ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.139
Committed: Tue Oct 14 14:39:06 2014 UTC (9 years, 8 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.138: +4 -3 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 1.114 #define ECB_VERSION 0x00010003
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 root 1.91 #define ECB_CPP (__cplusplus+0)
110     #define ECB_CPP11 (__cplusplus >= 201103L)
111 root 1.90
112 root 1.102 #if ECB_CPP
113 root 1.127 #define ECB_C 0
114     #define ECB_STDC_VERSION 0
115     #else
116     #define ECB_C 1
117     #define ECB_STDC_VERSION __STDC_VERSION__
118     #endif
119    
120     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
121     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
122    
123     #if ECB_CPP
124 root 1.102 #define ECB_EXTERN_C extern "C"
125     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
126     #define ECB_EXTERN_C_END }
127     #else
128     #define ECB_EXTERN_C extern
129     #define ECB_EXTERN_C_BEG
130     #define ECB_EXTERN_C_END
131     #endif
132    
133 root 1.52 /*****************************************************************************/
134    
135 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
136     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
137    
138 root 1.79 #if ECB_NO_THREADS
139 root 1.95 #define ECB_NO_SMP 1
140 root 1.79 #endif
141    
142 root 1.93 #if ECB_NO_SMP
143 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
144 root 1.58 #endif
145    
146 root 1.52 #ifndef ECB_MEMORY_FENCE
147 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
148 root 1.73 #if __i386 || __i386__
149 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
150 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
151     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
152 root 1.73 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
153 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
154     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
155     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
156 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
157 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
158 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
159     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
160 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
161 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
162 root 1.94 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
163     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
164 root 1.129 #elif __aarch64__
165     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
166 root 1.124 #elif (__sparc || __sparc__) && !__sparcv8
167 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
168     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
169     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
170 root 1.85 #elif defined __s390__ || defined __s390x__
171 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
172 root 1.85 #elif defined __mips__
173 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
174 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
175     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
176 root 1.86 #elif defined __alpha__
177 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
178     #elif defined __hppa__
179     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
180     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
181     #elif defined __ia64__
182     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
183 root 1.117 #elif defined __m68k__
184     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
185     #elif defined __m88k__
186     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
187     #elif defined __sh__
188     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
189 root 1.52 #endif
190     #endif
191     #endif
192    
193     #ifndef ECB_MEMORY_FENCE
194 root 1.93 #if ECB_GCC_VERSION(4,7)
195 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
196 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
197 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
198     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
199 root 1.110
200     /* The __has_feature syntax from clang is so misdesigned that we cannot use it
201     * without risking compile time errors with other compilers. We *could*
202     * define our own ecb_clang_has_feature, but I just can't be bothered to work
203 root 1.112 * around this shit time and again.
204 root 1.110 * #elif defined __clang && __has_feature (cxx_atomic)
205     * // see comment below (stdatomic.h) about the C11 memory model.
206     * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
207 root 1.128 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
208     * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
209 root 1.110 */
210    
211 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
212 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
213 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
214     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
215     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
216     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
217     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
218     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
219 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
220     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
221     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
222     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
223     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
224 root 1.85 #elif defined _WIN32
225 root 1.55 #include <WinNT.h>
226 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
227 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
228     #include <mbarrier.h>
229     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
230     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
231     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
232 root 1.82 #elif __xlC__
233 root 1.83 #define ECB_MEMORY_FENCE __sync ()
234 root 1.52 #endif
235     #endif
236    
237 root 1.53 #ifndef ECB_MEMORY_FENCE
238 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
239     /* we assume that these memory fences work on all variables/all memory accesses, */
240     /* not just C11 atomics and atomic accesses */
241     #include <stdatomic.h>
242 root 1.97 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
243     /* any fence other than seq_cst, which isn't very efficient for us. */
244     /* Why that is, we don't know - either the C11 memory model is quite useless */
245     /* for most usages, or gcc and clang have a bug */
246     /* I *currently* lean towards the latter, and inefficiently implement */
247     /* all three of ecb's fences as a seq_cst fence */
248 root 1.128 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
249     /* for all __atomic_thread_fence's except seq_cst */
250 root 1.94 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
251     #endif
252     #endif
253    
254     #ifndef ECB_MEMORY_FENCE
255 root 1.62 #if !ECB_AVOID_PTHREADS
256     /*
257     * if you get undefined symbol references to pthread_mutex_lock,
258     * or failure to find pthread.h, then you should implement
259     * the ECB_MEMORY_FENCE operations for your cpu/compiler
260     * OR provide pthread.h and link against the posix thread library
261     * of your system.
262     */
263     #include <pthread.h>
264     #define ECB_NEEDS_PTHREADS 1
265     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
266 root 1.52
267 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
268     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
269     #endif
270     #endif
271    
272 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
273 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
274 root 1.62 #endif
275    
276 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
277 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
278     #endif
279    
280     /*****************************************************************************/
281    
282 root 1.39 #if __cplusplus
283 root 1.46 #define ecb_inline static inline
284 root 1.38 #elif ECB_GCC_VERSION(2,5)
285 root 1.46 #define ecb_inline static __inline__
286 root 1.39 #elif ECB_C99
287 root 1.46 #define ecb_inline static inline
288 root 1.29 #else
289 root 1.46 #define ecb_inline static
290 root 1.38 #endif
291    
292     #if ECB_GCC_VERSION(3,3)
293     #define ecb_restrict __restrict__
294 root 1.39 #elif ECB_C99
295 root 1.38 #define ecb_restrict restrict
296     #else
297     #define ecb_restrict
298 root 1.4 #endif
299    
300 root 1.38 typedef int ecb_bool;
301    
302 root 1.8 #define ECB_CONCAT_(a, b) a ## b
303     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
304     #define ECB_STRINGIFY_(a) # a
305     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
306    
307 root 1.46 #define ecb_function_ ecb_inline
308 root 1.3
309 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
310 root 1.37 #define ecb_attribute(attrlist) __attribute__(attrlist)
311     #else
312     #define ecb_attribute(attrlist)
313 sf-exg 1.138 #endif
314 root 1.127
315 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
316     #define ecb_is_constant(expr) __builtin_constant_p (expr)
317     #else
318 root 1.127 /* possible C11 impl for integral types
319     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
320     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
321    
322 root 1.37 #define ecb_is_constant(expr) 0
323 sf-exg 1.138 #endif
324    
325     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
326     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
327     #else
328 root 1.37 #define ecb_expect(expr,value) (expr)
329 sf-exg 1.138 #endif
330    
331     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
332     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
333     #else
334 root 1.37 #define ecb_prefetch(addr,rw,locality)
335 root 1.1 #endif
336    
337 root 1.2 /* no emulation for ecb_decltype */
338 root 1.1 #if ECB_GCC_VERSION(4,5)
339 root 1.37 #define ecb_decltype(x) __decltype(x)
340 root 1.1 #elif ECB_GCC_VERSION(3,0)
341 root 1.39 #define ecb_decltype(x) __typeof(x)
342 root 1.1 #endif
343    
344 root 1.135 #if _MSC_VER >= 1300
345     #define ecb_deprecated __declspec(deprecated)
346     #else
347     #define ecb_deprecated ecb_attribute ((__deprecated__))
348     #endif
349    
350 root 1.24 #define ecb_noinline ecb_attribute ((__noinline__))
351     #define ecb_unused ecb_attribute ((__unused__))
352     #define ecb_const ecb_attribute ((__const__))
353     #define ecb_pure ecb_attribute ((__pure__))
354 root 1.35
355 root 1.136 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
356 root 1.90 #if ECB_C11
357     #define ecb_noreturn _Noreturn
358     #else
359     #define ecb_noreturn ecb_attribute ((__noreturn__))
360     #endif
361    
362 root 1.35 #if ECB_GCC_VERSION(4,3)
363 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
364     #define ecb_hot ecb_attribute ((__hot__))
365     #define ecb_cold ecb_attribute ((__cold__))
366 root 1.35 #else
367     #define ecb_artificial
368     #define ecb_hot
369     #define ecb_cold
370     #endif
371 root 1.1
372 root 1.39 /* put around conditional expressions if you are very sure that the */
373     /* expression is mostly true or mostly false. note that these return */
374     /* booleans, not the expression. */
375 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
376     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
377 root 1.36 /* for compatibility to the rest of the world */
378 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
379     #define ecb_unlikely(expr) ecb_expect_false (expr)
380 root 1.1
381 root 1.3 /* count trailing zero bits and count # of one bits */
382 root 1.139 #if ECB_GCC_VERSION(3,4) \
383     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
384     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
385     && ECB_CLANG_BUILTIN(__builtin_popcount))
386 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
387     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
388     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
389 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
390 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
391 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
392 root 1.49 /* no popcountll */
393 root 1.1 #else
394 root 1.39 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
395     ecb_function_ int
396 root 1.35 ecb_ctz32 (uint32_t x)
397     {
398     int r = 0;
399    
400 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
401 root 1.35
402 root 1.50 #if ECB_branchless_on_i386
403     r += !!(x & 0xaaaaaaaa) << 0;
404     r += !!(x & 0xcccccccc) << 1;
405     r += !!(x & 0xf0f0f0f0) << 2;
406     r += !!(x & 0xff00ff00) << 3;
407     r += !!(x & 0xffff0000) << 4;
408     #else
409 root 1.35 if (x & 0xaaaaaaaa) r += 1;
410     if (x & 0xcccccccc) r += 2;
411     if (x & 0xf0f0f0f0) r += 4;
412     if (x & 0xff00ff00) r += 8;
413     if (x & 0xffff0000) r += 16;
414 root 1.50 #endif
415 root 1.35
416     return r;
417     }
418    
419 root 1.49 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
420     ecb_function_ int
421     ecb_ctz64 (uint64_t x)
422     {
423     int shift = x & 0xffffffffU ? 0 : 32;
424 root 1.50 return ecb_ctz32 (x >> shift) + shift;
425 root 1.49 }
426    
427 root 1.39 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
428     ecb_function_ int
429 root 1.35 ecb_popcount32 (uint32_t x)
430     {
431     x -= (x >> 1) & 0x55555555;
432     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
433     x = ((x >> 4) + x) & 0x0f0f0f0f;
434     x *= 0x01010101;
435 root 1.1
436 root 1.35 return x >> 24;
437     }
438 root 1.49
439     ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
440     ecb_function_ int ecb_ld32 (uint32_t x)
441     {
442 root 1.50 int r = 0;
443 root 1.49
444 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
445     if (x >> 8) { x >>= 8; r += 8; }
446     if (x >> 4) { x >>= 4; r += 4; }
447     if (x >> 2) { x >>= 2; r += 2; }
448     if (x >> 1) { r += 1; }
449 root 1.49
450     return r;
451     }
452    
453     ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
454     ecb_function_ int ecb_ld64 (uint64_t x)
455     {
456 root 1.50 int r = 0;
457 root 1.49
458 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
459 root 1.49
460 root 1.50 return r + ecb_ld32 (x);
461 root 1.49 }
462 root 1.1 #endif
463    
464 root 1.88 ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
465     ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
466     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
467     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
468    
469 root 1.71 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
470 root 1.70 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
471     {
472     return ( (x * 0x0802U & 0x22110U)
473     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
474     }
475    
476 root 1.71 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
477 root 1.70 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
478     {
479     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
480     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
481     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
482     x = ( x >> 8 ) | ( x << 8);
483    
484     return x;
485     }
486    
487 root 1.71 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
488 root 1.70 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
489     {
490     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
491     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
492     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
493     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
494     x = ( x >> 16 ) | ( x << 16);
495    
496     return x;
497     }
498    
499 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
500     /* so for this version we are lazy */
501     ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
502     ecb_function_ int
503     ecb_popcount64 (uint64_t x)
504     {
505     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
506     }
507    
508 root 1.50 ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
509     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
510     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
511     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
512     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
513     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
514     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
515     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
516    
517     ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
518     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
519     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
520     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
521     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
522     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
523     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
524     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
525    
526 sf-exg 1.138 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
527 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
528     #define ecb_bswap32(x) __builtin_bswap32 (x)
529     #define ecb_bswap64(x) __builtin_bswap64 (x)
530 root 1.13 #else
531 root 1.50 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
532     ecb_function_ uint16_t
533     ecb_bswap16 (uint16_t x)
534 root 1.49 {
535 root 1.50 return ecb_rotl16 (x, 8);
536 root 1.49 }
537    
538 root 1.39 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
539     ecb_function_ uint32_t
540 root 1.35 ecb_bswap32 (uint32_t x)
541     {
542 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
543 root 1.35 }
544    
545 root 1.49 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
546     ecb_function_ uint64_t
547     ecb_bswap64 (uint64_t x)
548 root 1.35 {
549 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
550 root 1.35 }
551 root 1.13 #endif
552    
553 sf-exg 1.138 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
554 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
555 root 1.13 #else
556 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
557 root 1.75 ecb_inline void ecb_unreachable (void) ecb_noreturn;
558     ecb_inline void ecb_unreachable (void) { }
559 root 1.13 #endif
560    
561 root 1.41 /* try to tell the compiler that some condition is definitely true */
562 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
563 root 1.41
564 root 1.76 ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
565     ecb_inline unsigned char
566 root 1.23 ecb_byteorder_helper (void)
567 root 1.3 {
568 root 1.98 /* the union code still generates code under pressure in gcc, */
569 sf-exg 1.111 /* but less than using pointers, and always seems to */
570 root 1.98 /* successfully return a constant. */
571     /* the reason why we have this horrible preprocessor mess */
572 root 1.99 /* is to avoid it in all cases, at least on common architectures */
573 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
574 root 1.98 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
575     return 0x44;
576     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
577     return 0x44;
578     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
579 root 1.113 return 0x11;
580 root 1.98 #else
581     union
582     {
583     uint32_t i;
584     uint8_t c;
585     } u = { 0x11223344 };
586     return u.c;
587     #endif
588 root 1.3 }
589    
590 root 1.76 ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
591     ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
592     ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
593     ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
594 root 1.3
595 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
596 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
597 root 1.31 #else
598 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
599 root 1.31 #endif
600 root 1.21
601 sf-exg 1.68 #if __cplusplus
602     template<typename T>
603     static inline T ecb_div_rd (T val, T div)
604     {
605     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
606     }
607     template<typename T>
608     static inline T ecb_div_ru (T val, T div)
609     {
610     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
611     }
612     #else
613     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
614     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
615     #endif
616 sf-exg 1.67
617 root 1.5 #if ecb_cplusplus_does_not_suck
618 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
619 root 1.35 template<typename T, int N>
620     static inline int ecb_array_length (const T (&arr)[N])
621     {
622     return N;
623     }
624 root 1.5 #else
625 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
626 root 1.5 #endif
627    
628 root 1.104 /*******************************************************************************/
629     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
630    
631     /* basically, everything uses "ieee pure-endian" floating point numbers */
632     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
633     #if 0 \
634     || __i386 || __i386__ \
635     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
636     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
637     || defined __s390__ || defined __s390x__ \
638     || defined __mips__ \
639     || defined __alpha__ \
640     || defined __hppa__ \
641     || defined __ia64__ \
642 root 1.117 || defined __m68k__ \
643     || defined __m88k__ \
644     || defined __sh__ \
645 root 1.130 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \
646 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
647 root 1.132 || defined __aarch64__
648 root 1.104 #define ECB_STDFP 1
649     #include <string.h> /* for memcpy */
650 root 1.102 #else
651 root 1.104 #define ECB_STDFP 0
652 root 1.102 #endif
653    
654 root 1.104 #ifndef ECB_NO_LIBM
655 root 1.103
656 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
657    
658 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
659     #ifdef INFINITY
660     #define ECB_INFINITY INFINITY
661     #else
662     #define ECB_INFINITY HUGE_VAL
663     #endif
664    
665     #ifdef NAN
666 root 1.121 #define ECB_NAN NAN
667     #else
668 root 1.122 #define ECB_NAN ECB_INFINITY
669 root 1.121 #endif
670 root 1.120
671     /* converts an ieee half/binary16 to a float */
672     ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
673     ecb_function_ float
674     ecb_binary16_to_float (uint16_t x)
675     {
676     int e = (x >> 10) & 0x1f;
677     int m = x & 0x3ff;
678     float r;
679    
680 root 1.121 if (!e ) r = ldexpf (m , -24);
681     else if (e != 31) r = ldexpf (m + 0x400, e - 25);
682     else if (m ) r = ECB_NAN;
683 root 1.123 else r = ECB_INFINITY;
684 root 1.120
685     return x & 0x8000 ? -r : r;
686     }
687    
688 root 1.104 /* convert a float to ieee single/binary32 */
689 root 1.103 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
690     ecb_function_ uint32_t
691     ecb_float_to_binary32 (float x)
692     {
693     uint32_t r;
694    
695     #if ECB_STDFP
696 root 1.104 memcpy (&r, &x, 4);
697 root 1.103 #else
698 root 1.105 /* slow emulation, works for anything but -0 */
699 root 1.103 uint32_t m;
700     int e;
701    
702 root 1.108 if (x == 0e0f ) return 0x00000000U;
703 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
704     if (x < -3.40282346638528860e+38f) return 0xff800000U;
705 root 1.105 if (x != x ) return 0x7fbfffffU;
706 root 1.103
707     m = frexpf (x, &e) * 0x1000000U;
708    
709     r = m & 0x80000000U;
710    
711     if (r)
712     m = -m;
713    
714 root 1.108 if (e <= -126)
715 root 1.103 {
716     m &= 0xffffffU;
717     m >>= (-125 - e);
718     e = -126;
719     }
720    
721     r |= (e + 126) << 23;
722     r |= m & 0x7fffffU;
723     #endif
724    
725     return r;
726     }
727    
728 root 1.104 /* converts an ieee single/binary32 to a float */
729 root 1.103 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
730     ecb_function_ float
731     ecb_binary32_to_float (uint32_t x)
732     {
733     float r;
734    
735     #if ECB_STDFP
736 root 1.104 memcpy (&r, &x, 4);
737 root 1.103 #else
738     /* emulation, only works for normals and subnormals and +0 */
739     int neg = x >> 31;
740     int e = (x >> 23) & 0xffU;
741    
742     x &= 0x7fffffU;
743    
744     if (e)
745     x |= 0x800000U;
746 root 1.104 else
747     e = 1;
748 root 1.103
749     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
750 root 1.108 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
751 root 1.103
752     r = neg ? -r : r;
753     #endif
754    
755     return r;
756     }
757    
758 root 1.104 /* convert a double to ieee double/binary64 */
759 root 1.103 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
760     ecb_function_ uint64_t
761     ecb_double_to_binary64 (double x)
762     {
763 root 1.104 uint64_t r;
764    
765     #if ECB_STDFP
766     memcpy (&r, &x, 8);
767     #else
768 root 1.105 /* slow emulation, works for anything but -0 */
769 root 1.104 uint64_t m;
770     int e;
771    
772 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
773 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
774     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
775 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
776 root 1.104
777     m = frexp (x, &e) * 0x20000000000000U;
778    
779     r = m & 0x8000000000000000;;
780    
781     if (r)
782     m = -m;
783    
784 root 1.108 if (e <= -1022)
785 root 1.104 {
786     m &= 0x1fffffffffffffU;
787     m >>= (-1021 - e);
788     e = -1022;
789     }
790    
791     r |= ((uint64_t)(e + 1022)) << 52;
792     r |= m & 0xfffffffffffffU;
793     #endif
794    
795     return r;
796     }
797    
798     /* converts an ieee double/binary64 to a double */
799     ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
800     ecb_function_ double
801     ecb_binary64_to_double (uint64_t x)
802     {
803     double r;
804    
805     #if ECB_STDFP
806     memcpy (&r, &x, 8);
807     #else
808     /* emulation, only works for normals and subnormals and +0 */
809     int neg = x >> 63;
810     int e = (x >> 52) & 0x7ffU;
811    
812     x &= 0xfffffffffffffU;
813    
814     if (e)
815     x |= 0x10000000000000U;
816     else
817     e = 1;
818    
819 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
820 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
821 root 1.104
822     r = neg ? -r : r;
823     #endif
824    
825     return r;
826 root 1.103 }
827    
828     #endif
829 root 1.102
830 root 1.1 #endif
831