ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.146
Committed: Fri Oct 17 11:07:14 2014 UTC (9 years, 8 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.145: +2 -2 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 1.129 * Copyright (©) 2009-2014 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 root 1.114 /* work around x32 idiocy by defining proper macros */
80 root 1.125 #if __amd64 || __x86_64 || _M_AMD64 || _M_X64
81 root 1.119 #if _ILP32
82 root 1.115 #define ECB_AMD64_X32 1
83 root 1.114 #else
84 root 1.115 #define ECB_AMD64 1
85 root 1.114 #endif
86     #endif
87    
88 root 1.12 /* many compilers define _GNUC_ to some versions but then only implement
89     * what their idiot authors think are the "more important" extensions,
90 sf-exg 1.59 * causing enormous grief in return for some better fake benchmark numbers.
91 root 1.18 * or so.
92 root 1.12 * we try to detect these and simply assume they are not gcc - if they have
93     * an issue with that they should have done it right in the first place.
94     */
95 root 1.137 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
96     #define ECB_GCC_VERSION(major,minor) 0
97     #else
98     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
99 root 1.12 #endif
100 root 1.1
101 sf-exg 1.138 #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
102    
103     #if __clang__ && defined(__has_builtin)
104     #define ECB_CLANG_BUILTIN(x) __has_builtin(x)
105     #else
106     #define ECB_CLANG_BUILTIN(x) 0
107     #endif
108    
109 sf-exg 1.140 #if __clang__ && defined(__has_extension)
110     #define ECB_CLANG_EXTENSION(x) __has_extension(x)
111     #else
112     #define ECB_CLANG_EXTENSION(x) 0
113     #endif
114    
115 root 1.91 #define ECB_CPP (__cplusplus+0)
116     #define ECB_CPP11 (__cplusplus >= 201103L)
117 root 1.90
118 root 1.102 #if ECB_CPP
119 root 1.127 #define ECB_C 0
120     #define ECB_STDC_VERSION 0
121     #else
122     #define ECB_C 1
123     #define ECB_STDC_VERSION __STDC_VERSION__
124     #endif
125    
126     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
127     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
128    
129     #if ECB_CPP
130 root 1.102 #define ECB_EXTERN_C extern "C"
131     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
132     #define ECB_EXTERN_C_END }
133     #else
134     #define ECB_EXTERN_C extern
135     #define ECB_EXTERN_C_BEG
136     #define ECB_EXTERN_C_END
137     #endif
138    
139 root 1.52 /*****************************************************************************/
140    
141 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
142     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
143    
144 root 1.79 #if ECB_NO_THREADS
145 root 1.95 #define ECB_NO_SMP 1
146 root 1.79 #endif
147    
148 root 1.93 #if ECB_NO_SMP
149 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
150 root 1.58 #endif
151    
152 root 1.52 #ifndef ECB_MEMORY_FENCE
153 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
154 root 1.73 #if __i386 || __i386__
155 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
156 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
157     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
158 root 1.73 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
159 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
160     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
161     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
162 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
163 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
164 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
165     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
166 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
167 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
168 root 1.94 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
169     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
170 root 1.129 #elif __aarch64__
171     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
172 root 1.124 #elif (__sparc || __sparc__) && !__sparcv8
173 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
174     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
175     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
176 root 1.85 #elif defined __s390__ || defined __s390x__
177 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
178 root 1.85 #elif defined __mips__
179 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
180 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
181     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
182 root 1.86 #elif defined __alpha__
183 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
184     #elif defined __hppa__
185     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
186     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
187     #elif defined __ia64__
188     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
189 root 1.117 #elif defined __m68k__
190     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
191     #elif defined __m88k__
192     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
193     #elif defined __sh__
194     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
195 root 1.52 #endif
196     #endif
197     #endif
198    
199     #ifndef ECB_MEMORY_FENCE
200 root 1.93 #if ECB_GCC_VERSION(4,7)
201 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
202 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
203 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
204     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
205 root 1.110
206 sf-exg 1.140 #elif ECB_CLANG_EXTENSION(c_atomic)
207     /* see comment below (stdatomic.h) about the C11 memory model. */
208     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
209     #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
210     #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
211 root 1.110
212 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
213 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
214 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
215     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
216     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
217     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
218     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
219     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
220 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
221     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
222     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
223     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
224     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
225 root 1.85 #elif defined _WIN32
226 root 1.55 #include <WinNT.h>
227 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
228 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
229     #include <mbarrier.h>
230     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
231     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
232     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
233 root 1.82 #elif __xlC__
234 root 1.83 #define ECB_MEMORY_FENCE __sync ()
235 root 1.52 #endif
236     #endif
237    
238 root 1.53 #ifndef ECB_MEMORY_FENCE
239 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
240     /* we assume that these memory fences work on all variables/all memory accesses, */
241     /* not just C11 atomics and atomic accesses */
242     #include <stdatomic.h>
243 root 1.97 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
244     /* any fence other than seq_cst, which isn't very efficient for us. */
245     /* Why that is, we don't know - either the C11 memory model is quite useless */
246     /* for most usages, or gcc and clang have a bug */
247     /* I *currently* lean towards the latter, and inefficiently implement */
248     /* all three of ecb's fences as a seq_cst fence */
249 root 1.128 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
250     /* for all __atomic_thread_fence's except seq_cst */
251 root 1.94 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
252     #endif
253     #endif
254    
255     #ifndef ECB_MEMORY_FENCE
256 root 1.62 #if !ECB_AVOID_PTHREADS
257     /*
258     * if you get undefined symbol references to pthread_mutex_lock,
259     * or failure to find pthread.h, then you should implement
260     * the ECB_MEMORY_FENCE operations for your cpu/compiler
261     * OR provide pthread.h and link against the posix thread library
262     * of your system.
263     */
264     #include <pthread.h>
265     #define ECB_NEEDS_PTHREADS 1
266     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
267 root 1.52
268 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
269     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
270     #endif
271     #endif
272    
273 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
274 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
275 root 1.62 #endif
276    
277 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
278 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
279     #endif
280    
281     /*****************************************************************************/
282    
283 root 1.39 #if __cplusplus
284 root 1.46 #define ecb_inline static inline
285 root 1.38 #elif ECB_GCC_VERSION(2,5)
286 root 1.46 #define ecb_inline static __inline__
287 root 1.39 #elif ECB_C99
288 root 1.46 #define ecb_inline static inline
289 root 1.29 #else
290 root 1.46 #define ecb_inline static
291 root 1.38 #endif
292    
293     #if ECB_GCC_VERSION(3,3)
294     #define ecb_restrict __restrict__
295 root 1.39 #elif ECB_C99
296 root 1.38 #define ecb_restrict restrict
297     #else
298     #define ecb_restrict
299 root 1.4 #endif
300    
301 root 1.38 typedef int ecb_bool;
302    
303 root 1.8 #define ECB_CONCAT_(a, b) a ## b
304     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
305     #define ECB_STRINGIFY_(a) # a
306     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
307    
308 root 1.46 #define ecb_function_ ecb_inline
309 root 1.3
310 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
311 root 1.142 #define ecb_attribute(attrlist) __attribute__ (attrlist)
312 root 1.37 #else
313     #define ecb_attribute(attrlist)
314 sf-exg 1.138 #endif
315 root 1.127
316 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
317     #define ecb_is_constant(expr) __builtin_constant_p (expr)
318     #else
319 root 1.127 /* possible C11 impl for integral types
320     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
321     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
322    
323 root 1.37 #define ecb_is_constant(expr) 0
324 sf-exg 1.138 #endif
325    
326     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
327     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
328     #else
329 root 1.37 #define ecb_expect(expr,value) (expr)
330 sf-exg 1.138 #endif
331    
332     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
333     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
334     #else
335 root 1.37 #define ecb_prefetch(addr,rw,locality)
336 root 1.1 #endif
337    
338 root 1.2 /* no emulation for ecb_decltype */
339 root 1.143 #if ECB_CPP11
340 root 1.144 // older implementations might have problems with decltype(x)::type, work around it
341 root 1.146 template<class T> struct ecb_decltype_t { typedef T type; };
342     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
343 root 1.143 #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
344     #define ecb_decltype(x) __typeof__ (x)
345 root 1.1 #endif
346    
347 root 1.135 #if _MSC_VER >= 1300
348     #define ecb_deprecated __declspec(deprecated)
349     #else
350     #define ecb_deprecated ecb_attribute ((__deprecated__))
351     #endif
352    
353 root 1.24 #define ecb_noinline ecb_attribute ((__noinline__))
354     #define ecb_unused ecb_attribute ((__unused__))
355     #define ecb_const ecb_attribute ((__const__))
356     #define ecb_pure ecb_attribute ((__pure__))
357 root 1.35
358 root 1.145 /* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
359     #if ECB_C11 || __IBMC_NORETURN
360     /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */
361 root 1.90 #define ecb_noreturn _Noreturn
362     #else
363     #define ecb_noreturn ecb_attribute ((__noreturn__))
364     #endif
365    
366 root 1.35 #if ECB_GCC_VERSION(4,3)
367 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
368     #define ecb_hot ecb_attribute ((__hot__))
369     #define ecb_cold ecb_attribute ((__cold__))
370 root 1.35 #else
371     #define ecb_artificial
372     #define ecb_hot
373     #define ecb_cold
374     #endif
375 root 1.1
376 root 1.39 /* put around conditional expressions if you are very sure that the */
377     /* expression is mostly true or mostly false. note that these return */
378     /* booleans, not the expression. */
379 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
380     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
381 root 1.36 /* for compatibility to the rest of the world */
382 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
383     #define ecb_unlikely(expr) ecb_expect_false (expr)
384 root 1.1
385 root 1.3 /* count trailing zero bits and count # of one bits */
386 root 1.139 #if ECB_GCC_VERSION(3,4) \
387     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
388     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
389     && ECB_CLANG_BUILTIN(__builtin_popcount))
390 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
391     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
392     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
393 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
394 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
395 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
396 root 1.49 /* no popcountll */
397 root 1.1 #else
398 root 1.39 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
399     ecb_function_ int
400 root 1.35 ecb_ctz32 (uint32_t x)
401     {
402     int r = 0;
403    
404 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
405 root 1.35
406 root 1.50 #if ECB_branchless_on_i386
407     r += !!(x & 0xaaaaaaaa) << 0;
408     r += !!(x & 0xcccccccc) << 1;
409     r += !!(x & 0xf0f0f0f0) << 2;
410     r += !!(x & 0xff00ff00) << 3;
411     r += !!(x & 0xffff0000) << 4;
412     #else
413 root 1.35 if (x & 0xaaaaaaaa) r += 1;
414     if (x & 0xcccccccc) r += 2;
415     if (x & 0xf0f0f0f0) r += 4;
416     if (x & 0xff00ff00) r += 8;
417     if (x & 0xffff0000) r += 16;
418 root 1.50 #endif
419 root 1.35
420     return r;
421     }
422    
423 root 1.49 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
424     ecb_function_ int
425     ecb_ctz64 (uint64_t x)
426     {
427     int shift = x & 0xffffffffU ? 0 : 32;
428 root 1.50 return ecb_ctz32 (x >> shift) + shift;
429 root 1.49 }
430    
431 root 1.39 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
432     ecb_function_ int
433 root 1.35 ecb_popcount32 (uint32_t x)
434     {
435     x -= (x >> 1) & 0x55555555;
436     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
437     x = ((x >> 4) + x) & 0x0f0f0f0f;
438     x *= 0x01010101;
439 root 1.1
440 root 1.35 return x >> 24;
441     }
442 root 1.49
443     ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
444     ecb_function_ int ecb_ld32 (uint32_t x)
445     {
446 root 1.50 int r = 0;
447 root 1.49
448 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
449     if (x >> 8) { x >>= 8; r += 8; }
450     if (x >> 4) { x >>= 4; r += 4; }
451     if (x >> 2) { x >>= 2; r += 2; }
452     if (x >> 1) { r += 1; }
453 root 1.49
454     return r;
455     }
456    
457     ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
458     ecb_function_ int ecb_ld64 (uint64_t x)
459     {
460 root 1.50 int r = 0;
461 root 1.49
462 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
463 root 1.49
464 root 1.50 return r + ecb_ld32 (x);
465 root 1.49 }
466 root 1.1 #endif
467    
468 root 1.88 ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
469     ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
470     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
471     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
472    
473 root 1.71 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
474 root 1.70 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
475     {
476     return ( (x * 0x0802U & 0x22110U)
477     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
478     }
479    
480 root 1.71 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
481 root 1.70 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
482     {
483     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
484     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
485     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
486     x = ( x >> 8 ) | ( x << 8);
487    
488     return x;
489     }
490    
491 root 1.71 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
492 root 1.70 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
493     {
494     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
495     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
496     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
497     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
498     x = ( x >> 16 ) | ( x << 16);
499    
500     return x;
501     }
502    
503 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
504     /* so for this version we are lazy */
505     ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
506     ecb_function_ int
507     ecb_popcount64 (uint64_t x)
508     {
509     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
510     }
511    
512 root 1.50 ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
513     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
514     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
515     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
516     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
517     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
518     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
519     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
520    
521     ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
522     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
523     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
524     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
525     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
526     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
527     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
528     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
529    
530 sf-exg 1.138 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
531 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
532     #define ecb_bswap32(x) __builtin_bswap32 (x)
533     #define ecb_bswap64(x) __builtin_bswap64 (x)
534 root 1.13 #else
535 root 1.50 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
536     ecb_function_ uint16_t
537     ecb_bswap16 (uint16_t x)
538 root 1.49 {
539 root 1.50 return ecb_rotl16 (x, 8);
540 root 1.49 }
541    
542 root 1.39 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
543     ecb_function_ uint32_t
544 root 1.35 ecb_bswap32 (uint32_t x)
545     {
546 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
547 root 1.35 }
548    
549 root 1.49 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
550     ecb_function_ uint64_t
551     ecb_bswap64 (uint64_t x)
552 root 1.35 {
553 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
554 root 1.35 }
555 root 1.13 #endif
556    
557 sf-exg 1.138 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
558 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
559 root 1.13 #else
560 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
561 root 1.75 ecb_inline void ecb_unreachable (void) ecb_noreturn;
562     ecb_inline void ecb_unreachable (void) { }
563 root 1.13 #endif
564    
565 root 1.41 /* try to tell the compiler that some condition is definitely true */
566 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
567 root 1.41
568 root 1.76 ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
569     ecb_inline unsigned char
570 root 1.23 ecb_byteorder_helper (void)
571 root 1.3 {
572 root 1.98 /* the union code still generates code under pressure in gcc, */
573 sf-exg 1.111 /* but less than using pointers, and always seems to */
574 root 1.98 /* successfully return a constant. */
575     /* the reason why we have this horrible preprocessor mess */
576 root 1.99 /* is to avoid it in all cases, at least on common architectures */
577 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
578 root 1.98 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
579     return 0x44;
580     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
581     return 0x44;
582     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
583 root 1.113 return 0x11;
584 root 1.98 #else
585     union
586     {
587     uint32_t i;
588     uint8_t c;
589     } u = { 0x11223344 };
590     return u.c;
591     #endif
592 root 1.3 }
593    
594 root 1.76 ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
595     ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
596     ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
597     ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
598 root 1.3
599 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
600 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
601 root 1.31 #else
602 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
603 root 1.31 #endif
604 root 1.21
605 sf-exg 1.68 #if __cplusplus
606     template<typename T>
607     static inline T ecb_div_rd (T val, T div)
608     {
609     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
610     }
611     template<typename T>
612     static inline T ecb_div_ru (T val, T div)
613     {
614     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
615     }
616     #else
617     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
618     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
619     #endif
620 sf-exg 1.67
621 root 1.5 #if ecb_cplusplus_does_not_suck
622 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
623 root 1.35 template<typename T, int N>
624     static inline int ecb_array_length (const T (&arr)[N])
625     {
626     return N;
627     }
628 root 1.5 #else
629 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
630 root 1.5 #endif
631    
632 root 1.104 /*******************************************************************************/
633     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
634    
635     /* basically, everything uses "ieee pure-endian" floating point numbers */
636     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
637     #if 0 \
638     || __i386 || __i386__ \
639     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
640     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
641     || defined __s390__ || defined __s390x__ \
642     || defined __mips__ \
643     || defined __alpha__ \
644     || defined __hppa__ \
645     || defined __ia64__ \
646 root 1.117 || defined __m68k__ \
647     || defined __m88k__ \
648     || defined __sh__ \
649 root 1.130 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \
650 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
651 root 1.132 || defined __aarch64__
652 root 1.104 #define ECB_STDFP 1
653     #include <string.h> /* for memcpy */
654 root 1.102 #else
655 root 1.104 #define ECB_STDFP 0
656 root 1.102 #endif
657    
658 root 1.104 #ifndef ECB_NO_LIBM
659 root 1.103
660 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
661    
662 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
663     #ifdef INFINITY
664     #define ECB_INFINITY INFINITY
665     #else
666     #define ECB_INFINITY HUGE_VAL
667     #endif
668    
669     #ifdef NAN
670 root 1.121 #define ECB_NAN NAN
671     #else
672 root 1.122 #define ECB_NAN ECB_INFINITY
673 root 1.121 #endif
674 root 1.120
675     /* converts an ieee half/binary16 to a float */
676     ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
677     ecb_function_ float
678     ecb_binary16_to_float (uint16_t x)
679     {
680     int e = (x >> 10) & 0x1f;
681     int m = x & 0x3ff;
682     float r;
683    
684 root 1.121 if (!e ) r = ldexpf (m , -24);
685     else if (e != 31) r = ldexpf (m + 0x400, e - 25);
686     else if (m ) r = ECB_NAN;
687 root 1.123 else r = ECB_INFINITY;
688 root 1.120
689     return x & 0x8000 ? -r : r;
690     }
691    
692 root 1.104 /* convert a float to ieee single/binary32 */
693 root 1.103 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
694     ecb_function_ uint32_t
695     ecb_float_to_binary32 (float x)
696     {
697     uint32_t r;
698    
699     #if ECB_STDFP
700 root 1.104 memcpy (&r, &x, 4);
701 root 1.103 #else
702 root 1.105 /* slow emulation, works for anything but -0 */
703 root 1.103 uint32_t m;
704     int e;
705    
706 root 1.108 if (x == 0e0f ) return 0x00000000U;
707 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
708     if (x < -3.40282346638528860e+38f) return 0xff800000U;
709 root 1.105 if (x != x ) return 0x7fbfffffU;
710 root 1.103
711     m = frexpf (x, &e) * 0x1000000U;
712    
713     r = m & 0x80000000U;
714    
715     if (r)
716     m = -m;
717    
718 root 1.108 if (e <= -126)
719 root 1.103 {
720     m &= 0xffffffU;
721     m >>= (-125 - e);
722     e = -126;
723     }
724    
725     r |= (e + 126) << 23;
726     r |= m & 0x7fffffU;
727     #endif
728    
729     return r;
730     }
731    
732 root 1.104 /* converts an ieee single/binary32 to a float */
733 root 1.103 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
734     ecb_function_ float
735     ecb_binary32_to_float (uint32_t x)
736     {
737     float r;
738    
739     #if ECB_STDFP
740 root 1.104 memcpy (&r, &x, 4);
741 root 1.103 #else
742     /* emulation, only works for normals and subnormals and +0 */
743     int neg = x >> 31;
744     int e = (x >> 23) & 0xffU;
745    
746     x &= 0x7fffffU;
747    
748     if (e)
749     x |= 0x800000U;
750 root 1.104 else
751     e = 1;
752 root 1.103
753     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
754 root 1.108 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
755 root 1.103
756     r = neg ? -r : r;
757     #endif
758    
759     return r;
760     }
761    
762 root 1.104 /* convert a double to ieee double/binary64 */
763 root 1.103 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
764     ecb_function_ uint64_t
765     ecb_double_to_binary64 (double x)
766     {
767 root 1.104 uint64_t r;
768    
769     #if ECB_STDFP
770     memcpy (&r, &x, 8);
771     #else
772 root 1.105 /* slow emulation, works for anything but -0 */
773 root 1.104 uint64_t m;
774     int e;
775    
776 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
777 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
778     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
779 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
780 root 1.104
781     m = frexp (x, &e) * 0x20000000000000U;
782    
783     r = m & 0x8000000000000000;;
784    
785     if (r)
786     m = -m;
787    
788 root 1.108 if (e <= -1022)
789 root 1.104 {
790     m &= 0x1fffffffffffffU;
791     m >>= (-1021 - e);
792     e = -1022;
793     }
794    
795     r |= ((uint64_t)(e + 1022)) << 52;
796     r |= m & 0xfffffffffffffU;
797     #endif
798    
799     return r;
800     }
801    
802     /* converts an ieee double/binary64 to a double */
803     ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
804     ecb_function_ double
805     ecb_binary64_to_double (uint64_t x)
806     {
807     double r;
808    
809     #if ECB_STDFP
810     memcpy (&r, &x, 8);
811     #else
812     /* emulation, only works for normals and subnormals and +0 */
813     int neg = x >> 63;
814     int e = (x >> 52) & 0x7ffU;
815    
816     x &= 0xfffffffffffffU;
817    
818     if (e)
819     x |= 0x10000000000000U;
820     else
821     e = 1;
822    
823 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
824 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
825 root 1.104
826     r = neg ? -r : r;
827     #endif
828    
829     return r;
830 root 1.103 }
831    
832     #endif
833 root 1.102
834 root 1.1 #endif
835