ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/cvsroot/Async-Interrupt/ecb.h
Revision: 1.5
Committed: Thu Jan 18 16:43:35 2018 UTC (8 years, 7 months ago) by root
Content type: text/plain
Branch: MAIN
CVS Tags: rel-1_22
Changes since 1.4: +162 -30 lines
Log Message: 
1.22

File Contents

# User Rev Content
1 root 1.1 /*
2     * libecb - http://software.schmorp.de/pkg/libecb
3     *
4 root 1.4 * 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 1.4 *
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 1.2 /* 16 bits major, 16 bits minor */
45 root 1.5 #define ECB_VERSION 0x00010005
46 root 1.2
47 root 1.1 #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     #else /* _MSC_VER || __BORLANDC__ */
58     typedef signed __int64 int64_t;
59     typedef unsigned __int64 uint64_t;
60     #endif
61 root 1.2 #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 1.1 #else
71     #include <inttypes.h>
72 root 1.2 #if UINTMAX_MAX > 0xffffffffU
73     #define ECB_PTRSIZE 8
74     #else
75     #define ECB_PTRSIZE 4
76     #endif
77     #endif
78    
79 root 1.5 #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
80     #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
81    
82 root 1.2 /* work around x32 idiocy by defining proper macros */
83 root 1.5 #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
84 root 1.3 #if _ILP32
85 root 1.2 #define ECB_AMD64_X32 1
86     #else
87     #define ECB_AMD64 1
88     #endif
89 root 1.1 #endif
90    
91     /* many compilers define _GNUC_ to some versions but then only implement
92     * what their idiot authors think are the "more important" extensions,
93     * causing enormous grief in return for some better fake benchmark numbers.
94     * or so.
95     * we try to detect these and simply assume they are not gcc - if they have
96     * an issue with that they should have done it right in the first place.
97     */
98 root 1.4 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
99     #define ECB_GCC_VERSION(major,minor) 0
100     #else
101     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
102     #endif
103    
104     #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
105    
106     #if __clang__ && defined __has_builtin
107     #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
108     #else
109     #define ECB_CLANG_BUILTIN(x) 0
110     #endif
111    
112     #if __clang__ && defined __has_extension
113     #define ECB_CLANG_EXTENSION(x) __has_extension (x)
114     #else
115     #define ECB_CLANG_EXTENSION(x) 0
116 root 1.1 #endif
117    
118 root 1.2 #define ECB_CPP (__cplusplus+0)
119     #define ECB_CPP11 (__cplusplus >= 201103L)
120    
121     #if ECB_CPP
122 root 1.3 #define ECB_C 0
123     #define ECB_STDC_VERSION 0
124     #else
125     #define ECB_C 1
126     #define ECB_STDC_VERSION __STDC_VERSION__
127     #endif
128    
129     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
130     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
131    
132     #if ECB_CPP
133 root 1.2 #define ECB_EXTERN_C extern "C"
134     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
135     #define ECB_EXTERN_C_END }
136     #else
137     #define ECB_EXTERN_C extern
138     #define ECB_EXTERN_C_BEG
139     #define ECB_EXTERN_C_END
140     #endif
141    
142 root 1.1 /*****************************************************************************/
143    
144     /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
145     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
146    
147     #if ECB_NO_THREADS
148 root 1.2 #define ECB_NO_SMP 1
149 root 1.1 #endif
150    
151 root 1.2 #if ECB_NO_SMP
152 root 1.1 #define ECB_MEMORY_FENCE do { } while (0)
153     #endif
154    
155 root 1.5 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
156     #if __xlC__ && ECB_CPP
157     #include <builtins.h>
158     #endif
159    
160 root 1.1 #ifndef ECB_MEMORY_FENCE
161     #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
162     #if __i386 || __i386__
163     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
164 root 1.2 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
165     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
166 root 1.5 #elif ECB_GCC_AMD64
167 root 1.2 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
168     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
169     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
170 root 1.1 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
171 root 1.2 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
172 root 1.1 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
173     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
174     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
175     #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
176 root 1.2 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
177     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
178 root 1.3 #elif __aarch64__
179     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
180 root 1.5 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
181 root 1.2 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
182     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
183     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
184 root 1.1 #elif defined __s390__ || defined __s390x__
185     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
186     #elif defined __mips__
187 root 1.3 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
188     /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
189     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
190 root 1.1 #elif defined __alpha__
191 root 1.2 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
192     #elif defined __hppa__
193     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
194     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
195     #elif defined __ia64__
196     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
197 root 1.3 #elif defined __m68k__
198     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
199     #elif defined __m88k__
200     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
201     #elif defined __sh__
202     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
203 root 1.1 #endif
204     #endif
205     #endif
206    
207     #ifndef ECB_MEMORY_FENCE
208 root 1.2 #if ECB_GCC_VERSION(4,7)
209     /* see comment below (stdatomic.h) about the C11 memory model. */
210     #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
211 root 1.3 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
212     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
213 root 1.2
214 root 1.4 #elif ECB_CLANG_EXTENSION(c_atomic)
215     /* see comment below (stdatomic.h) about the C11 memory model. */
216     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
217     #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
218     #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
219 root 1.2
220     #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
221 root 1.1 #define ECB_MEMORY_FENCE __sync_synchronize ()
222 root 1.3 #elif _MSC_VER >= 1500 /* VC++ 2008 */
223     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
224     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
225     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
226     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
227     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
228 root 1.1 #elif _MSC_VER >= 1400 /* VC++ 2005 */
229     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
230     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
231     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
232     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
233     #elif defined _WIN32
234     #include <WinNT.h>
235     #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
236     #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
237     #include <mbarrier.h>
238     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
239     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
240     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
241     #elif __xlC__
242     #define ECB_MEMORY_FENCE __sync ()
243     #endif
244     #endif
245    
246     #ifndef ECB_MEMORY_FENCE
247 root 1.2 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
248     /* we assume that these memory fences work on all variables/all memory accesses, */
249     /* not just C11 atomics and atomic accesses */
250     #include <stdatomic.h>
251     /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
252     /* any fence other than seq_cst, which isn't very efficient for us. */
253     /* Why that is, we don't know - either the C11 memory model is quite useless */
254     /* for most usages, or gcc and clang have a bug */
255     /* I *currently* lean towards the latter, and inefficiently implement */
256     /* all three of ecb's fences as a seq_cst fence */
257 root 1.3 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
258     /* for all __atomic_thread_fence's except seq_cst */
259 root 1.2 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
260     #endif
261     #endif
262    
263     #ifndef ECB_MEMORY_FENCE
264 root 1.1 #if !ECB_AVOID_PTHREADS
265     /*
266     * if you get undefined symbol references to pthread_mutex_lock,
267     * or failure to find pthread.h, then you should implement
268     * the ECB_MEMORY_FENCE operations for your cpu/compiler
269     * OR provide pthread.h and link against the posix thread library
270     * of your system.
271     */
272     #include <pthread.h>
273     #define ECB_NEEDS_PTHREADS 1
274     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
275    
276     static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
277     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
278     #endif
279     #endif
280    
281     #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
282     #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
283     #endif
284    
285     #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
286     #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
287     #endif
288    
289     /*****************************************************************************/
290    
291 root 1.4 #if ECB_CPP
292 root 1.1 #define ecb_inline static inline
293     #elif ECB_GCC_VERSION(2,5)
294     #define ecb_inline static __inline__
295     #elif ECB_C99
296     #define ecb_inline static inline
297     #else
298     #define ecb_inline static
299     #endif
300    
301     #if ECB_GCC_VERSION(3,3)
302     #define ecb_restrict __restrict__
303     #elif ECB_C99
304     #define ecb_restrict restrict
305     #else
306     #define ecb_restrict
307     #endif
308    
309     typedef int ecb_bool;
310    
311     #define ECB_CONCAT_(a, b) a ## b
312     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
313     #define ECB_STRINGIFY_(a) # a
314     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
315 root 1.5 #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
316 root 1.1
317     #define ecb_function_ ecb_inline
318    
319 root 1.4 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
320     #define ecb_attribute(attrlist) __attribute__ (attrlist)
321 root 1.1 #else
322     #define ecb_attribute(attrlist)
323 root 1.4 #endif
324 root 1.3
325 root 1.4 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
326     #define ecb_is_constant(expr) __builtin_constant_p (expr)
327     #else
328 root 1.3 /* possible C11 impl for integral types
329     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
330     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
331    
332 root 1.1 #define ecb_is_constant(expr) 0
333 root 1.4 #endif
334    
335     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
336     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
337     #else
338 root 1.1 #define ecb_expect(expr,value) (expr)
339 root 1.4 #endif
340    
341     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
342     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
343     #else
344 root 1.1 #define ecb_prefetch(addr,rw,locality)
345     #endif
346    
347     /* no emulation for ecb_decltype */
348 root 1.4 #if ECB_CPP11
349     // older implementations might have problems with decltype(x)::type, work around it
350     template<class T> struct ecb_decltype_t { typedef T type; };
351     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
352     #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
353     #define ecb_decltype(x) __typeof__ (x)
354     #endif
355    
356     #if _MSC_VER >= 1300
357     #define ecb_deprecated __declspec (deprecated)
358     #else
359     #define ecb_deprecated ecb_attribute ((__deprecated__))
360 root 1.1 #endif
361    
362 root 1.5 #if _MSC_VER >= 1500
363     #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
364     #elif ECB_GCC_VERSION(4,5)
365     #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
366     #else
367     #define ecb_deprecated_message(msg) ecb_deprecated
368     #endif
369    
370     #if _MSC_VER >= 1400
371     #define ecb_noinline __declspec (noinline)
372     #else
373     #define ecb_noinline ecb_attribute ((__noinline__))
374     #endif
375    
376 root 1.1 #define ecb_unused ecb_attribute ((__unused__))
377     #define ecb_const ecb_attribute ((__const__))
378     #define ecb_pure ecb_attribute ((__pure__))
379    
380 root 1.4 #if ECB_C11 || __IBMC_NORETURN
381 root 1.5 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
382 root 1.2 #define ecb_noreturn _Noreturn
383 root 1.5 #elif ECB_CPP11
384     #define ecb_noreturn [[noreturn]]
385     #elif _MSC_VER >= 1200
386     /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
387     #define ecb_noreturn __declspec (noreturn)
388 root 1.2 #else
389     #define ecb_noreturn ecb_attribute ((__noreturn__))
390     #endif
391    
392 root 1.1 #if ECB_GCC_VERSION(4,3)
393     #define ecb_artificial ecb_attribute ((__artificial__))
394     #define ecb_hot ecb_attribute ((__hot__))
395     #define ecb_cold ecb_attribute ((__cold__))
396     #else
397     #define ecb_artificial
398     #define ecb_hot
399     #define ecb_cold
400     #endif
401    
402     /* put around conditional expressions if you are very sure that the */
403     /* expression is mostly true or mostly false. note that these return */
404     /* booleans, not the expression. */
405     #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
406     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
407     /* for compatibility to the rest of the world */
408     #define ecb_likely(expr) ecb_expect_true (expr)
409     #define ecb_unlikely(expr) ecb_expect_false (expr)
410    
411     /* count trailing zero bits and count # of one bits */
412 root 1.4 #if ECB_GCC_VERSION(3,4) \
413     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
414     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
415     && ECB_CLANG_BUILTIN(__builtin_popcount))
416 root 1.1 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
417     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
418     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
419     #define ecb_ctz32(x) __builtin_ctz (x)
420     #define ecb_ctz64(x) __builtin_ctzll (x)
421     #define ecb_popcount32(x) __builtin_popcount (x)
422     /* no popcountll */
423     #else
424 root 1.4 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
425     ecb_function_ ecb_const int
426 root 1.1 ecb_ctz32 (uint32_t x)
427     {
428     int r = 0;
429    
430     x &= ~x + 1; /* this isolates the lowest bit */
431    
432     #if ECB_branchless_on_i386
433     r += !!(x & 0xaaaaaaaa) << 0;
434     r += !!(x & 0xcccccccc) << 1;
435     r += !!(x & 0xf0f0f0f0) << 2;
436     r += !!(x & 0xff00ff00) << 3;
437     r += !!(x & 0xffff0000) << 4;
438     #else
439     if (x & 0xaaaaaaaa) r += 1;
440     if (x & 0xcccccccc) r += 2;
441     if (x & 0xf0f0f0f0) r += 4;
442     if (x & 0xff00ff00) r += 8;
443     if (x & 0xffff0000) r += 16;
444     #endif
445    
446     return r;
447     }
448    
449 root 1.4 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
450     ecb_function_ ecb_const int
451 root 1.1 ecb_ctz64 (uint64_t x)
452     {
453 root 1.5 int shift = x & 0xffffffff ? 0 : 32;
454 root 1.1 return ecb_ctz32 (x >> shift) + shift;
455     }
456    
457 root 1.4 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
458     ecb_function_ ecb_const int
459 root 1.1 ecb_popcount32 (uint32_t x)
460     {
461     x -= (x >> 1) & 0x55555555;
462     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
463     x = ((x >> 4) + x) & 0x0f0f0f0f;
464     x *= 0x01010101;
465    
466     return x >> 24;
467     }
468    
469 root 1.4 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
470     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
471 root 1.1 {
472     int r = 0;
473    
474     if (x >> 16) { x >>= 16; r += 16; }
475     if (x >> 8) { x >>= 8; r += 8; }
476     if (x >> 4) { x >>= 4; r += 4; }
477     if (x >> 2) { x >>= 2; r += 2; }
478     if (x >> 1) { r += 1; }
479    
480     return r;
481     }
482    
483 root 1.4 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
484     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
485 root 1.1 {
486     int r = 0;
487    
488     if (x >> 32) { x >>= 32; r += 32; }
489    
490     return r + ecb_ld32 (x);
491     }
492     #endif
493    
494 root 1.4 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
495     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
496     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
497     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
498 root 1.2
499 root 1.4 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
500     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
501 root 1.1 {
502     return ( (x * 0x0802U & 0x22110U)
503 root 1.4 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
504 root 1.1 }
505    
506 root 1.4 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
507     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
508 root 1.1 {
509     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
510     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
511     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
512     x = ( x >> 8 ) | ( x << 8);
513    
514     return x;
515     }
516    
517 root 1.4 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
518     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
519 root 1.1 {
520     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
521     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
522     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
523     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
524     x = ( x >> 16 ) | ( x << 16);
525    
526     return x;
527     }
528    
529     /* popcount64 is only available on 64 bit cpus as gcc builtin */
530     /* so for this version we are lazy */
531 root 1.4 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
532     ecb_function_ ecb_const int
533 root 1.1 ecb_popcount64 (uint64_t x)
534     {
535     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
536     }
537    
538 root 1.4 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
539     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
540     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
541     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
542     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
543     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
544     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
545     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
546    
547     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
548     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
549     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
550     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
551     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
552     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
553     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
554     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
555 root 1.1
556 root 1.4 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
557 root 1.5 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
558     #define ecb_bswap16(x) __builtin_bswap16 (x)
559     #else
560 root 1.1 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
561 root 1.5 #endif
562 root 1.1 #define ecb_bswap32(x) __builtin_bswap32 (x)
563     #define ecb_bswap64(x) __builtin_bswap64 (x)
564 root 1.5 #elif _MSC_VER
565     #include <stdlib.h>
566     #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
567     #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
568     #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
569 root 1.1 #else
570 root 1.4 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
571     ecb_function_ ecb_const uint16_t
572 root 1.1 ecb_bswap16 (uint16_t x)
573     {
574     return ecb_rotl16 (x, 8);
575     }
576    
577 root 1.4 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
578     ecb_function_ ecb_const uint32_t
579 root 1.1 ecb_bswap32 (uint32_t x)
580     {
581     return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
582     }
583    
584 root 1.4 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
585     ecb_function_ ecb_const uint64_t
586 root 1.1 ecb_bswap64 (uint64_t x)
587     {
588     return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
589     }
590     #endif
591    
592 root 1.4 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
593 root 1.1 #define ecb_unreachable() __builtin_unreachable ()
594     #else
595     /* this seems to work fine, but gcc always emits a warning for it :/ */
596 root 1.4 ecb_inline ecb_noreturn void ecb_unreachable (void);
597     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
598 root 1.1 #endif
599    
600     /* try to tell the compiler that some condition is definitely true */
601 root 1.2 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
602 root 1.1
603 root 1.4 ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);
604     ecb_inline ecb_const unsigned char
605 root 1.1 ecb_byteorder_helper (void)
606     {
607 root 1.2 /* the union code still generates code under pressure in gcc, */
608     /* but less than using pointers, and always seems to */
609     /* successfully return a constant. */
610     /* the reason why we have this horrible preprocessor mess */
611     /* is to avoid it in all cases, at least on common architectures */
612     /* or when using a recent enough gcc version (>= 4.6) */
613 root 1.5 #if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64
614 root 1.2 return 0x44;
615     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
616     return 0x44;
617     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
618     return 0x11;
619     #else
620     union
621     {
622     uint32_t i;
623     uint8_t c;
624     } u = { 0x11223344 };
625     return u.c;
626     #endif
627 root 1.1 }
628    
629 root 1.4 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
630     ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
631     ecb_inline ecb_const ecb_bool ecb_little_endian (void);
632     ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
633 root 1.1
634     #if ECB_GCC_VERSION(3,0) || ECB_C99
635     #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
636     #else
637     #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
638     #endif
639    
640 root 1.4 #if ECB_CPP
641 root 1.1 template<typename T>
642     static inline T ecb_div_rd (T val, T div)
643     {
644     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
645     }
646     template<typename T>
647     static inline T ecb_div_ru (T val, T div)
648     {
649     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
650     }
651     #else
652     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
653     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
654     #endif
655    
656     #if ecb_cplusplus_does_not_suck
657     /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
658     template<typename T, int N>
659     static inline int ecb_array_length (const T (&arr)[N])
660     {
661     return N;
662     }
663     #else
664     #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
665     #endif
666    
667 root 1.5 ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
668     ecb_function_ ecb_const uint32_t
669     ecb_binary16_to_binary32 (uint32_t x)
670     {
671     unsigned int s = (x & 0x8000) << (31 - 15);
672     int e = (x >> 10) & 0x001f;
673     unsigned int m = x & 0x03ff;
674    
675     if (ecb_expect_false (e == 31))
676     /* infinity or NaN */
677     e = 255 - (127 - 15);
678     else if (ecb_expect_false (!e))
679     {
680     if (ecb_expect_true (!m))
681     /* zero, handled by code below by forcing e to 0 */
682     e = 0 - (127 - 15);
683     else
684     {
685     /* subnormal, renormalise */
686     unsigned int s = 10 - ecb_ld32 (m);
687    
688     m = (m << s) & 0x3ff; /* mask implicit bit */
689     e -= s - 1;
690     }
691     }
692    
693     /* e and m now are normalised, or zero, (or inf or nan) */
694     e += 127 - 15;
695    
696     return s | (e << 23) | (m << (23 - 10));
697     }
698    
699     ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
700     ecb_function_ ecb_const uint16_t
701     ecb_binary32_to_binary16 (uint32_t x)
702     {
703     unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
704     unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
705     unsigned int m = x & 0x007fffff;
706    
707     x &= 0x7fffffff;
708    
709     /* if it's within range of binary16 normals, use fast path */
710     if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
711     {
712     /* mantissa round-to-even */
713     m += 0x00000fff + ((m >> (23 - 10)) & 1);
714    
715     /* handle overflow */
716     if (ecb_expect_false (m >= 0x00800000))
717     {
718     m >>= 1;
719     e += 1;
720     }
721    
722     return s | (e << 10) | (m >> (23 - 10));
723     }
724    
725     /* handle large numbers and infinity */
726     if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
727     return s | 0x7c00;
728    
729     /* handle zero, subnormals and small numbers */
730     if (ecb_expect_true (x < 0x38800000))
731     {
732     /* zero */
733     if (ecb_expect_true (!x))
734     return s;
735    
736     /* handle subnormals */
737    
738     /* too small, will be zero */
739     if (e < (14 - 24)) /* might not be sharp, but is good enough */
740     return s;
741    
742     m |= 0x00800000; /* make implicit bit explicit */
743    
744     /* very tricky - we need to round to the nearest e (+10) bit value */
745     {
746     unsigned int bits = 14 - e;
747     unsigned int half = (1 << (bits - 1)) - 1;
748     unsigned int even = (m >> bits) & 1;
749    
750     /* if this overflows, we will end up with a normalised number */
751     m = (m + half + even) >> bits;
752     }
753    
754     return s | m;
755     }
756    
757     /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
758     m >>= 13;
759    
760     return s | 0x7c00 | m | !m;
761     }
762    
763 root 1.2 /*******************************************************************************/
764     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
765    
766     /* basically, everything uses "ieee pure-endian" floating point numbers */
767     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
768     #if 0 \
769     || __i386 || __i386__ \
770 root 1.5 || ECB_GCC_AMD64 \
771 root 1.2 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
772     || defined __s390__ || defined __s390x__ \
773     || defined __mips__ \
774     || defined __alpha__ \
775     || defined __hppa__ \
776     || defined __ia64__ \
777 root 1.3 || defined __m68k__ \
778     || defined __m88k__ \
779     || defined __sh__ \
780 root 1.5 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
781 root 1.3 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
782     || defined __aarch64__
783 root 1.2 #define ECB_STDFP 1
784     #include <string.h> /* for memcpy */
785     #else
786     #define ECB_STDFP 0
787     #endif
788    
789     #ifndef ECB_NO_LIBM
790    
791 root 1.3 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
792    
793     /* only the oldest of old doesn't have this one. solaris. */
794     #ifdef INFINITY
795     #define ECB_INFINITY INFINITY
796     #else
797     #define ECB_INFINITY HUGE_VAL
798     #endif
799    
800     #ifdef NAN
801     #define ECB_NAN NAN
802     #else
803     #define ECB_NAN ECB_INFINITY
804     #endif
805    
806 root 1.4 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
807     #define ecb_ldexpf(x,e) ldexpf ((x), (e))
808 root 1.5 #define ecb_frexpf(x,e) frexpf ((x), (e))
809 root 1.4 #else
810 root 1.5 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
811     #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
812 root 1.4 #endif
813    
814 root 1.2 /* convert a float to ieee single/binary32 */
815 root 1.4 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
816     ecb_function_ ecb_const uint32_t
817 root 1.2 ecb_float_to_binary32 (float x)
818     {
819     uint32_t r;
820    
821     #if ECB_STDFP
822     memcpy (&r, &x, 4);
823     #else
824     /* slow emulation, works for anything but -0 */
825     uint32_t m;
826     int e;
827    
828     if (x == 0e0f ) return 0x00000000U;
829     if (x > +3.40282346638528860e+38f) return 0x7f800000U;
830     if (x < -3.40282346638528860e+38f) return 0xff800000U;
831     if (x != x ) return 0x7fbfffffU;
832    
833 root 1.5 m = ecb_frexpf (x, &e) * 0x1000000U;
834 root 1.2
835     r = m & 0x80000000U;
836    
837     if (r)
838     m = -m;
839    
840     if (e <= -126)
841     {
842     m &= 0xffffffU;
843     m >>= (-125 - e);
844     e = -126;
845     }
846    
847     r |= (e + 126) << 23;
848     r |= m & 0x7fffffU;
849     #endif
850    
851     return r;
852     }
853    
854     /* converts an ieee single/binary32 to a float */
855 root 1.4 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
856     ecb_function_ ecb_const float
857 root 1.2 ecb_binary32_to_float (uint32_t x)
858     {
859     float r;
860    
861     #if ECB_STDFP
862     memcpy (&r, &x, 4);
863     #else
864     /* emulation, only works for normals and subnormals and +0 */
865     int neg = x >> 31;
866     int e = (x >> 23) & 0xffU;
867    
868     x &= 0x7fffffU;
869    
870     if (e)
871     x |= 0x800000U;
872     else
873     e = 1;
874    
875     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
876 root 1.4 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
877 root 1.2
878     r = neg ? -r : r;
879     #endif
880    
881     return r;
882     }
883    
884     /* convert a double to ieee double/binary64 */
885 root 1.4 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
886     ecb_function_ ecb_const uint64_t
887 root 1.2 ecb_double_to_binary64 (double x)
888     {
889     uint64_t r;
890    
891     #if ECB_STDFP
892     memcpy (&r, &x, 8);
893     #else
894     /* slow emulation, works for anything but -0 */
895     uint64_t m;
896     int e;
897    
898     if (x == 0e0 ) return 0x0000000000000000U;
899     if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
900     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
901     if (x != x ) return 0X7ff7ffffffffffffU;
902    
903     m = frexp (x, &e) * 0x20000000000000U;
904    
905     r = m & 0x8000000000000000;;
906    
907     if (r)
908     m = -m;
909    
910     if (e <= -1022)
911     {
912     m &= 0x1fffffffffffffU;
913     m >>= (-1021 - e);
914     e = -1022;
915     }
916    
917     r |= ((uint64_t)(e + 1022)) << 52;
918     r |= m & 0xfffffffffffffU;
919     #endif
920    
921     return r;
922     }
923    
924     /* converts an ieee double/binary64 to a double */
925 root 1.4 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
926     ecb_function_ ecb_const double
927 root 1.2 ecb_binary64_to_double (uint64_t x)
928     {
929     double r;
930    
931     #if ECB_STDFP
932     memcpy (&r, &x, 8);
933     #else
934     /* emulation, only works for normals and subnormals and +0 */
935     int neg = x >> 63;
936     int e = (x >> 52) & 0x7ffU;
937    
938     x &= 0xfffffffffffffU;
939    
940     if (e)
941     x |= 0x10000000000000U;
942     else
943     e = 1;
944    
945     /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
946     r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
947    
948     r = neg ? -r : r;
949     #endif
950    
951     return r;
952     }
953    
954 root 1.5 /* convert a float to ieee half/binary16 */
955     ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
956     ecb_function_ ecb_const uint16_t
957     ecb_float_to_binary16 (float x)
958     {
959     return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
960     }
961    
962     /* convert an ieee half/binary16 to float */
963     ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
964     ecb_function_ ecb_const float
965     ecb_binary16_to_float (uint16_t x)
966     {
967     return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
968     }
969    
970 root 1.2 #endif
971    
972 root 1.1 #endif
973