inline int ecb_array_length (const T (&arr)[N])
844     {
845     return N;
846     }
847 root 1.5 #else
848 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
849 root 1.5 #endif
850    
851 root 1.180 /*****************************************************************************/
852 root 1.188 /* IEEE 754-2008 half float conversions */
853 root 1.180
854 root 1.170 ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
855 root 1.167 ecb_function_ ecb_const uint32_t
856 root 1.170 ecb_binary16_to_binary32 (uint32_t x)
857 root 1.167 {
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    
886     ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
887     ecb_function_ ecb_const uint16_t
888     ecb_binary32_to_binary16 (uint32_t x)
889     {
890     unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
891 root 1.188 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
892 root 1.167 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 root 1.169 /* handle zero, subnormals and small numbers */
917 root 1.167 if (ecb_expect_true (x < 0x38800000))
918     {
919     /* zero */
920     if (ecb_expect_true (!x))
921     return s;
922    
923     /* handle subnormals */
924    
925 root 1.169 /* too small, will be zero */
926     if (e < (14 - 24)) /* might not be sharp, but is good enough */
927     return s;
928    
929 root 1.167 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    
950 root 1.104 /*******************************************************************************/
951     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
952    
953     /* basically, everything uses "ieee pure-endian" floating point numbers */
954     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
955     #if 0 \
956     || __i386 || __i386__ \
957 sf-exg 1.159 || ECB_GCC_AMD64 \
958 root 1.104 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
959     || defined __s390__ || defined __s390x__ \
960     || defined __mips__ \
961     || defined __alpha__ \
962     || defined __hppa__ \
963     || defined __ia64__ \
964 root 1.117 || defined __m68k__ \
965     || defined __m88k__ \
966     || defined __sh__ \
967 sf-exg 1.159 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
968 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
969 root 1.132 || defined __aarch64__
970 root 1.104 #define ECB_STDFP 1
971 root 1.102 #else
972 root 1.104 #define ECB_STDFP 0
973 root 1.102 #endif
974    
975 root 1.104 #ifndef ECB_NO_LIBM
976 root 1.103
977 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
978    
979 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
980     #ifdef INFINITY
981     #define ECB_INFINITY INFINITY
982     #else
983     #define ECB_INFINITY HUGE_VAL
984     #endif
985    
986     #ifdef NAN
987 root 1.121 #define ECB_NAN NAN
988     #else
989 root 1.122 #define ECB_NAN ECB_INFINITY
990 root 1.121 #endif
991 root 1.120
992 root 1.148 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
993 root 1.150 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
994 sf-exg 1.163 #define ecb_frexpf(x,e) frexpf ((x), (e))
995 root 1.148 #else
996 sf-exg 1.161 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
997 sf-exg 1.163 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
998 root 1.148 #endif
999    
1000 root 1.104 /* convert a float to ieee single/binary32 */
1001 root 1.151 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1002     ecb_function_ ecb_const uint32_t
1003 root 1.103 ecb_float_to_binary32 (float x)
1004     {
1005     uint32_t r;
1006    
1007     #if ECB_STDFP
1008 root 1.104 memcpy (&r, &x, 4);
1009 root 1.103 #else
1010 root 1.105 /* slow emulation, works for anything but -0 */
1011 root 1.103 uint32_t m;
1012     int e;
1013    
1014 root 1.108 if (x == 0e0f ) return 0x00000000U;
1015 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1016     if (x < -3.40282346638528860e+38f) return 0xff800000U;
1017 root 1.105 if (x != x ) return 0x7fbfffffU;
1018 root 1.103
1019 sf-exg 1.163 m = ecb_frexpf (x, &e) * 0x1000000U;
1020 root 1.103
1021     r = m & 0x80000000U;
1022    
1023     if (r)
1024     m = -m;
1025    
1026 root 1.108 if (e <= -126)
1027 root 1.103 {
1028     m &= 0xffffffU;
1029     m >>= (-125 - e);
1030     e = -126;
1031     }
1032    
1033     r |= (e + 126) << 23;
1034     r |= m & 0x7fffffU;
1035     #endif
1036    
1037     return r;
1038     }
1039    
1040 root 1.104 /* converts an ieee single/binary32 to a float */
1041 root 1.151 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1042     ecb_function_ ecb_const float
1043 root 1.103 ecb_binary32_to_float (uint32_t x)
1044     {
1045     float r;
1046    
1047     #if ECB_STDFP
1048 root 1.104 memcpy (&r, &x, 4);
1049 root 1.103 #else
1050     /* emulation, only works for normals and subnormals and +0 */
1051     int neg = x >> 31;
1052     int e = (x >> 23) & 0xffU;
1053    
1054     x &= 0x7fffffU;
1055    
1056     if (e)
1057     x |= 0x800000U;
1058 root 1.104 else
1059     e = 1;
1060 root 1.103
1061     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1062 root 1.148 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1063 root 1.103
1064     r = neg ? -r : r;
1065     #endif
1066    
1067     return r;
1068     }
1069    
1070 root 1.104 /* convert a double to ieee double/binary64 */
1071 root 1.151 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1072     ecb_function_ ecb_const uint64_t
1073 root 1.103 ecb_double_to_binary64 (double x)
1074     {
1075 root 1.104 uint64_t r;
1076    
1077     #if ECB_STDFP
1078     memcpy (&r, &x, 8);
1079     #else
1080 root 1.105 /* slow emulation, works for anything but -0 */
1081 root 1.104 uint64_t m;
1082     int e;
1083    
1084 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
1085 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1086     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1087 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
1088 root 1.104
1089     m = frexp (x, &e) * 0x20000000000000U;
1090    
1091     r = m & 0x8000000000000000;;
1092    
1093     if (r)
1094     m = -m;
1095    
1096 root 1.108 if (e <= -1022)
1097 root 1.104 {
1098     m &= 0x1fffffffffffffU;
1099     m >>= (-1021 - e);
1100     e = -1022;
1101     }
1102    
1103     r |= ((uint64_t)(e + 1022)) << 52;
1104     r |= m & 0xfffffffffffffU;
1105     #endif
1106    
1107     return r;
1108     }
1109    
1110     /* converts an ieee double/binary64 to a double */
1111 root 1.151 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1112     ecb_function_ ecb_const double
1113 root 1.104 ecb_binary64_to_double (uint64_t x)
1114     {
1115     double r;
1116    
1117     #if ECB_STDFP
1118     memcpy (&r, &x, 8);
1119     #else
1120     /* emulation, only works for normals and subnormals and +0 */
1121     int neg = x >> 63;
1122     int e = (x >> 52) & 0x7ffU;
1123    
1124     x &= 0xfffffffffffffU;
1125    
1126     if (e)
1127     x |= 0x10000000000000U;
1128     else
1129     e = 1;
1130    
1131 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1132 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1133 root 1.104
1134     r = neg ? -r : r;
1135     #endif
1136    
1137     return r;
1138 root 1.103 }
1139    
1140 root 1.167 /* 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     }
1147    
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     }
1155    
1156 root 1.103 #endif
1157 root 1.102
1158 root 1.1 #endif
1159