ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/ecb.h
Revision: 1.24
Committed: Sat Jun 13 00:06:13 2015 UTC (8 years, 11 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.23: +19 -1 lines
Log Message: 
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 /*
2     * libecb - http://software.schmorp.de/pkg/libecb
3     *
4 root 1.23 * Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de>
5 root 1.1 * Copyright (©) 2011 Emanuele Giaquinta
6     * 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 1.21 *
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 1.12 /* 16 bits major, 16 bits minor */
45 root 1.23 #define ECB_VERSION 0x00010004
46 root 1.12
47 root 1.2 #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 1.3 #else /* _MSC_VER || __BORLANDC__ */
58 root 1.2 typedef signed __int64 int64_t;
59     typedef unsigned __int64 uint64_t;
60     #endif
61 root 1.12 #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 1.2 #else
71     #include <inttypes.h>
72 root 1.12 #if UINTMAX_MAX > 0xffffffffU
73     #define ECB_PTRSIZE 8
74     #else
75     #define ECB_PTRSIZE 4
76     #endif
77 root 1.2 #endif
78 root 1.1
79 root 1.17 /* work around x32 idiocy by defining proper macros */
80 root 1.19 #if __amd64 || __x86_64 || _M_AMD64 || _M_X64
81     #if _ILP32
82 root 1.17 #define ECB_AMD64_X32 1
83     #else
84     #define ECB_AMD64 1
85     #endif
86     #endif
87    
88 root 1.1 /* many compilers define _GNUC_ to some versions but then only implement
89     * what their idiot authors think are the "more important" extensions,
90 root 1.8 * causing enormous grief in return for some better fake benchmark numbers.
91 root 1.1 * or so.
92     * 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 1.23 #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     #endif
100    
101     #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     #if __clang__ && defined __has_extension
110     #define ECB_CLANG_EXTENSION(x) __has_extension (x)
111     #else
112     #define ECB_CLANG_EXTENSION(x) 0
113 root 1.1 #endif
114    
115 root 1.13 #define ECB_CPP (__cplusplus+0)
116     #define ECB_CPP11 (__cplusplus >= 201103L)
117    
118 root 1.15 #if ECB_CPP
119 root 1.19 #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 1.15 #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 1.4 /*****************************************************************************/
140    
141 root 1.8 /* 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 1.11 #if ECB_NO_THREADS
145 root 1.13 #define ECB_NO_SMP 1
146 root 1.11 #endif
147    
148 root 1.13 #if ECB_NO_SMP
149 root 1.8 #define ECB_MEMORY_FENCE do { } while (0)
150     #endif
151    
152 root 1.4 #ifndef ECB_MEMORY_FENCE
153 root 1.11 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
154     #if __i386 || __i386__
155 root 1.5 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
156 root 1.13 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
157     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
158 root 1.11 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
159 root 1.13 #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 1.8 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
163 root 1.13 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
164 root 1.11 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
165     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
166     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
167     #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
168 root 1.13 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
169     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
170 root 1.20 #elif __aarch64__
171     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
172 root 1.19 #elif (__sparc || __sparc__) && !__sparcv8
173 root 1.13 #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 1.11 #elif defined __s390__ || defined __s390x__
177     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
178     #elif defined __mips__
179 root 1.19 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
180 root 1.18 /* 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 1.11 #elif defined __alpha__
183 root 1.13 #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 1.19 #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 1.4 #endif
196     #endif
197     #endif
198    
199     #ifndef ECB_MEMORY_FENCE
200 root 1.13 #if ECB_GCC_VERSION(4,7)
201     /* see comment below (stdatomic.h) about the C11 memory model. */
202     #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
203 root 1.19 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
204     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
205 root 1.15
206 root 1.23 #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 1.15
212 root 1.13 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
213 root 1.4 #define ECB_MEMORY_FENCE __sync_synchronize ()
214 root 1.19 #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 1.7 #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 1.11 #elif defined _WIN32
226 root 1.6 #include <WinNT.h>
227 root 1.7 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
228 root 1.11 #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     #elif __xlC__
234     #define ECB_MEMORY_FENCE __sync ()
235 root 1.4 #endif
236     #endif
237    
238     #ifndef ECB_MEMORY_FENCE
239 root 1.13 #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     /* 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 1.19 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
250     /* for all __atomic_thread_fence's except seq_cst */
251 root 1.13 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
252     #endif
253     #endif
254    
255     #ifndef ECB_MEMORY_FENCE
256 root 1.8 #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    
268     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 root 1.4
273 root 1.11 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
274 root 1.4 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
275 root 1.8 #endif
276    
277 root 1.11 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
278 root 1.4 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
279     #endif
280    
281     /*****************************************************************************/
282    
283 root 1.23 #if ECB_CPP
284 root 1.2 #define ecb_inline static inline
285     #elif ECB_GCC_VERSION(2,5)
286     #define ecb_inline static __inline__
287     #elif ECB_C99
288     #define ecb_inline static inline
289     #else
290     #define ecb_inline static
291     #endif
292    
293     #if ECB_GCC_VERSION(3,3)
294     #define ecb_restrict __restrict__
295     #elif ECB_C99
296     #define ecb_restrict restrict
297 root 1.1 #else
298 root 1.2 #define ecb_restrict
299 root 1.1 #endif
300    
301 root 1.2 typedef int ecb_bool;
302    
303 root 1.1 #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 root 1.24 #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
308 root 1.1
309 root 1.2 #define ecb_function_ ecb_inline
310 root 1.1
311 root 1.23 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
312     #define ecb_attribute(attrlist) __attribute__ (attrlist)
313 root 1.2 #else
314     #define ecb_attribute(attrlist)
315 root 1.23 #endif
316 root 1.19
317 root 1.23 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
318     #define ecb_is_constant(expr) __builtin_constant_p (expr)
319     #else
320 root 1.19 /* possible C11 impl for integral types
321     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
322     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
323    
324 root 1.2 #define ecb_is_constant(expr) 0
325 root 1.23 #endif
326    
327     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
328     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
329     #else
330 root 1.2 #define ecb_expect(expr,value) (expr)
331 root 1.23 #endif
332    
333     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
334     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
335     #else
336 root 1.2 #define ecb_prefetch(addr,rw,locality)
337 root 1.1 #endif
338    
339     /* no emulation for ecb_decltype */
340 root 1.23 #if ECB_CPP11
341     // older implementations might have problems with decltype(x)::type, work around it
342     template<class T> struct ecb_decltype_t { typedef T type; };
343     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
344     #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
345     #define ecb_decltype(x) __typeof__ (x)
346 root 1.1 #endif
347    
348 root 1.22 #if _MSC_VER >= 1300
349 root 1.23 #define ecb_deprecated __declspec (deprecated)
350 root 1.22 #else
351     #define ecb_deprecated ecb_attribute ((__deprecated__))
352     #endif
353    
354 root 1.24 #if __MSC_VER >= 1500
355     #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
356     #elif ECB_GCC_VERSION(4,5)
357     #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
358     #else
359     #define ecb_deprecated_message(msg) ecb_deprecated
360     #endif
361    
362     #if _MSC_VER >= 1400
363     #define ecb_noinline __declspec (noinline)
364     #else
365     #define ecb_noinline ecb_attribute ((__noinline__))
366     #endif
367    
368 root 1.1 #define ecb_unused ecb_attribute ((__unused__))
369     #define ecb_const ecb_attribute ((__const__))
370     #define ecb_pure ecb_attribute ((__pure__))
371    
372 root 1.23 /* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
373     #if ECB_C11 || __IBMC_NORETURN
374     /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */
375 root 1.13 #define ecb_noreturn _Noreturn
376 root 1.24 #elif ECB_CPP11
377     #define ecb_noreturn [[noreturn]]
378     #elif _MSC_VER >= 1200
379     #define ecb_noreturn __declspec (noreturn)
380 root 1.13 #else
381     #define ecb_noreturn ecb_attribute ((__noreturn__))
382     #endif
383    
384 root 1.2 #if ECB_GCC_VERSION(4,3)
385     #define ecb_artificial ecb_attribute ((__artificial__))
386     #define ecb_hot ecb_attribute ((__hot__))
387     #define ecb_cold ecb_attribute ((__cold__))
388     #else
389     #define ecb_artificial
390     #define ecb_hot
391     #define ecb_cold
392     #endif
393    
394     /* put around conditional expressions if you are very sure that the */
395     /* expression is mostly true or mostly false. note that these return */
396     /* booleans, not the expression. */
397 root 1.1 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
398     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
399 root 1.2 /* for compatibility to the rest of the world */
400 root 1.1 #define ecb_likely(expr) ecb_expect_true (expr)
401     #define ecb_unlikely(expr) ecb_expect_false (expr)
402    
403     /* count trailing zero bits and count # of one bits */
404 root 1.23 #if ECB_GCC_VERSION(3,4) \
405     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
406     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
407     && ECB_CLANG_BUILTIN(__builtin_popcount))
408 root 1.2 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
409     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
410     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
411     #define ecb_ctz32(x) __builtin_ctz (x)
412     #define ecb_ctz64(x) __builtin_ctzll (x)
413     #define ecb_popcount32(x) __builtin_popcount (x)
414     /* no popcountll */
415 root 1.1 #else
416 root 1.23 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
417     ecb_function_ ecb_const int
418 root 1.2 ecb_ctz32 (uint32_t x)
419     {
420     int r = 0;
421    
422     x &= ~x + 1; /* this isolates the lowest bit */
423    
424 root 1.3 #if ECB_branchless_on_i386
425     r += !!(x & 0xaaaaaaaa) << 0;
426     r += !!(x & 0xcccccccc) << 1;
427     r += !!(x & 0xf0f0f0f0) << 2;
428     r += !!(x & 0xff00ff00) << 3;
429     r += !!(x & 0xffff0000) << 4;
430     #else
431 root 1.2 if (x & 0xaaaaaaaa) r += 1;
432     if (x & 0xcccccccc) r += 2;
433     if (x & 0xf0f0f0f0) r += 4;
434     if (x & 0xff00ff00) r += 8;
435     if (x & 0xffff0000) r += 16;
436 root 1.3 #endif
437 root 1.2
438     return r;
439     }
440    
441 root 1.23 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
442     ecb_function_ ecb_const int
443 root 1.2 ecb_ctz64 (uint64_t x)
444     {
445     int shift = x & 0xffffffffU ? 0 : 32;
446 root 1.3 return ecb_ctz32 (x >> shift) + shift;
447 root 1.2 }
448    
449 root 1.23 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
450     ecb_function_ ecb_const int
451 root 1.2 ecb_popcount32 (uint32_t x)
452     {
453     x -= (x >> 1) & 0x55555555;
454     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
455     x = ((x >> 4) + x) & 0x0f0f0f0f;
456     x *= 0x01010101;
457    
458     return x >> 24;
459     }
460    
461 root 1.23 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
462     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
463 root 1.2 {
464 root 1.3 int r = 0;
465 root 1.2
466 root 1.3 if (x >> 16) { x >>= 16; r += 16; }
467     if (x >> 8) { x >>= 8; r += 8; }
468     if (x >> 4) { x >>= 4; r += 4; }
469     if (x >> 2) { x >>= 2; r += 2; }
470     if (x >> 1) { r += 1; }
471 root 1.2
472     return r;
473     }
474    
475 root 1.23 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
476     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
477 root 1.2 {
478 root 1.3 int r = 0;
479 root 1.2
480 root 1.3 if (x >> 32) { x >>= 32; r += 32; }
481 root 1.1
482 root 1.3 return r + ecb_ld32 (x);
483 root 1.2 }
484     #endif
485 root 1.1
486 root 1.23 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
487     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
488     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
489     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
490 root 1.12
491 root 1.23 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
492     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
493 root 1.11 {
494     return ( (x * 0x0802U & 0x22110U)
495 root 1.23 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
496 root 1.11 }
497    
498 root 1.23 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
499     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
500 root 1.11 {
501     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
502     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
503     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
504     x = ( x >> 8 ) | ( x << 8);
505    
506     return x;
507     }
508    
509 root 1.23 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
510     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
511 root 1.11 {
512     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
513     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
514     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
515     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
516     x = ( x >> 16 ) | ( x << 16);
517    
518     return x;
519     }
520    
521 root 1.2 /* popcount64 is only available on 64 bit cpus as gcc builtin */
522     /* so for this version we are lazy */
523 root 1.23 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
524     ecb_function_ ecb_const int
525 root 1.2 ecb_popcount64 (uint64_t x)
526 root 1.1 {
527 root 1.2 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
528 root 1.1 }
529    
530 root 1.23 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
531     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
532     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
533     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
534     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
535     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
536     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
537     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
538    
539     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
540     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
541     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
542     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
543     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
544     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
545     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
546     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
547 root 1.3
548 root 1.23 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
549 root 1.2 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
550     #define ecb_bswap32(x) __builtin_bswap32 (x)
551     #define ecb_bswap64(x) __builtin_bswap64 (x)
552 root 1.1 #else
553 root 1.23 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
554     ecb_function_ ecb_const uint16_t
555 root 1.3 ecb_bswap16 (uint16_t x)
556 root 1.2 {
557 root 1.3 return ecb_rotl16 (x, 8);
558 root 1.2 }
559    
560 root 1.23 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
561     ecb_function_ ecb_const uint32_t
562 root 1.2 ecb_bswap32 (uint32_t x)
563     {
564 root 1.3 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
565 root 1.2 }
566    
567 root 1.23 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
568     ecb_function_ ecb_const uint64_t
569 root 1.2 ecb_bswap64 (uint64_t x)
570     {
571 root 1.3 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
572 root 1.2 }
573 root 1.1 #endif
574    
575 root 1.23 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
576 root 1.2 #define ecb_unreachable() __builtin_unreachable ()
577 root 1.1 #else
578 root 1.2 /* this seems to work fine, but gcc always emits a warning for it :/ */
579 root 1.23 ecb_inline ecb_noreturn void ecb_unreachable (void);
580     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
581 root 1.1 #endif
582    
583 root 1.2 /* try to tell the compiler that some condition is definitely true */
584 root 1.14 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
585 root 1.2
586 root 1.23 ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);
587     ecb_inline ecb_const unsigned char
588 root 1.1 ecb_byteorder_helper (void)
589     {
590 root 1.14 /* the union code still generates code under pressure in gcc, */
591 root 1.16 /* but less than using pointers, and always seems to */
592 root 1.14 /* successfully return a constant. */
593     /* the reason why we have this horrible preprocessor mess */
594     /* is to avoid it in all cases, at least on common architectures */
595 root 1.16 /* or when using a recent enough gcc version (>= 4.6) */
596 root 1.14 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
597     return 0x44;
598     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
599     return 0x44;
600     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
601 root 1.16 return 0x11;
602 root 1.14 #else
603     union
604     {
605     uint32_t i;
606     uint8_t c;
607     } u = { 0x11223344 };
608     return u.c;
609     #endif
610 root 1.1 }
611    
612 root 1.23 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
613     ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
614     ecb_inline ecb_const ecb_bool ecb_little_endian (void);
615     ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
616 root 1.1
617 root 1.2 #if ECB_GCC_VERSION(3,0) || ECB_C99
618     #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
619 root 1.1 #else
620 root 1.2 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
621 root 1.1 #endif
622    
623 root 1.23 #if ECB_CPP
624 root 1.10 template<typename T>
625     static inline T ecb_div_rd (T val, T div)
626     {
627     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
628     }
629     template<typename T>
630     static inline T ecb_div_ru (T val, T div)
631     {
632     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
633     }
634     #else
635     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
636     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
637     #endif
638    
639 root 1.1 #if ecb_cplusplus_does_not_suck
640 root 1.2 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
641     template<typename T, int N>
642     static inline int ecb_array_length (const T (&arr)[N])
643     {
644     return N;
645     }
646 root 1.1 #else
647 root 1.2 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
648 root 1.1 #endif
649    
650 root 1.15 /*******************************************************************************/
651     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
652    
653     /* basically, everything uses "ieee pure-endian" floating point numbers */
654     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
655     #if 0 \
656     || __i386 || __i386__ \
657     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
658     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
659     || defined __s390__ || defined __s390x__ \
660     || defined __mips__ \
661     || defined __alpha__ \
662     || defined __hppa__ \
663     || defined __ia64__ \
664 root 1.19 || defined __m68k__ \
665     || defined __m88k__ \
666     || defined __sh__ \
667 root 1.20 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \
668     || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
669     || defined __aarch64__
670 root 1.15 #define ECB_STDFP 1
671     #include <string.h> /* for memcpy */
672     #else
673     #define ECB_STDFP 0
674     #endif
675    
676     #ifndef ECB_NO_LIBM
677    
678 root 1.19 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
679    
680     /* only the oldest of old doesn't have this one. solaris. */
681     #ifdef INFINITY
682     #define ECB_INFINITY INFINITY
683     #else
684     #define ECB_INFINITY HUGE_VAL
685     #endif
686    
687     #ifdef NAN
688     #define ECB_NAN NAN
689     #else
690     #define ECB_NAN ECB_INFINITY
691     #endif
692    
693 root 1.23 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
694     #define ecb_ldexpf(x,e) ldexpf ((x), (e))
695     #else
696     #define ecb_ldexpf(x,e) (float) ldexp ((x), (e))
697     #endif
698    
699 root 1.19 /* converts an ieee half/binary16 to a float */
700 root 1.23 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
701     ecb_function_ ecb_const float
702 root 1.19 ecb_binary16_to_float (uint16_t x)
703     {
704     int e = (x >> 10) & 0x1f;
705     int m = x & 0x3ff;
706     float r;
707    
708 root 1.23 if (!e ) r = ecb_ldexpf (m , -24);
709     else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
710 root 1.19 else if (m ) r = ECB_NAN;
711     else r = ECB_INFINITY;
712    
713     return x & 0x8000 ? -r : r;
714     }
715    
716 root 1.15 /* convert a float to ieee single/binary32 */
717 root 1.23 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
718     ecb_function_ ecb_const uint32_t
719 root 1.15 ecb_float_to_binary32 (float x)
720     {
721     uint32_t r;
722    
723     #if ECB_STDFP
724     memcpy (&r, &x, 4);
725     #else
726     /* slow emulation, works for anything but -0 */
727     uint32_t m;
728     int e;
729    
730     if (x == 0e0f ) return 0x00000000U;
731     if (x > +3.40282346638528860e+38f) return 0x7f800000U;
732     if (x < -3.40282346638528860e+38f) return 0xff800000U;
733     if (x != x ) return 0x7fbfffffU;
734    
735     m = frexpf (x, &e) * 0x1000000U;
736    
737     r = m & 0x80000000U;
738    
739     if (r)
740     m = -m;
741    
742     if (e <= -126)
743     {
744     m &= 0xffffffU;
745     m >>= (-125 - e);
746     e = -126;
747     }
748    
749     r |= (e + 126) << 23;
750     r |= m & 0x7fffffU;
751     #endif
752    
753     return r;
754     }
755    
756     /* converts an ieee single/binary32 to a float */
757 root 1.23 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
758     ecb_function_ ecb_const float
759 root 1.15 ecb_binary32_to_float (uint32_t x)
760     {
761     float r;
762    
763     #if ECB_STDFP
764     memcpy (&r, &x, 4);
765     #else
766     /* emulation, only works for normals and subnormals and +0 */
767     int neg = x >> 31;
768     int e = (x >> 23) & 0xffU;
769    
770     x &= 0x7fffffU;
771    
772     if (e)
773     x |= 0x800000U;
774     else
775     e = 1;
776    
777     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
778 root 1.23 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
779 root 1.15
780     r = neg ? -r : r;
781     #endif
782    
783     return r;
784     }
785    
786     /* convert a double to ieee double/binary64 */
787 root 1.23 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
788     ecb_function_ ecb_const uint64_t
789 root 1.15 ecb_double_to_binary64 (double x)
790     {
791     uint64_t r;
792    
793     #if ECB_STDFP
794     memcpy (&r, &x, 8);
795     #else
796     /* slow emulation, works for anything but -0 */
797     uint64_t m;
798     int e;
799    
800     if (x == 0e0 ) return 0x0000000000000000U;
801     if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
802     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
803     if (x != x ) return 0X7ff7ffffffffffffU;
804    
805     m = frexp (x, &e) * 0x20000000000000U;
806    
807     r = m & 0x8000000000000000;;
808    
809     if (r)
810     m = -m;
811    
812     if (e <= -1022)
813     {
814     m &= 0x1fffffffffffffU;
815     m >>= (-1021 - e);
816     e = -1022;
817     }
818    
819     r |= ((uint64_t)(e + 1022)) << 52;
820     r |= m & 0xfffffffffffffU;
821     #endif
822    
823     return r;
824     }
825    
826     /* converts an ieee double/binary64 to a double */
827 root 1.23 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
828     ecb_function_ ecb_const double
829 root 1.15 ecb_binary64_to_double (uint64_t x)
830     {
831     double r;
832    
833     #if ECB_STDFP
834     memcpy (&r, &x, 8);
835     #else
836     /* emulation, only works for normals and subnormals and +0 */
837     int neg = x >> 63;
838     int e = (x >> 52) & 0x7ffU;
839    
840     x &= 0xfffffffffffffU;
841    
842     if (e)
843     x |= 0x10000000000000U;
844     else
845     e = 1;
846    
847     /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
848     r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
849    
850     r = neg ? -r : r;
851     #endif
852    
853     return r;
854     }
855    
856     #endif
857    
858 root 1.1 #endif
859