ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.134
Committed: Mon Jun 9 17:31:57 2014 UTC (10 years ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.133: +1 -0 lines
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 /*
2 root 1.17 * libecb - http://software.schmorp.de/pkg/libecb
3 root 1.1 *
4 root 1.129 * 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 1.14 #ifndef ECB_GCC_VERSION
96 root 1.85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
97 root 1.37 #define ECB_GCC_VERSION(major,minor) 0
98     #else
99     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
100     #endif
101 root 1.12 #endif
102 root 1.1
103 root 1.91 #define ECB_CPP (__cplusplus+0)
104     #define ECB_CPP11 (__cplusplus >= 201103L)
105 root 1.90
106 root 1.102 #if ECB_CPP
107 root 1.127 #define ECB_C 0
108     #define ECB_STDC_VERSION 0
109     #else
110     #define ECB_C 1
111     #define ECB_STDC_VERSION __STDC_VERSION__
112     #endif
113    
114     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
115     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
116    
117     #if ECB_CPP
118 root 1.102 #define ECB_EXTERN_C extern "C"
119     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
120     #define ECB_EXTERN_C_END }
121     #else
122     #define ECB_EXTERN_C extern
123     #define ECB_EXTERN_C_BEG
124     #define ECB_EXTERN_C_END
125     #endif
126    
127 root 1.52 /*****************************************************************************/
128    
129 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
130     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
131    
132 root 1.79 #if ECB_NO_THREADS
133 root 1.95 #define ECB_NO_SMP 1
134 root 1.79 #endif
135    
136 root 1.93 #if ECB_NO_SMP
137 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
138 root 1.58 #endif
139    
140 root 1.52 #ifndef ECB_MEMORY_FENCE
141 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
142 root 1.73 #if __i386 || __i386__
143 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
144 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
145     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
146 root 1.73 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
147 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
148     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
149     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
150 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
151 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
152 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
153     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
154 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
155 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
156 root 1.94 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
157     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
158 root 1.129 #elif __aarch64__
159     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
160 root 1.124 #elif (__sparc || __sparc__) && !__sparcv8
161 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
162     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
163     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
164 root 1.85 #elif defined __s390__ || defined __s390x__
165 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
166 root 1.85 #elif defined __mips__
167 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
168 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
169     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
170 root 1.86 #elif defined __alpha__
171 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
172     #elif defined __hppa__
173     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
174     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
175     #elif defined __ia64__
176     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
177 root 1.117 #elif defined __m68k__
178     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
179     #elif defined __m88k__
180     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
181     #elif defined __sh__
182     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
183 root 1.52 #endif
184     #endif
185     #endif
186    
187     #ifndef ECB_MEMORY_FENCE
188 root 1.93 #if ECB_GCC_VERSION(4,7)
189 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
190 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
191 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
192     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
193 root 1.110
194     /* The __has_feature syntax from clang is so misdesigned that we cannot use it
195     * without risking compile time errors with other compilers. We *could*
196     * define our own ecb_clang_has_feature, but I just can't be bothered to work
197 root 1.112 * around this shit time and again.
198 root 1.110 * #elif defined __clang && __has_feature (cxx_atomic)
199     * // see comment below (stdatomic.h) about the C11 memory model.
200     * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
201 root 1.128 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
202     * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
203 root 1.110 */
204    
205 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
206 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
207 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
208     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
209     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
210     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
211     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
212     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
213 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
214     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
215     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
216     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
217     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
218 root 1.85 #elif defined _WIN32
219 root 1.55 #include <WinNT.h>
220 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
221 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
222     #include <mbarrier.h>
223     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
224     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
225     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
226 root 1.82 #elif __xlC__
227 root 1.83 #define ECB_MEMORY_FENCE __sync ()
228 root 1.52 #endif
229     #endif
230    
231 root 1.53 #ifndef ECB_MEMORY_FENCE
232 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
233     /* we assume that these memory fences work on all variables/all memory accesses, */
234     /* not just C11 atomics and atomic accesses */
235     #include <stdatomic.h>
236 root 1.97 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
237     /* any fence other than seq_cst, which isn't very efficient for us. */
238     /* Why that is, we don't know - either the C11 memory model is quite useless */
239     /* for most usages, or gcc and clang have a bug */
240     /* I *currently* lean towards the latter, and inefficiently implement */
241     /* all three of ecb's fences as a seq_cst fence */
242 root 1.128 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
243     /* for all __atomic_thread_fence's except seq_cst */
244 root 1.94 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
245     #endif
246     #endif
247    
248     #ifndef ECB_MEMORY_FENCE
249 root 1.62 #if !ECB_AVOID_PTHREADS
250     /*
251     * if you get undefined symbol references to pthread_mutex_lock,
252     * or failure to find pthread.h, then you should implement
253     * the ECB_MEMORY_FENCE operations for your cpu/compiler
254     * OR provide pthread.h and link against the posix thread library
255     * of your system.
256     */
257     #include <pthread.h>
258     #define ECB_NEEDS_PTHREADS 1
259     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
260 root 1.52
261 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
262     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
263     #endif
264     #endif
265    
266 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
267 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
268 root 1.62 #endif
269    
270 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
271 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
272     #endif
273    
274     /*****************************************************************************/
275    
276 root 1.39 #if __cplusplus
277 root 1.46 #define ecb_inline static inline
278 root 1.38 #elif ECB_GCC_VERSION(2,5)
279 root 1.46 #define ecb_inline static __inline__
280 root 1.39 #elif ECB_C99
281 root 1.46 #define ecb_inline static inline
282 root 1.29 #else
283 root 1.46 #define ecb_inline static
284 root 1.38 #endif
285    
286     #if ECB_GCC_VERSION(3,3)
287     #define ecb_restrict __restrict__
288 root 1.39 #elif ECB_C99
289 root 1.38 #define ecb_restrict restrict
290     #else
291     #define ecb_restrict
292 root 1.4 #endif
293    
294 root 1.38 typedef int ecb_bool;
295    
296 root 1.8 #define ECB_CONCAT_(a, b) a ## b
297     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
298     #define ECB_STRINGIFY_(a) # a
299     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
300    
301 root 1.46 #define ecb_function_ ecb_inline
302 root 1.3
303 root 1.1 #if ECB_GCC_VERSION(3,1)
304 root 1.37 #define ecb_attribute(attrlist) __attribute__(attrlist)
305     #define ecb_is_constant(expr) __builtin_constant_p (expr)
306     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
307     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
308     #else
309     #define ecb_attribute(attrlist)
310 root 1.127
311     /* possible C11 impl for integral types
312     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
313     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
314    
315 root 1.37 #define ecb_is_constant(expr) 0
316     #define ecb_expect(expr,value) (expr)
317     #define ecb_prefetch(addr,rw,locality)
318 root 1.1 #endif
319    
320 root 1.2 /* no emulation for ecb_decltype */
321 root 1.1 #if ECB_GCC_VERSION(4,5)
322 root 1.37 #define ecb_decltype(x) __decltype(x)
323 root 1.1 #elif ECB_GCC_VERSION(3,0)
324 root 1.39 #define ecb_decltype(x) __typeof(x)
325 root 1.1 #endif
326    
327 root 1.134 #define ecb_deprecated ecb_attribute ((__deprecated__))
328 root 1.24 #define ecb_noinline ecb_attribute ((__noinline__))
329     #define ecb_unused ecb_attribute ((__unused__))
330     #define ecb_const ecb_attribute ((__const__))
331     #define ecb_pure ecb_attribute ((__pure__))
332 root 1.35
333 root 1.90 #if ECB_C11
334     #define ecb_noreturn _Noreturn
335     #else
336     #define ecb_noreturn ecb_attribute ((__noreturn__))
337     #endif
338    
339 root 1.35 #if ECB_GCC_VERSION(4,3)
340 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
341     #define ecb_hot ecb_attribute ((__hot__))
342     #define ecb_cold ecb_attribute ((__cold__))
343 root 1.35 #else
344     #define ecb_artificial
345     #define ecb_hot
346     #define ecb_cold
347     #endif
348 root 1.1
349 root 1.39 /* put around conditional expressions if you are very sure that the */
350     /* expression is mostly true or mostly false. note that these return */
351     /* booleans, not the expression. */
352 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
353     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
354 root 1.36 /* for compatibility to the rest of the world */
355 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
356     #define ecb_unlikely(expr) ecb_expect_false (expr)
357 root 1.1
358 root 1.3 /* count trailing zero bits and count # of one bits */
359 root 1.1 #if ECB_GCC_VERSION(3,4)
360 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
361     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
362     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
363 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
364 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
365 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
366 root 1.49 /* no popcountll */
367 root 1.1 #else
368 root 1.39 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
369     ecb_function_ int
370 root 1.35 ecb_ctz32 (uint32_t x)
371     {
372     int r = 0;
373    
374 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
375 root 1.35
376 root 1.50 #if ECB_branchless_on_i386
377     r += !!(x & 0xaaaaaaaa) << 0;
378     r += !!(x & 0xcccccccc) << 1;
379     r += !!(x & 0xf0f0f0f0) << 2;
380     r += !!(x & 0xff00ff00) << 3;
381     r += !!(x & 0xffff0000) << 4;
382     #else
383 root 1.35 if (x & 0xaaaaaaaa) r += 1;
384     if (x & 0xcccccccc) r += 2;
385     if (x & 0xf0f0f0f0) r += 4;
386     if (x & 0xff00ff00) r += 8;
387     if (x & 0xffff0000) r += 16;
388 root 1.50 #endif
389 root 1.35
390     return r;
391     }
392    
393 root 1.49 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
394     ecb_function_ int
395     ecb_ctz64 (uint64_t x)
396     {
397     int shift = x & 0xffffffffU ? 0 : 32;
398 root 1.50 return ecb_ctz32 (x >> shift) + shift;
399 root 1.49 }
400    
401 root 1.39 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
402     ecb_function_ int
403 root 1.35 ecb_popcount32 (uint32_t x)
404     {
405     x -= (x >> 1) & 0x55555555;
406     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
407     x = ((x >> 4) + x) & 0x0f0f0f0f;
408     x *= 0x01010101;
409 root 1.1
410 root 1.35 return x >> 24;
411     }
412 root 1.49
413     ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
414     ecb_function_ int ecb_ld32 (uint32_t x)
415     {
416 root 1.50 int r = 0;
417 root 1.49
418 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
419     if (x >> 8) { x >>= 8; r += 8; }
420     if (x >> 4) { x >>= 4; r += 4; }
421     if (x >> 2) { x >>= 2; r += 2; }
422     if (x >> 1) { r += 1; }
423 root 1.49
424     return r;
425     }
426    
427     ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
428     ecb_function_ int ecb_ld64 (uint64_t x)
429     {
430 root 1.50 int r = 0;
431 root 1.49
432 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
433 root 1.49
434 root 1.50 return r + ecb_ld32 (x);
435 root 1.49 }
436 root 1.1 #endif
437    
438 root 1.88 ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
439     ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
440     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
441     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
442    
443 root 1.71 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
444 root 1.70 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
445     {
446     return ( (x * 0x0802U & 0x22110U)
447     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
448     }
449    
450 root 1.71 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
451 root 1.70 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
452     {
453     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
454     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
455     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
456     x = ( x >> 8 ) | ( x << 8);
457    
458     return x;
459     }
460    
461 root 1.71 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
462 root 1.70 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
463     {
464     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
465     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
466     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
467     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
468     x = ( x >> 16 ) | ( x << 16);
469    
470     return x;
471     }
472    
473 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
474     /* so for this version we are lazy */
475     ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
476     ecb_function_ int
477     ecb_popcount64 (uint64_t x)
478     {
479     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
480     }
481    
482 root 1.50 ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
483     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
484     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
485     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
486     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
487     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
488     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
489     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
490    
491     ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
492     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
493     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
494     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
495     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
496     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
497     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
498     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
499    
500 root 1.13 #if ECB_GCC_VERSION(4,3)
501 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
502     #define ecb_bswap32(x) __builtin_bswap32 (x)
503     #define ecb_bswap64(x) __builtin_bswap64 (x)
504 root 1.13 #else
505 root 1.50 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
506     ecb_function_ uint16_t
507     ecb_bswap16 (uint16_t x)
508 root 1.49 {
509 root 1.50 return ecb_rotl16 (x, 8);
510 root 1.49 }
511    
512 root 1.39 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
513     ecb_function_ uint32_t
514 root 1.35 ecb_bswap32 (uint32_t x)
515     {
516 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
517 root 1.35 }
518    
519 root 1.49 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
520     ecb_function_ uint64_t
521     ecb_bswap64 (uint64_t x)
522 root 1.35 {
523 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
524 root 1.35 }
525 root 1.13 #endif
526    
527     #if ECB_GCC_VERSION(4,5)
528 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
529 root 1.13 #else
530 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
531 root 1.75 ecb_inline void ecb_unreachable (void) ecb_noreturn;
532     ecb_inline void ecb_unreachable (void) { }
533 root 1.13 #endif
534    
535 root 1.41 /* try to tell the compiler that some condition is definitely true */
536 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
537 root 1.41
538 root 1.76 ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
539     ecb_inline unsigned char
540 root 1.23 ecb_byteorder_helper (void)
541 root 1.3 {
542 root 1.98 /* the union code still generates code under pressure in gcc, */
543 sf-exg 1.111 /* but less than using pointers, and always seems to */
544 root 1.98 /* successfully return a constant. */
545     /* the reason why we have this horrible preprocessor mess */
546 root 1.99 /* is to avoid it in all cases, at least on common architectures */
547 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
548 root 1.98 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
549     return 0x44;
550     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
551     return 0x44;
552     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
553 root 1.113 return 0x11;
554 root 1.98 #else
555     union
556     {
557     uint32_t i;
558     uint8_t c;
559     } u = { 0x11223344 };
560     return u.c;
561     #endif
562 root 1.3 }
563    
564 root 1.76 ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
565     ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
566     ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
567     ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
568 root 1.3
569 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
570 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
571 root 1.31 #else
572 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
573 root 1.31 #endif
574 root 1.21
575 sf-exg 1.68 #if __cplusplus
576     template<typename T>
577     static inline T ecb_div_rd (T val, T div)
578     {
579     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
580     }
581     template<typename T>
582     static inline T ecb_div_ru (T val, T div)
583     {
584     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
585     }
586     #else
587     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
588     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
589     #endif
590 sf-exg 1.67
591 root 1.5 #if ecb_cplusplus_does_not_suck
592 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
593 root 1.35 template<typename T, int N>
594     static inline int ecb_array_length (const T (&arr)[N])
595     {
596     return N;
597     }
598 root 1.5 #else
599 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
600 root 1.5 #endif
601    
602 root 1.104 /*******************************************************************************/
603     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
604    
605     /* basically, everything uses "ieee pure-endian" floating point numbers */
606     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
607     #if 0 \
608     || __i386 || __i386__ \
609     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
610     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
611     || defined __s390__ || defined __s390x__ \
612     || defined __mips__ \
613     || defined __alpha__ \
614     || defined __hppa__ \
615     || defined __ia64__ \
616 root 1.117 || defined __m68k__ \
617     || defined __m88k__ \
618     || defined __sh__ \
619 root 1.130 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \
620 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
621 root 1.132 || defined __aarch64__
622 root 1.104 #define ECB_STDFP 1
623     #include <string.h> /* for memcpy */
624 root 1.102 #else
625 root 1.104 #define ECB_STDFP 0
626 root 1.102 #endif
627    
628 root 1.104 #ifndef ECB_NO_LIBM
629 root 1.103
630 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
631    
632 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
633     #ifdef INFINITY
634     #define ECB_INFINITY INFINITY
635     #else
636     #define ECB_INFINITY HUGE_VAL
637     #endif
638    
639     #ifdef NAN
640 root 1.121 #define ECB_NAN NAN
641     #else
642 root 1.122 #define ECB_NAN ECB_INFINITY
643 root 1.121 #endif
644 root 1.120
645     /* converts an ieee half/binary16 to a float */
646     ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
647     ecb_function_ float
648     ecb_binary16_to_float (uint16_t x)
649     {
650     int e = (x >> 10) & 0x1f;
651     int m = x & 0x3ff;
652     float r;
653    
654 root 1.121 if (!e ) r = ldexpf (m , -24);
655     else if (e != 31) r = ldexpf (m + 0x400, e - 25);
656     else if (m ) r = ECB_NAN;
657 root 1.123 else r = ECB_INFINITY;
658 root 1.120
659     return x & 0x8000 ? -r : r;
660     }
661    
662 root 1.104 /* convert a float to ieee single/binary32 */
663 root 1.103 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
664     ecb_function_ uint32_t
665     ecb_float_to_binary32 (float x)
666     {
667     uint32_t r;
668    
669     #if ECB_STDFP
670 root 1.104 memcpy (&r, &x, 4);
671 root 1.103 #else
672 root 1.105 /* slow emulation, works for anything but -0 */
673 root 1.103 uint32_t m;
674     int e;
675    
676 root 1.108 if (x == 0e0f ) return 0x00000000U;
677 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
678     if (x < -3.40282346638528860e+38f) return 0xff800000U;
679 root 1.105 if (x != x ) return 0x7fbfffffU;
680 root 1.103
681     m = frexpf (x, &e) * 0x1000000U;
682    
683     r = m & 0x80000000U;
684    
685     if (r)
686     m = -m;
687    
688 root 1.108 if (e <= -126)
689 root 1.103 {
690     m &= 0xffffffU;
691     m >>= (-125 - e);
692     e = -126;
693     }
694    
695     r |= (e + 126) << 23;
696     r |= m & 0x7fffffU;
697     #endif
698    
699     return r;
700     }
701    
702 root 1.104 /* converts an ieee single/binary32 to a float */
703 root 1.103 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
704     ecb_function_ float
705     ecb_binary32_to_float (uint32_t x)
706     {
707     float r;
708    
709     #if ECB_STDFP
710 root 1.104 memcpy (&r, &x, 4);
711 root 1.103 #else
712     /* emulation, only works for normals and subnormals and +0 */
713     int neg = x >> 31;
714     int e = (x >> 23) & 0xffU;
715    
716     x &= 0x7fffffU;
717    
718     if (e)
719     x |= 0x800000U;
720 root 1.104 else
721     e = 1;
722 root 1.103
723     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
724 root 1.108 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
725 root 1.103
726     r = neg ? -r : r;
727     #endif
728    
729     return r;
730     }
731    
732 root 1.104 /* convert a double to ieee double/binary64 */
733 root 1.103 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
734     ecb_function_ uint64_t
735     ecb_double_to_binary64 (double x)
736     {
737 root 1.104 uint64_t r;
738    
739     #if ECB_STDFP
740     memcpy (&r, &x, 8);
741     #else
742 root 1.105 /* slow emulation, works for anything but -0 */
743 root 1.104 uint64_t m;
744     int e;
745    
746 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
747 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
748     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
749 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
750 root 1.104
751     m = frexp (x, &e) * 0x20000000000000U;
752    
753     r = m & 0x8000000000000000;;
754    
755     if (r)
756     m = -m;
757    
758 root 1.108 if (e <= -1022)
759 root 1.104 {
760     m &= 0x1fffffffffffffU;
761     m >>= (-1021 - e);
762     e = -1022;
763     }
764    
765     r |= ((uint64_t)(e + 1022)) << 52;
766     r |= m & 0xfffffffffffffU;
767     #endif
768    
769     return r;
770     }
771    
772     /* converts an ieee double/binary64 to a double */
773     ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
774     ecb_function_ double
775     ecb_binary64_to_double (uint64_t x)
776     {
777     double r;
778    
779     #if ECB_STDFP
780     memcpy (&r, &x, 8);
781     #else
782     /* emulation, only works for normals and subnormals and +0 */
783     int neg = x >> 63;
784     int e = (x >> 52) & 0x7ffU;
785    
786     x &= 0xfffffffffffffU;
787    
788     if (e)
789     x |= 0x10000000000000U;
790     else
791     e = 1;
792    
793 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
794 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
795 root 1.104
796     r = neg ? -r : r;
797     #endif
798    
799     return r;
800 root 1.103 }
801    
802     #endif
803 root 1.102
804 root 1.1 #endif
805