ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.164
Committed: Mon Apr 20 20:06:30 2015 UTC (9 years, 2 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.163: +9 -0 lines
Log Message:
msc bswap, gcc bswap16

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 1.152 * Copyright (©) 2009-2015 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 1.141 #define ECB_VERSION 0x00010004
46 root 1.87
47 root 1.44 #ifdef _WIN32
48     typedef signed char int8_t;
49     typedef unsigned char uint8_t;
50     typedef signed short int16_t;
51     typedef unsigned short uint16_t;
52     typedef signed int int32_t;
53     typedef unsigned int uint32_t;
54     #if __GNUC__
55     typedef signed long long int64_t;
56     typedef unsigned long long uint64_t;
57 root 1.51 #else /* _MSC_VER || __BORLANDC__ */
58 root 1.44 typedef signed __int64 int64_t;
59     typedef unsigned __int64 uint64_t;
60     #endif
61 root 1.87 #ifdef _WIN64
62     #define ECB_PTRSIZE 8
63     typedef uint64_t uintptr_t;
64     typedef int64_t intptr_t;
65     #else
66     #define ECB_PTRSIZE 4
67     typedef uint32_t uintptr_t;
68     typedef int32_t intptr_t;
69     #endif
70 root 1.44 #else
71     #include <inttypes.h>
72 root 1.87 #if UINTMAX_MAX > 0xffffffffU
73     #define ECB_PTRSIZE 8
74     #else
75     #define ECB_PTRSIZE 4
76     #endif
77 root 1.44 #endif
78 root 1.6
79 sf-exg 1.159 #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
80     #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
81    
82 root 1.114 /* work around x32 idiocy by defining proper macros */
83 sf-exg 1.159 #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
84 root 1.119 #if _ILP32
85 root 1.115 #define ECB_AMD64_X32 1
86 root 1.114 #else
87 root 1.115 #define ECB_AMD64 1
88 root 1.114 #endif
89     #endif
90    
91 root 1.12 /* many compilers define _GNUC_ to some versions but then only implement
92     * what their idiot authors think are the "more important" extensions,
93 sf-exg 1.59 * causing enormous grief in return for some better fake benchmark numbers.
94 root 1.18 * or so.
95 root 1.12 * we try to detect these and simply assume they are not gcc - if they have
96     * an issue with that they should have done it right in the first place.
97     */
98 root 1.137 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
99     #define ECB_GCC_VERSION(major,minor) 0
100     #else
101     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
102 root 1.12 #endif
103 root 1.1
104 sf-exg 1.138 #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
105    
106 root 1.147 #if __clang__ && defined __has_builtin
107     #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
108 sf-exg 1.138 #else
109     #define ECB_CLANG_BUILTIN(x) 0
110     #endif
111    
112 root 1.147 #if __clang__ && defined __has_extension
113     #define ECB_CLANG_EXTENSION(x) __has_extension (x)
114 sf-exg 1.140 #else
115     #define ECB_CLANG_EXTENSION(x) 0
116     #endif
117    
118 root 1.91 #define ECB_CPP (__cplusplus+0)
119     #define ECB_CPP11 (__cplusplus >= 201103L)
120 root 1.90
121 root 1.102 #if ECB_CPP
122 root 1.127 #define ECB_C 0
123     #define ECB_STDC_VERSION 0
124     #else
125     #define ECB_C 1
126     #define ECB_STDC_VERSION __STDC_VERSION__
127     #endif
128    
129     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
130     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
131    
132     #if ECB_CPP
133 root 1.102 #define ECB_EXTERN_C extern "C"
134     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
135     #define ECB_EXTERN_C_END }
136     #else
137     #define ECB_EXTERN_C extern
138     #define ECB_EXTERN_C_BEG
139     #define ECB_EXTERN_C_END
140     #endif
141    
142 root 1.52 /*****************************************************************************/
143    
144 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
145     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
146    
147 root 1.79 #if ECB_NO_THREADS
148 root 1.95 #define ECB_NO_SMP 1
149 root 1.79 #endif
150    
151 root 1.93 #if ECB_NO_SMP
152 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
153 root 1.58 #endif
154    
155 root 1.52 #ifndef ECB_MEMORY_FENCE
156 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
157 root 1.73 #if __i386 || __i386__
158 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
159 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
160     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
161 sf-exg 1.159 #elif ECB_GCC_AMD64
162 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
163     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
164     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
165 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
166 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
167 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
168     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
169 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
170 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
171 root 1.94 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
172     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
173 root 1.129 #elif __aarch64__
174     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
175 root 1.124 #elif (__sparc || __sparc__) && !__sparcv8
176 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
177     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
178     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
179 root 1.85 #elif defined __s390__ || defined __s390x__
180 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
181 root 1.85 #elif defined __mips__
182 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
183 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
184     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
185 root 1.86 #elif defined __alpha__
186 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
187     #elif defined __hppa__
188     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
189     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
190     #elif defined __ia64__
191     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
192 root 1.117 #elif defined __m68k__
193     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
194     #elif defined __m88k__
195     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
196     #elif defined __sh__
197     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
198 root 1.52 #endif
199     #endif
200     #endif
201    
202     #ifndef ECB_MEMORY_FENCE
203 root 1.93 #if ECB_GCC_VERSION(4,7)
204 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
205 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
206 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
207     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
208 root 1.110
209 sf-exg 1.140 #elif ECB_CLANG_EXTENSION(c_atomic)
210     /* see comment below (stdatomic.h) about the C11 memory model. */
211     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
212     #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
213     #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
214 root 1.110
215 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
216 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
217 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
218     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
219     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
220     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
221     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
222     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
223 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
224     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
225     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
226     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
227     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
228 root 1.85 #elif defined _WIN32
229 root 1.55 #include <WinNT.h>
230 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
231 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
232     #include <mbarrier.h>
233     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
234     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
235     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
236 root 1.82 #elif __xlC__
237 root 1.83 #define ECB_MEMORY_FENCE __sync ()
238 root 1.52 #endif
239     #endif
240    
241 root 1.53 #ifndef ECB_MEMORY_FENCE
242 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
243     /* we assume that these memory fences work on all variables/all memory accesses, */
244     /* not just C11 atomics and atomic accesses */
245     #include <stdatomic.h>
246 root 1.97 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
247     /* any fence other than seq_cst, which isn't very efficient for us. */
248     /* Why that is, we don't know - either the C11 memory model is quite useless */
249     /* for most usages, or gcc and clang have a bug */
250     /* I *currently* lean towards the latter, and inefficiently implement */
251     /* all three of ecb's fences as a seq_cst fence */
252 root 1.128 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
253     /* for all __atomic_thread_fence's except seq_cst */
254 root 1.94 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
255     #endif
256     #endif
257    
258     #ifndef ECB_MEMORY_FENCE
259 root 1.62 #if !ECB_AVOID_PTHREADS
260     /*
261     * if you get undefined symbol references to pthread_mutex_lock,
262     * or failure to find pthread.h, then you should implement
263     * the ECB_MEMORY_FENCE operations for your cpu/compiler
264     * OR provide pthread.h and link against the posix thread library
265     * of your system.
266     */
267     #include <pthread.h>
268     #define ECB_NEEDS_PTHREADS 1
269     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
270 root 1.52
271 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
272     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
273     #endif
274     #endif
275    
276 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
277 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
278 root 1.62 #endif
279    
280 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
281 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
282     #endif
283    
284     /*****************************************************************************/
285    
286 root 1.149 #if ECB_CPP
287 root 1.46 #define ecb_inline static inline
288 root 1.38 #elif ECB_GCC_VERSION(2,5)
289 root 1.46 #define ecb_inline static __inline__
290 root 1.39 #elif ECB_C99
291 root 1.46 #define ecb_inline static inline
292 root 1.29 #else
293 root 1.46 #define ecb_inline static
294 root 1.38 #endif
295    
296     #if ECB_GCC_VERSION(3,3)
297     #define ecb_restrict __restrict__
298 root 1.39 #elif ECB_C99
299 root 1.38 #define ecb_restrict restrict
300     #else
301     #define ecb_restrict
302 root 1.4 #endif
303    
304 root 1.38 typedef int ecb_bool;
305    
306 root 1.8 #define ECB_CONCAT_(a, b) a ## b
307     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
308     #define ECB_STRINGIFY_(a) # a
309     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
310 root 1.155 #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
311 root 1.8
312 root 1.46 #define ecb_function_ ecb_inline
313 root 1.3
314 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
315 root 1.142 #define ecb_attribute(attrlist) __attribute__ (attrlist)
316 root 1.37 #else
317     #define ecb_attribute(attrlist)
318 sf-exg 1.138 #endif
319 root 1.127
320 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
321     #define ecb_is_constant(expr) __builtin_constant_p (expr)
322     #else
323 root 1.127 /* possible C11 impl for integral types
324     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
325     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
326    
327 root 1.37 #define ecb_is_constant(expr) 0
328 sf-exg 1.138 #endif
329    
330     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
331     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
332     #else
333 root 1.37 #define ecb_expect(expr,value) (expr)
334 sf-exg 1.138 #endif
335    
336     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
337     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
338     #else
339 root 1.37 #define ecb_prefetch(addr,rw,locality)
340 root 1.1 #endif
341    
342 root 1.2 /* no emulation for ecb_decltype */
343 root 1.143 #if ECB_CPP11
344 root 1.144 // older implementations might have problems with decltype(x)::type, work around it
345 root 1.146 template<class T> struct ecb_decltype_t { typedef T type; };
346     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
347 root 1.143 #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
348     #define ecb_decltype(x) __typeof__ (x)
349 root 1.1 #endif
350    
351 root 1.135 #if _MSC_VER >= 1300
352 root 1.149 #define ecb_deprecated __declspec (deprecated)
353 root 1.135 #else
354     #define ecb_deprecated ecb_attribute ((__deprecated__))
355     #endif
356    
357 sf-exg 1.162 #if _MSC_VER >= 1500
358 root 1.154 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
359     #elif ECB_GCC_VERSION(4,5)
360     #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
361     #else
362     #define ecb_deprecated_message(msg) ecb_deprecated
363     #endif
364    
365     #if _MSC_VER >= 1400
366     #define ecb_noinline __declspec (noinline)
367     #else
368     #define ecb_noinline ecb_attribute ((__noinline__))
369     #endif
370    
371 root 1.24 #define ecb_unused ecb_attribute ((__unused__))
372     #define ecb_const ecb_attribute ((__const__))
373     #define ecb_pure ecb_attribute ((__pure__))
374 root 1.35
375 root 1.145 #if ECB_C11 || __IBMC_NORETURN
376     /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */
377 root 1.90 #define ecb_noreturn _Noreturn
378 root 1.153 #elif ECB_CPP11
379     #define ecb_noreturn [[noreturn]]
380     #elif _MSC_VER >= 1200
381 sf-exg 1.156 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
382 root 1.153 #define ecb_noreturn __declspec (noreturn)
383 root 1.90 #else
384     #define ecb_noreturn ecb_attribute ((__noreturn__))
385     #endif
386    
387 root 1.35 #if ECB_GCC_VERSION(4,3)
388 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
389     #define ecb_hot ecb_attribute ((__hot__))
390     #define ecb_cold ecb_attribute ((__cold__))
391 root 1.35 #else
392     #define ecb_artificial
393     #define ecb_hot
394     #define ecb_cold
395     #endif
396 root 1.1
397 root 1.39 /* put around conditional expressions if you are very sure that the */
398     /* expression is mostly true or mostly false. note that these return */
399     /* booleans, not the expression. */
400 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
401     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
402 root 1.36 /* for compatibility to the rest of the world */
403 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
404     #define ecb_unlikely(expr) ecb_expect_false (expr)
405 root 1.1
406 root 1.3 /* count trailing zero bits and count # of one bits */
407 root 1.139 #if ECB_GCC_VERSION(3,4) \
408     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
409     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
410     && ECB_CLANG_BUILTIN(__builtin_popcount))
411 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
412     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
413     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
414 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
415 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
416 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
417 root 1.49 /* no popcountll */
418 root 1.1 #else
419 root 1.151 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
420     ecb_function_ ecb_const int
421 root 1.35 ecb_ctz32 (uint32_t x)
422     {
423     int r = 0;
424    
425 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
426 root 1.35
427 root 1.50 #if ECB_branchless_on_i386
428     r += !!(x & 0xaaaaaaaa) << 0;
429     r += !!(x & 0xcccccccc) << 1;
430     r += !!(x & 0xf0f0f0f0) << 2;
431     r += !!(x & 0xff00ff00) << 3;
432     r += !!(x & 0xffff0000) << 4;
433     #else
434 root 1.35 if (x & 0xaaaaaaaa) r += 1;
435     if (x & 0xcccccccc) r += 2;
436     if (x & 0xf0f0f0f0) r += 4;
437     if (x & 0xff00ff00) r += 8;
438     if (x & 0xffff0000) r += 16;
439 root 1.50 #endif
440 root 1.35
441     return r;
442     }
443    
444 root 1.151 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
445     ecb_function_ ecb_const int
446 root 1.49 ecb_ctz64 (uint64_t x)
447     {
448     int shift = x & 0xffffffffU ? 0 : 32;
449 root 1.50 return ecb_ctz32 (x >> shift) + shift;
450 root 1.49 }
451    
452 root 1.151 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
453     ecb_function_ ecb_const int
454 root 1.35 ecb_popcount32 (uint32_t x)
455     {
456     x -= (x >> 1) & 0x55555555;
457     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
458     x = ((x >> 4) + x) & 0x0f0f0f0f;
459     x *= 0x01010101;
460 root 1.1
461 root 1.35 return x >> 24;
462     }
463 root 1.49
464 root 1.151 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
465     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
466 root 1.49 {
467 root 1.50 int r = 0;
468 root 1.49
469 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
470     if (x >> 8) { x >>= 8; r += 8; }
471     if (x >> 4) { x >>= 4; r += 4; }
472     if (x >> 2) { x >>= 2; r += 2; }
473     if (x >> 1) { r += 1; }
474 root 1.49
475     return r;
476     }
477    
478 root 1.151 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
479     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
480 root 1.49 {
481 root 1.50 int r = 0;
482 root 1.49
483 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
484 root 1.49
485 root 1.50 return r + ecb_ld32 (x);
486 root 1.49 }
487 root 1.1 #endif
488    
489 root 1.151 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
490     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
491     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
492     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
493 root 1.88
494 root 1.151 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
495     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
496 root 1.70 {
497     return ( (x * 0x0802U & 0x22110U)
498 root 1.151 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
499 root 1.70 }
500    
501 root 1.151 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
502     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
503 root 1.70 {
504     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
505     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
506     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
507     x = ( x >> 8 ) | ( x << 8);
508    
509     return x;
510     }
511    
512 root 1.151 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
513     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
514 root 1.70 {
515     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
516     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
517     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
518     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
519     x = ( x >> 16 ) | ( x << 16);
520    
521     return x;
522     }
523    
524 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
525     /* so for this version we are lazy */
526 root 1.151 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
527     ecb_function_ ecb_const int
528 root 1.49 ecb_popcount64 (uint64_t x)
529     {
530     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
531     }
532    
533 root 1.151 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
534     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
535     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
536     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
537     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
538     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
539     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
540     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
541    
542     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
543     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
544     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
545     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
546     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
547     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
548     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
549     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
550 root 1.50
551 sf-exg 1.138 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
552 root 1.164 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
553     #define ecb_bswap16(x) __builtin_bswap16 (x)
554     #else
555 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
556 root 1.164 #endif
557 root 1.49 #define ecb_bswap32(x) __builtin_bswap32 (x)
558     #define ecb_bswap64(x) __builtin_bswap64 (x)
559 root 1.164 #elif _MSC_VER
560     #include <stdlib.h>
561     #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
562     #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
563     #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
564 root 1.13 #else
565 root 1.151 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
566     ecb_function_ ecb_const uint16_t
567 root 1.50 ecb_bswap16 (uint16_t x)
568 root 1.49 {
569 root 1.50 return ecb_rotl16 (x, 8);
570 root 1.49 }
571    
572 root 1.151 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
573     ecb_function_ ecb_const uint32_t
574 root 1.35 ecb_bswap32 (uint32_t x)
575     {
576 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
577 root 1.35 }
578    
579 root 1.151 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
580     ecb_function_ ecb_const uint64_t
581 root 1.49 ecb_bswap64 (uint64_t x)
582 root 1.35 {
583 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
584 root 1.35 }
585 root 1.13 #endif
586    
587 sf-exg 1.138 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
588 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
589 root 1.13 #else
590 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
591 root 1.151 ecb_inline ecb_noreturn void ecb_unreachable (void);
592     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
593 root 1.13 #endif
594    
595 root 1.41 /* try to tell the compiler that some condition is definitely true */
596 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
597 root 1.41
598 root 1.151 ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);
599     ecb_inline ecb_const unsigned char
600 root 1.23 ecb_byteorder_helper (void)
601 root 1.3 {
602 root 1.98 /* the union code still generates code under pressure in gcc, */
603 sf-exg 1.111 /* but less than using pointers, and always seems to */
604 root 1.98 /* successfully return a constant. */
605     /* the reason why we have this horrible preprocessor mess */
606 root 1.99 /* is to avoid it in all cases, at least on common architectures */
607 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
608 sf-exg 1.159 #if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64
609 root 1.98 return 0x44;
610     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
611     return 0x44;
612     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
613 root 1.113 return 0x11;
614 root 1.98 #else
615     union
616     {
617     uint32_t i;
618     uint8_t c;
619     } u = { 0x11223344 };
620     return u.c;
621     #endif
622 root 1.3 }
623    
624 root 1.151 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
625     ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
626     ecb_inline ecb_const ecb_bool ecb_little_endian (void);
627     ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
628 root 1.3
629 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
630 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
631 root 1.31 #else
632 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
633 root 1.31 #endif
634 root 1.21
635 root 1.149 #if ECB_CPP
636 sf-exg 1.68 template<typename T>
637     static inline T ecb_div_rd (T val, T div)
638     {
639     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
640     }
641     template<typename T>
642     static inline T ecb_div_ru (T val, T div)
643     {
644     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
645     }
646     #else
647     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
648     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
649     #endif
650 sf-exg 1.67
651 root 1.5 #if ecb_cplusplus_does_not_suck
652 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
653 root 1.35 template<typename T, int N>
654     static inline int ecb_array_length (const T (&arr)[N])
655     {
656     return N;
657     }
658 root 1.5 #else
659 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
660 root 1.5 #endif
661    
662 root 1.104 /*******************************************************************************/
663     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
664    
665     /* basically, everything uses "ieee pure-endian" floating point numbers */
666     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
667     #if 0 \
668     || __i386 || __i386__ \
669 sf-exg 1.159 || ECB_GCC_AMD64 \
670 root 1.104 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
671     || defined __s390__ || defined __s390x__ \
672     || defined __mips__ \
673     || defined __alpha__ \
674     || defined __hppa__ \
675     || defined __ia64__ \
676 root 1.117 || defined __m68k__ \
677     || defined __m88k__ \
678     || defined __sh__ \
679 sf-exg 1.159 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
680 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
681 root 1.132 || defined __aarch64__
682 root 1.104 #define ECB_STDFP 1
683     #include <string.h> /* for memcpy */
684 root 1.102 #else
685 root 1.104 #define ECB_STDFP 0
686 root 1.102 #endif
687    
688 root 1.104 #ifndef ECB_NO_LIBM
689 root 1.103
690 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
691    
692 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
693     #ifdef INFINITY
694     #define ECB_INFINITY INFINITY
695     #else
696     #define ECB_INFINITY HUGE_VAL
697     #endif
698    
699     #ifdef NAN
700 root 1.121 #define ECB_NAN NAN
701     #else
702 root 1.122 #define ECB_NAN ECB_INFINITY
703 root 1.121 #endif
704 root 1.120
705 root 1.148 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
706 root 1.150 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
707 sf-exg 1.163 #define ecb_frexpf(x,e) frexpf ((x), (e))
708 root 1.148 #else
709 sf-exg 1.161 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
710 sf-exg 1.163 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
711 root 1.148 #endif
712    
713 root 1.120 /* converts an ieee half/binary16 to a float */
714 root 1.151 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
715     ecb_function_ ecb_const float
716 root 1.120 ecb_binary16_to_float (uint16_t x)
717     {
718     int e = (x >> 10) & 0x1f;
719     int m = x & 0x3ff;
720     float r;
721    
722 root 1.148 if (!e ) r = ecb_ldexpf (m , -24);
723     else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
724 root 1.121 else if (m ) r = ECB_NAN;
725 root 1.123 else r = ECB_INFINITY;
726 root 1.120
727     return x & 0x8000 ? -r : r;
728     }
729    
730 root 1.104 /* convert a float to ieee single/binary32 */
731 root 1.151 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
732     ecb_function_ ecb_const uint32_t
733 root 1.103 ecb_float_to_binary32 (float x)
734     {
735     uint32_t r;
736    
737     #if ECB_STDFP
738 root 1.104 memcpy (&r, &x, 4);
739 root 1.103 #else
740 root 1.105 /* slow emulation, works for anything but -0 */
741 root 1.103 uint32_t m;
742     int e;
743    
744 root 1.108 if (x == 0e0f ) return 0x00000000U;
745 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
746     if (x < -3.40282346638528860e+38f) return 0xff800000U;
747 root 1.105 if (x != x ) return 0x7fbfffffU;
748 root 1.103
749 sf-exg 1.163 m = ecb_frexpf (x, &e) * 0x1000000U;
750 root 1.103
751     r = m & 0x80000000U;
752    
753     if (r)
754     m = -m;
755    
756 root 1.108 if (e <= -126)
757 root 1.103 {
758     m &= 0xffffffU;
759     m >>= (-125 - e);
760     e = -126;
761     }
762    
763     r |= (e + 126) << 23;
764     r |= m & 0x7fffffU;
765     #endif
766    
767     return r;
768     }
769    
770 root 1.104 /* converts an ieee single/binary32 to a float */
771 root 1.151 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
772     ecb_function_ ecb_const float
773 root 1.103 ecb_binary32_to_float (uint32_t x)
774     {
775     float r;
776    
777     #if ECB_STDFP
778 root 1.104 memcpy (&r, &x, 4);
779 root 1.103 #else
780     /* emulation, only works for normals and subnormals and +0 */
781     int neg = x >> 31;
782     int e = (x >> 23) & 0xffU;
783    
784     x &= 0x7fffffU;
785    
786     if (e)
787     x |= 0x800000U;
788 root 1.104 else
789     e = 1;
790 root 1.103
791     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
792 root 1.148 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
793 root 1.103
794     r = neg ? -r : r;
795     #endif
796    
797     return r;
798     }
799    
800 root 1.104 /* convert a double to ieee double/binary64 */
801 root 1.151 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
802     ecb_function_ ecb_const uint64_t
803 root 1.103 ecb_double_to_binary64 (double x)
804     {
805 root 1.104 uint64_t r;
806    
807     #if ECB_STDFP
808     memcpy (&r, &x, 8);
809     #else
810 root 1.105 /* slow emulation, works for anything but -0 */
811 root 1.104 uint64_t m;
812     int e;
813    
814 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
815 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
816     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
817 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
818 root 1.104
819     m = frexp (x, &e) * 0x20000000000000U;
820    
821     r = m & 0x8000000000000000;;
822    
823     if (r)
824     m = -m;
825    
826 root 1.108 if (e <= -1022)
827 root 1.104 {
828     m &= 0x1fffffffffffffU;
829     m >>= (-1021 - e);
830     e = -1022;
831     }
832    
833     r |= ((uint64_t)(e + 1022)) << 52;
834     r |= m & 0xfffffffffffffU;
835     #endif
836    
837     return r;
838     }
839    
840     /* converts an ieee double/binary64 to a double */
841 root 1.151 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
842     ecb_function_ ecb_const double
843 root 1.104 ecb_binary64_to_double (uint64_t x)
844     {
845     double r;
846    
847     #if ECB_STDFP
848     memcpy (&r, &x, 8);
849     #else
850     /* emulation, only works for normals and subnormals and +0 */
851     int neg = x >> 63;
852     int e = (x >> 52) & 0x7ffU;
853    
854     x &= 0xfffffffffffffU;
855    
856     if (e)
857     x |= 0x10000000000000U;
858     else
859     e = 1;
860    
861 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
862 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
863 root 1.104
864     r = neg ? -r : r;
865     #endif
866    
867     return r;
868 root 1.103 }
869    
870     #endif
871 root 1.102
872 root 1.1 #endif
873