ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/cf.schmorp.de/Deliantra-Client/ecb.h
Revision: 1.1
Committed: Mon Nov 19 00:10:34 2018 UTC (5 years, 7 months ago) by root
Content type: text/plain
Branch: MAIN
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 /*
2     * libecb - http://software.schmorp.de/pkg/libecb
3     *
4     * Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de>
5     * 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     *
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     */
40    
41     #ifndef ECB_H
42     #define ECB_H
43    
44     /* 16 bits major, 16 bits minor */
45     #define ECB_VERSION 0x00010005
46    
47     #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     #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     #else
71     #include <inttypes.h>
72     #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
73     #define ECB_PTRSIZE 8
74     #else
75     #define ECB_PTRSIZE 4
76     #endif
77     #endif
78    
79     #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
80     #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
81    
82     /* work around x32 idiocy by defining proper macros */
83     #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
84     #if _ILP32
85     #define ECB_AMD64_X32 1
86     #else
87     #define ECB_AMD64 1
88     #endif
89     #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     #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     #endif
117    
118     #define ECB_CPP (__cplusplus+0)
119     #define ECB_CPP11 (__cplusplus >= 201103L)
120    
121     #if ECB_CPP
122     #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     #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     /*****************************************************************************/
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     #define ECB_NO_SMP 1
149     #endif
150    
151     #if ECB_NO_SMP
152     #define ECB_MEMORY_FENCE do { } while (0)
153     #endif
154    
155     /* 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     #if 1400 <= _MSC_VER
161     #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
162     #endif
163    
164     #ifndef ECB_MEMORY_FENCE
165     #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
166     #if __i386 || __i386__
167     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
168     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
169     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
170     #elif ECB_GCC_AMD64
171     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
172     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
173     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
174     #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
175     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
176     #elif defined __ARM_ARCH_2__ \
177     || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
178     || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
179     || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
180     || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
181     || defined __ARM_ARCH_5TEJ__
182     /* should not need any, unless running old code on newer cpu - arm doesn't support that */
183     #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
184     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
185     || defined __ARM_ARCH_6T2__
186     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
187     #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
188     || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
189     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
190     #elif __aarch64__
191     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
192     #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
193     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
194     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
195     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
196     #elif defined __s390__ || defined __s390x__
197     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
198     #elif defined __mips__
199     /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
200     /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
201     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
202     #elif defined __alpha__
203     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
204     #elif defined __hppa__
205     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
206     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
207     #elif defined __ia64__
208     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
209     #elif defined __m68k__
210     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
211     #elif defined __m88k__
212     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
213     #elif defined __sh__
214     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
215     #endif
216     #endif
217     #endif
218    
219     #ifndef ECB_MEMORY_FENCE
220     #if ECB_GCC_VERSION(4,7)
221     /* see comment below (stdatomic.h) about the C11 memory model. */
222     #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
223     #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
224     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
225    
226     #elif ECB_CLANG_EXTENSION(c_atomic)
227     /* see comment below (stdatomic.h) about the C11 memory model. */
228     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
229     #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
230     #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
231    
232     #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
233     #define ECB_MEMORY_FENCE __sync_synchronize ()
234     #elif _MSC_VER >= 1500 /* VC++ 2008 */
235     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
236     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
237     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
238     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
239     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
240     #elif _MSC_VER >= 1400 /* VC++ 2005 */
241     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
242     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
243     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
244     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
245     #elif defined _WIN32
246     #include <WinNT.h>
247     #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
248     #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
249     #include <mbarrier.h>
250     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
251     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
252     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
253     #elif __xlC__
254     #define ECB_MEMORY_FENCE __sync ()
255     #endif
256     #endif
257    
258     #ifndef ECB_MEMORY_FENCE
259     #if ECB_C11 && !defined __STDC_NO_ATOMICS__
260     /* we assume that these memory fences work on all variables/all memory accesses, */
261     /* not just C11 atomics and atomic accesses */
262     #include <stdatomic.h>
263     /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
264     /* any fence other than seq_cst, which isn't very efficient for us. */
265     /* Why that is, we don't know - either the C11 memory model is quite useless */
266     /* for most usages, or gcc and clang have a bug */
267     /* I *currently* lean towards the latter, and inefficiently implement */
268     /* all three of ecb's fences as a seq_cst fence */
269     /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
270     /* for all __atomic_thread_fence's except seq_cst */
271     #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
272     #endif
273     #endif
274    
275     #ifndef ECB_MEMORY_FENCE
276     #if !ECB_AVOID_PTHREADS
277     /*
278     * if you get undefined symbol references to pthread_mutex_lock,
279     * or failure to find pthread.h, then you should implement
280     * the ECB_MEMORY_FENCE operations for your cpu/compiler
281     * OR provide pthread.h and link against the posix thread library
282     * of your system.
283     */
284     #include <pthread.h>
285     #define ECB_NEEDS_PTHREADS 1
286     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
287    
288     static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
289     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
290     #endif
291     #endif
292    
293     #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
294     #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
295     #endif
296    
297     #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
298     #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
299     #endif
300    
301     /*****************************************************************************/
302    
303     #if ECB_CPP
304     #define ecb_inline static inline
305     #elif ECB_GCC_VERSION(2,5)
306     #define ecb_inline static __inline__
307     #elif ECB_C99
308     #define ecb_inline static inline
309     #else
310     #define ecb_inline static
311     #endif
312    
313     #if ECB_GCC_VERSION(3,3)
314     #define ecb_restrict __restrict__
315     #elif ECB_C99
316     #define ecb_restrict restrict
317     #else
318     #define ecb_restrict
319     #endif
320    
321     typedef int ecb_bool;
322    
323     #define ECB_CONCAT_(a, b) a ## b
324     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
325     #define ECB_STRINGIFY_(a) # a
326     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
327     #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
328    
329     #define ecb_function_ ecb_inline
330    
331     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
332     #define ecb_attribute(attrlist) __attribute__ (attrlist)
333     #else
334     #define ecb_attribute(attrlist)
335     #endif
336    
337     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
338     #define ecb_is_constant(expr) __builtin_constant_p (expr)
339     #else
340     /* possible C11 impl for integral types
341     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
342     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
343    
344     #define ecb_is_constant(expr) 0
345     #endif
346    
347     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
348     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
349     #else
350     #define ecb_expect(expr,value) (expr)
351     #endif
352    
353     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
354     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
355     #else
356     #define ecb_prefetch(addr,rw,locality)
357     #endif
358    
359     /* no emulation for ecb_decltype */
360     #if ECB_CPP11
361     // older implementations might have problems with decltype(x)::type, work around it
362     template<class T> struct ecb_decltype_t { typedef T type; };
363     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
364     #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
365     #define ecb_decltype(x) __typeof__ (x)
366     #endif
367    
368     #if _MSC_VER >= 1300
369     #define ecb_deprecated __declspec (deprecated)
370     #else
371     #define ecb_deprecated ecb_attribute ((__deprecated__))
372     #endif
373    
374     #if _MSC_VER >= 1500
375     #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
376     #elif ECB_GCC_VERSION(4,5)
377     #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
378     #else
379     #define ecb_deprecated_message(msg) ecb_deprecated
380     #endif
381    
382     #if _MSC_VER >= 1400
383     #define ecb_noinline __declspec (noinline)
384     #else
385     #define ecb_noinline ecb_attribute ((__noinline__))
386     #endif
387    
388     #define ecb_unused ecb_attribute ((__unused__))
389     #define ecb_const ecb_attribute ((__const__))
390     #define ecb_pure ecb_attribute ((__pure__))
391    
392     #if ECB_C11 || __IBMC_NORETURN
393     /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
394     #define ecb_noreturn _Noreturn
395     #elif ECB_CPP11
396     #define ecb_noreturn [[noreturn]]
397     #elif _MSC_VER >= 1200
398     /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
399     #define ecb_noreturn __declspec (noreturn)
400     #else
401     #define ecb_noreturn ecb_attribute ((__noreturn__))
402     #endif
403    
404     #if ECB_GCC_VERSION(4,3)
405     #define ecb_artificial ecb_attribute ((__artificial__))
406     #define ecb_hot ecb_attribute ((__hot__))
407     #define ecb_cold ecb_attribute ((__cold__))
408     #else
409     #define ecb_artificial
410     #define ecb_hot
411     #define ecb_cold
412     #endif
413    
414     /* put around conditional expressions if you are very sure that the */
415     /* expression is mostly true or mostly false. note that these return */
416     /* booleans, not the expression. */
417     #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
418     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
419     /* for compatibility to the rest of the world */
420     #define ecb_likely(expr) ecb_expect_true (expr)
421     #define ecb_unlikely(expr) ecb_expect_false (expr)
422    
423     /* count trailing zero bits and count # of one bits */
424     #if ECB_GCC_VERSION(3,4) \
425     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
426     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
427     && ECB_CLANG_BUILTIN(__builtin_popcount))
428     /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
429     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
430     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
431     #define ecb_ctz32(x) __builtin_ctz (x)
432     #define ecb_ctz64(x) __builtin_ctzll (x)
433     #define ecb_popcount32(x) __builtin_popcount (x)
434     /* no popcountll */
435     #else
436     ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
437     ecb_function_ ecb_const int
438     ecb_ctz32 (uint32_t x)
439     {
440     #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
441     unsigned long r;
442     _BitScanForward (&r, x);
443     return (int)r;
444     #else
445     int r = 0;
446    
447     x &= ~x + 1; /* this isolates the lowest bit */
448    
449     #if ECB_branchless_on_i386
450     r += !!(x & 0xaaaaaaaa) << 0;
451     r += !!(x & 0xcccccccc) << 1;
452     r += !!(x & 0xf0f0f0f0) << 2;
453     r += !!(x & 0xff00ff00) << 3;
454     r += !!(x & 0xffff0000) << 4;
455     #else
456     if (x & 0xaaaaaaaa) r += 1;
457     if (x & 0xcccccccc) r += 2;
458     if (x & 0xf0f0f0f0) r += 4;
459     if (x & 0xff00ff00) r += 8;
460     if (x & 0xffff0000) r += 16;
461     #endif
462    
463     return r;
464     #endif
465     }
466    
467     ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
468     ecb_function_ ecb_const int
469     ecb_ctz64 (uint64_t x)
470     {
471     #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
472     unsigned long r;
473     _BitScanForward64 (&r, x);
474     return (int)r;
475     #else
476     int shift = x & 0xffffffff ? 0 : 32;
477     return ecb_ctz32 (x >> shift) + shift;
478     #endif
479     }
480    
481     ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
482     ecb_function_ ecb_const int
483     ecb_popcount32 (uint32_t x)
484     {
485     x -= (x >> 1) & 0x55555555;
486     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
487     x = ((x >> 4) + x) & 0x0f0f0f0f;
488     x *= 0x01010101;
489    
490     return x >> 24;
491     }
492    
493     ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
494     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
495     {
496     #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
497     unsigned long r;
498     _BitScanReverse (&r, x);
499     return (int)r;
500     #else
501     int r = 0;
502    
503     if (x >> 16) { x >>= 16; r += 16; }
504     if (x >> 8) { x >>= 8; r += 8; }
505     if (x >> 4) { x >>= 4; r += 4; }
506     if (x >> 2) { x >>= 2; r += 2; }
507     if (x >> 1) { r += 1; }
508    
509     return r;
510     #endif
511     }
512    
513     ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
514     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
515     {
516     #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
517     unsigned long r;
518     _BitScanReverse64 (&r, x);
519     return (int)r;
520     #else
521     int r = 0;
522    
523     if (x >> 32) { x >>= 32; r += 32; }
524    
525     return r + ecb_ld32 (x);
526     #endif
527     }
528     #endif
529    
530     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
531     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
532     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
533     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
534    
535     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
536     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
537     {
538     return ( (x * 0x0802U & 0x22110U)
539     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
540     }
541    
542     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
543     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
544     {
545     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
546     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
547     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
548     x = ( x >> 8 ) | ( x << 8);
549    
550     return x;
551     }
552    
553     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
554     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
555     {
556     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
557     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
558     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
559     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
560     x = ( x >> 16 ) | ( x << 16);
561    
562     return x;
563     }
564    
565     /* popcount64 is only available on 64 bit cpus as gcc builtin */
566     /* so for this version we are lazy */
567     ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
568     ecb_function_ ecb_const int
569     ecb_popcount64 (uint64_t x)
570     {
571     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
572     }
573    
574     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
575     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
576     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
577     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
578     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
579     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
580     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
581     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
582    
583     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
584     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
585     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
586     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
587     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
588     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
589     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
590     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
591    
592     #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
593     #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
594     #define ecb_bswap16(x) __builtin_bswap16 (x)
595     #else
596     #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
597     #endif
598     #define ecb_bswap32(x) __builtin_bswap32 (x)
599     #define ecb_bswap64(x) __builtin_bswap64 (x)
600     #elif _MSC_VER
601     #include <stdlib.h>
602     #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
603     #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
604     #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
605     #else
606     ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
607     ecb_function_ ecb_const uint16_t
608     ecb_bswap16 (uint16_t x)
609     {
610     return ecb_rotl16 (x, 8);
611     }
612    
613     ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
614     ecb_function_ ecb_const uint32_t
615     ecb_bswap32 (uint32_t x)
616     {
617     return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
618     }
619    
620     ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
621     ecb_function_ ecb_const uint64_t
622     ecb_bswap64 (uint64_t x)
623     {
624     return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
625     }
626     #endif
627    
628     #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
629     #define ecb_unreachable() __builtin_unreachable ()
630     #else
631     /* this seems to work fine, but gcc always emits a warning for it :/ */
632     ecb_inline ecb_noreturn void ecb_unreachable (void);
633     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
634     #endif
635    
636     /* try to tell the compiler that some condition is definitely true */
637     #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
638    
639     ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
640     ecb_inline ecb_const uint32_t
641     ecb_byteorder_helper (void)
642     {
643     /* the union code still generates code under pressure in gcc, */
644     /* but less than using pointers, and always seems to */
645     /* successfully return a constant. */
646     /* the reason why we have this horrible preprocessor mess */
647     /* is to avoid it in all cases, at least on common architectures */
648     /* or when using a recent enough gcc version (>= 4.6) */
649     #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
650     || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
651     #define ECB_LITTLE_ENDIAN 1
652     return 0x44332211;
653     #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
654     || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
655     #define ECB_BIG_ENDIAN 1
656     return 0x11223344;
657     #else
658     union
659     {
660     uint8_t c[4];
661     uint32_t u;
662     } u = { 0x11, 0x22, 0x33, 0x44 };
663     return u.u;
664     #endif
665     }
666    
667     ecb_inline ecb_const ecb_bool ecb_big_endian (void);
668     ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
669     ecb_inline ecb_const ecb_bool ecb_little_endian (void);
670     ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
671    
672     #if ECB_GCC_VERSION(3,0) || ECB_C99
673     #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
674     #else
675     #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
676     #endif
677    
678     #if ECB_CPP
679     template<typename T>
680     static inline T ecb_div_rd (T val, T div)
681     {
682     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
683     }
684     template<typename T>
685     static inline T ecb_div_ru (T val, T div)
686     {
687     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
688     }
689     #else
690     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
691     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
692     #endif
693    
694     #if ecb_cplusplus_does_not_suck
695     /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
696     template<typename T, int N>
697     static inline int ecb_array_length (const T (&arr)[N])
698     {
699     return N;
700     }
701     #else
702     #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
703     #endif
704    
705     ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
706     ecb_function_ ecb_const uint32_t
707     ecb_binary16_to_binary32 (uint32_t x)
708     {
709     unsigned int s = (x & 0x8000) << (31 - 15);
710     int e = (x >> 10) & 0x001f;
711     unsigned int m = x & 0x03ff;
712    
713     if (ecb_expect_false (e == 31))
714     /* infinity or NaN */
715     e = 255 - (127 - 15);
716     else if (ecb_expect_false (!e))
717     {
718     if (ecb_expect_true (!m))
719     /* zero, handled by code below by forcing e to 0 */
720     e = 0 - (127 - 15);
721     else
722     {
723     /* subnormal, renormalise */
724     unsigned int s = 10 - ecb_ld32 (m);
725    
726     m = (m << s) & 0x3ff; /* mask implicit bit */
727     e -= s - 1;
728     }
729     }
730    
731     /* e and m now are normalised, or zero, (or inf or nan) */
732     e += 127 - 15;
733    
734     return s | (e << 23) | (m << (23 - 10));
735     }
736    
737     ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
738     ecb_function_ ecb_const uint16_t
739     ecb_binary32_to_binary16 (uint32_t x)
740     {
741     unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
742     unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
743     unsigned int m = x & 0x007fffff;
744    
745     x &= 0x7fffffff;
746    
747     /* if it's within range of binary16 normals, use fast path */
748     if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
749     {
750     /* mantissa round-to-even */
751     m += 0x00000fff + ((m >> (23 - 10)) & 1);
752    
753     /* handle overflow */
754     if (ecb_expect_false (m >= 0x00800000))
755     {
756     m >>= 1;
757     e += 1;
758     }
759    
760     return s | (e << 10) | (m >> (23 - 10));
761     }
762    
763     /* handle large numbers and infinity */
764     if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
765     return s | 0x7c00;
766    
767     /* handle zero, subnormals and small numbers */
768     if (ecb_expect_true (x < 0x38800000))
769     {
770     /* zero */
771     if (ecb_expect_true (!x))
772     return s;
773    
774     /* handle subnormals */
775    
776     /* too small, will be zero */
777     if (e < (14 - 24)) /* might not be sharp, but is good enough */
778     return s;
779    
780     m |= 0x00800000; /* make implicit bit explicit */
781    
782     /* very tricky - we need to round to the nearest e (+10) bit value */
783     {
784     unsigned int bits = 14 - e;
785     unsigned int half = (1 << (bits - 1)) - 1;
786     unsigned int even = (m >> bits) & 1;
787    
788     /* if this overflows, we will end up with a normalised number */
789     m = (m + half + even) >> bits;
790     }
791    
792     return s | m;
793     }
794    
795     /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
796     m >>= 13;
797    
798     return s | 0x7c00 | m | !m;
799     }
800    
801     /*******************************************************************************/
802     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
803    
804     /* basically, everything uses "ieee pure-endian" floating point numbers */
805     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
806     #if 0 \
807     || __i386 || __i386__ \
808     || ECB_GCC_AMD64 \
809     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
810     || defined __s390__ || defined __s390x__ \
811     || defined __mips__ \
812     || defined __alpha__ \
813     || defined __hppa__ \
814     || defined __ia64__ \
815     || defined __m68k__ \
816     || defined __m88k__ \
817     || defined __sh__ \
818     || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
819     || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
820     || defined __aarch64__
821     #define ECB_STDFP 1
822     #include <string.h> /* for memcpy */
823     #else
824     #define ECB_STDFP 0
825     #endif
826    
827     #ifndef ECB_NO_LIBM
828    
829     #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
830    
831     /* only the oldest of old doesn't have this one. solaris. */
832     #ifdef INFINITY
833     #define ECB_INFINITY INFINITY
834     #else
835     #define ECB_INFINITY HUGE_VAL
836     #endif
837    
838     #ifdef NAN
839     #define ECB_NAN NAN
840     #else
841     #define ECB_NAN ECB_INFINITY
842     #endif
843    
844     #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
845     #define ecb_ldexpf(x,e) ldexpf ((x), (e))
846     #define ecb_frexpf(x,e) frexpf ((x), (e))
847     #else
848     #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
849     #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
850     #endif
851    
852     /* convert a float to ieee single/binary32 */
853     ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
854     ecb_function_ ecb_const uint32_t
855     ecb_float_to_binary32 (float x)
856     {
857     uint32_t r;
858    
859     #if ECB_STDFP
860     memcpy (&r, &x, 4);
861     #else
862     /* slow emulation, works for anything but -0 */
863     uint32_t m;
864     int e;
865    
866     if (x == 0e0f ) return 0x00000000U;
867     if (x > +3.40282346638528860e+38f) return 0x7f800000U;
868     if (x < -3.40282346638528860e+38f) return 0xff800000U;
869     if (x != x ) return 0x7fbfffffU;
870    
871     m = ecb_frexpf (x, &e) * 0x1000000U;
872    
873     r = m & 0x80000000U;
874    
875     if (r)
876     m = -m;
877    
878     if (e <= -126)
879     {
880     m &= 0xffffffU;
881     m >>= (-125 - e);
882     e = -126;
883     }
884    
885     r |= (e + 126) << 23;
886     r |= m & 0x7fffffU;
887     #endif
888    
889     return r;
890     }
891    
892     /* converts an ieee single/binary32 to a float */
893     ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
894     ecb_function_ ecb_const float
895     ecb_binary32_to_float (uint32_t x)
896     {
897     float r;
898    
899     #if ECB_STDFP
900     memcpy (&r, &x, 4);
901     #else
902     /* emulation, only works for normals and subnormals and +0 */
903     int neg = x >> 31;
904     int e = (x >> 23) & 0xffU;
905    
906     x &= 0x7fffffU;
907    
908     if (e)
909     x |= 0x800000U;
910     else
911     e = 1;
912    
913     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
914     r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
915    
916     r = neg ? -r : r;
917     #endif
918    
919     return r;
920     }
921    
922     /* convert a double to ieee double/binary64 */
923     ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
924     ecb_function_ ecb_const uint64_t
925     ecb_double_to_binary64 (double x)
926     {
927     uint64_t r;
928    
929     #if ECB_STDFP
930     memcpy (&r, &x, 8);
931     #else
932     /* slow emulation, works for anything but -0 */
933     uint64_t m;
934     int e;
935    
936     if (x == 0e0 ) return 0x0000000000000000U;
937     if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
938     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
939     if (x != x ) return 0X7ff7ffffffffffffU;
940    
941     m = frexp (x, &e) * 0x20000000000000U;
942    
943     r = m & 0x8000000000000000;;
944    
945     if (r)
946     m = -m;
947    
948     if (e <= -1022)
949     {
950     m &= 0x1fffffffffffffU;
951     m >>= (-1021 - e);
952     e = -1022;
953     }
954    
955     r |= ((uint64_t)(e + 1022)) << 52;
956     r |= m & 0xfffffffffffffU;
957     #endif
958    
959     return r;
960     }
961    
962     /* converts an ieee double/binary64 to a double */
963     ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
964     ecb_function_ ecb_const double
965     ecb_binary64_to_double (uint64_t x)
966     {
967     double r;
968    
969     #if ECB_STDFP
970     memcpy (&r, &x, 8);
971     #else
972     /* emulation, only works for normals and subnormals and +0 */
973     int neg = x >> 63;
974     int e = (x >> 52) & 0x7ffU;
975    
976     x &= 0xfffffffffffffU;
977    
978     if (e)
979     x |= 0x10000000000000U;
980     else
981     e = 1;
982    
983     /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
984     r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
985    
986     r = neg ? -r : r;
987     #endif
988    
989     return r;
990     }
991    
992     /* convert a float to ieee half/binary16 */
993     ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
994     ecb_function_ ecb_const uint16_t
995     ecb_float_to_binary16 (float x)
996     {
997     return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
998     }
999    
1000     /* convert an ieee half/binary16 to float */
1001     ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1002     ecb_function_ ecb_const float
1003     ecb_binary16_to_float (uint16_t x)
1004     {
1005     return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1006     }
1007    
1008     #endif
1009    
1010     #endif
1011