ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/cf.schmorp.de/Deliantra-Client/ecb.h
Revision: 1.3
Committed: Sat Jan 11 02:10:03 2020 UTC (4 years, 6 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.2: +22 -12 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     * 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 root 1.3 #define ECB_VERSION 0x00010007
46 root 1.1
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 root 1.3 #ifndef ECB_OPTIMIZE_SIZE
83     #if __OPTIMIZE_SIZE__
84     #define ECB_OPTIMIZE_SIZE 1
85     #else
86     #define ECB_OPTIMIZE_SIZE 0
87     #endif
88     #endif
89    
90 root 1.1 /* work around x32 idiocy by defining proper macros */
91     #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
92     #if _ILP32
93     #define ECB_AMD64_X32 1
94     #else
95     #define ECB_AMD64 1
96     #endif
97     #endif
98    
99     /* many compilers define _GNUC_ to some versions but then only implement
100     * what their idiot authors think are the "more important" extensions,
101     * causing enormous grief in return for some better fake benchmark numbers.
102     * or so.
103     * we try to detect these and simply assume they are not gcc - if they have
104     * an issue with that they should have done it right in the first place.
105     */
106     #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
107     #define ECB_GCC_VERSION(major,minor) 0
108     #else
109     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
110     #endif
111    
112     #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
113    
114     #if __clang__ && defined __has_builtin
115     #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
116     #else
117     #define ECB_CLANG_BUILTIN(x) 0
118     #endif
119    
120     #if __clang__ && defined __has_extension
121     #define ECB_CLANG_EXTENSION(x) __has_extension (x)
122     #else
123     #define ECB_CLANG_EXTENSION(x) 0
124     #endif
125    
126     #define ECB_CPP (__cplusplus+0)
127     #define ECB_CPP11 (__cplusplus >= 201103L)
128 root 1.2 #define ECB_CPP14 (__cplusplus >= 201402L)
129     #define ECB_CPP17 (__cplusplus >= 201703L)
130 root 1.1
131     #if ECB_CPP
132     #define ECB_C 0
133     #define ECB_STDC_VERSION 0
134     #else
135     #define ECB_C 1
136     #define ECB_STDC_VERSION __STDC_VERSION__
137     #endif
138    
139     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
140     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
141 root 1.2 #define ECB_C17 (ECB_STDC_VERSION >= 201710L)
142 root 1.1
143     #if ECB_CPP
144     #define ECB_EXTERN_C extern "C"
145     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
146     #define ECB_EXTERN_C_END }
147     #else
148     #define ECB_EXTERN_C extern
149     #define ECB_EXTERN_C_BEG
150     #define ECB_EXTERN_C_END
151     #endif
152    
153     /*****************************************************************************/
154    
155     /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
156     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
157    
158     #if ECB_NO_THREADS
159     #define ECB_NO_SMP 1
160     #endif
161    
162     #if ECB_NO_SMP
163     #define ECB_MEMORY_FENCE do { } while (0)
164     #endif
165    
166     /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
167     #if __xlC__ && ECB_CPP
168     #include <builtins.h>
169     #endif
170    
171     #if 1400 <= _MSC_VER
172     #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
173     #endif
174    
175     #ifndef ECB_MEMORY_FENCE
176     #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
177 root 1.3 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
178 root 1.1 #if __i386 || __i386__
179     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
180     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
181     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
182     #elif ECB_GCC_AMD64
183     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
184     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
185     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
186     #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
187     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
188     #elif defined __ARM_ARCH_2__ \
189     || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
190     || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
191     || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
192     || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
193     || defined __ARM_ARCH_5TEJ__
194     /* should not need any, unless running old code on newer cpu - arm doesn't support that */
195     #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
196     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
197     || defined __ARM_ARCH_6T2__
198     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
199     #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
200     || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
201     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
202     #elif __aarch64__
203     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
204     #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
205     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
206     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
207     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
208     #elif defined __s390__ || defined __s390x__
209     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
210     #elif defined __mips__
211     /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
212     /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
213     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
214     #elif defined __alpha__
215     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
216     #elif defined __hppa__
217     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
218     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
219     #elif defined __ia64__
220     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
221     #elif defined __m68k__
222     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
223     #elif defined __m88k__
224     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
225     #elif defined __sh__
226     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
227     #endif
228     #endif
229     #endif
230    
231     #ifndef ECB_MEMORY_FENCE
232     #if ECB_GCC_VERSION(4,7)
233     /* see comment below (stdatomic.h) about the C11 memory model. */
234     #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
235     #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
236     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
237 root 1.3 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
238 root 1.1
239     #elif ECB_CLANG_EXTENSION(c_atomic)
240     /* see comment below (stdatomic.h) about the C11 memory model. */
241     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
242     #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
243     #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
244 root 1.3 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
245 root 1.1
246     #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
247     #define ECB_MEMORY_FENCE __sync_synchronize ()
248     #elif _MSC_VER >= 1500 /* VC++ 2008 */
249     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
250     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
251     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
252     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
253     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
254     #elif _MSC_VER >= 1400 /* VC++ 2005 */
255     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
256     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
257     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
258     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
259     #elif defined _WIN32
260     #include <WinNT.h>
261     #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
262     #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
263     #include <mbarrier.h>
264 root 1.3 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
265     #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
266     #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
267     #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
268 root 1.1 #elif __xlC__
269     #define ECB_MEMORY_FENCE __sync ()
270     #endif
271     #endif
272    
273     #ifndef ECB_MEMORY_FENCE
274     #if ECB_C11 && !defined __STDC_NO_ATOMICS__
275     /* we assume that these memory fences work on all variables/all memory accesses, */
276     /* not just C11 atomics and atomic accesses */
277     #include <stdatomic.h>
278     #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
279 root 1.3 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
280     #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
281 root 1.1 #endif
282     #endif
283    
284     #ifndef ECB_MEMORY_FENCE
285     #if !ECB_AVOID_PTHREADS
286     /*
287     * if you get undefined symbol references to pthread_mutex_lock,
288     * or failure to find pthread.h, then you should implement
289     * the ECB_MEMORY_FENCE operations for your cpu/compiler
290     * OR provide pthread.h and link against the posix thread library
291     * of your system.
292     */
293     #include <pthread.h>
294     #define ECB_NEEDS_PTHREADS 1
295     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
296    
297     static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
298     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
299     #endif
300     #endif
301    
302     #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
303     #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
304     #endif
305    
306     #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
307     #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
308     #endif
309    
310 root 1.3 #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
311     #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
312     #endif
313    
314 root 1.1 /*****************************************************************************/
315    
316     #if ECB_CPP
317     #define ecb_inline static inline
318     #elif ECB_GCC_VERSION(2,5)
319     #define ecb_inline static __inline__
320     #elif ECB_C99
321     #define ecb_inline static inline
322     #else
323     #define ecb_inline static
324     #endif
325    
326     #if ECB_GCC_VERSION(3,3)
327     #define ecb_restrict __restrict__
328     #elif ECB_C99
329     #define ecb_restrict restrict
330     #else
331     #define ecb_restrict
332     #endif
333    
334     typedef int ecb_bool;
335    
336     #define ECB_CONCAT_(a, b) a ## b
337     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
338     #define ECB_STRINGIFY_(a) # a
339     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
340     #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
341    
342     #define ecb_function_ ecb_inline
343    
344     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
345     #define ecb_attribute(attrlist) __attribute__ (attrlist)
346     #else
347     #define ecb_attribute(attrlist)
348     #endif
349    
350     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
351     #define ecb_is_constant(expr) __builtin_constant_p (expr)
352     #else
353     /* possible C11 impl for integral types
354     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
355     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
356    
357     #define ecb_is_constant(expr) 0
358     #endif
359    
360     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
361     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
362     #else
363     #define ecb_expect(expr,value) (expr)
364     #endif
365    
366     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
367     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
368     #else
369     #define ecb_prefetch(addr,rw,locality)
370     #endif
371    
372     /* no emulation for ecb_decltype */
373     #if ECB_CPP11
374     // older implementations might have problems with decltype(x)::type, work around it
375     template<class T> struct ecb_decltype_t { typedef T type; };
376     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
377     #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
378     #define ecb_decltype(x) __typeof__ (x)
379     #endif
380    
381     #if _MSC_VER >= 1300
382     #define ecb_deprecated __declspec (deprecated)
383     #else
384     #define ecb_deprecated ecb_attribute ((__deprecated__))
385     #endif
386    
387     #if _MSC_VER >= 1500
388     #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
389     #elif ECB_GCC_VERSION(4,5)
390     #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
391     #else
392     #define ecb_deprecated_message(msg) ecb_deprecated
393     #endif
394    
395     #if _MSC_VER >= 1400
396     #define ecb_noinline __declspec (noinline)
397     #else
398     #define ecb_noinline ecb_attribute ((__noinline__))
399     #endif
400    
401     #define ecb_unused ecb_attribute ((__unused__))
402     #define ecb_const ecb_attribute ((__const__))
403     #define ecb_pure ecb_attribute ((__pure__))
404    
405     #if ECB_C11 || __IBMC_NORETURN
406     /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
407     #define ecb_noreturn _Noreturn
408     #elif ECB_CPP11
409     #define ecb_noreturn [[noreturn]]
410     #elif _MSC_VER >= 1200
411     /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
412     #define ecb_noreturn __declspec (noreturn)
413     #else
414     #define ecb_noreturn ecb_attribute ((__noreturn__))
415     #endif
416    
417     #if ECB_GCC_VERSION(4,3)
418     #define ecb_artificial ecb_attribute ((__artificial__))
419     #define ecb_hot ecb_attribute ((__hot__))
420     #define ecb_cold ecb_attribute ((__cold__))
421     #else
422     #define ecb_artificial
423     #define ecb_hot
424     #define ecb_cold
425     #endif
426    
427     /* put around conditional expressions if you are very sure that the */
428     /* expression is mostly true or mostly false. note that these return */
429     /* booleans, not the expression. */
430     #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
431     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
432     /* for compatibility to the rest of the world */
433     #define ecb_likely(expr) ecb_expect_true (expr)
434     #define ecb_unlikely(expr) ecb_expect_false (expr)
435    
436     /* count trailing zero bits and count # of one bits */
437     #if ECB_GCC_VERSION(3,4) \
438     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
439     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
440     && ECB_CLANG_BUILTIN(__builtin_popcount))
441     /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
442     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
443     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
444     #define ecb_ctz32(x) __builtin_ctz (x)
445     #define ecb_ctz64(x) __builtin_ctzll (x)
446     #define ecb_popcount32(x) __builtin_popcount (x)
447     /* no popcountll */
448     #else
449     ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
450     ecb_function_ ecb_const int
451     ecb_ctz32 (uint32_t x)
452     {
453     #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
454     unsigned long r;
455     _BitScanForward (&r, x);
456     return (int)r;
457     #else
458     int r = 0;
459    
460     x &= ~x + 1; /* this isolates the lowest bit */
461    
462     #if ECB_branchless_on_i386
463     r += !!(x & 0xaaaaaaaa) << 0;
464     r += !!(x & 0xcccccccc) << 1;
465     r += !!(x & 0xf0f0f0f0) << 2;
466     r += !!(x & 0xff00ff00) << 3;
467     r += !!(x & 0xffff0000) << 4;
468     #else
469     if (x & 0xaaaaaaaa) r += 1;
470     if (x & 0xcccccccc) r += 2;
471     if (x & 0xf0f0f0f0) r += 4;
472     if (x & 0xff00ff00) r += 8;
473     if (x & 0xffff0000) r += 16;
474     #endif
475    
476     return r;
477     #endif
478     }
479    
480     ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
481     ecb_function_ ecb_const int
482     ecb_ctz64 (uint64_t x)
483     {
484     #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
485     unsigned long r;
486     _BitScanForward64 (&r, x);
487     return (int)r;
488     #else
489     int shift = x & 0xffffffff ? 0 : 32;
490     return ecb_ctz32 (x >> shift) + shift;
491     #endif
492     }
493    
494     ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
495     ecb_function_ ecb_const int
496     ecb_popcount32 (uint32_t x)
497     {
498     x -= (x >> 1) & 0x55555555;
499     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
500     x = ((x >> 4) + x) & 0x0f0f0f0f;
501     x *= 0x01010101;
502    
503     return x >> 24;
504     }
505    
506     ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
507     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
508     {
509     #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
510     unsigned long r;
511     _BitScanReverse (&r, x);
512     return (int)r;
513     #else
514     int r = 0;
515    
516     if (x >> 16) { x >>= 16; r += 16; }
517     if (x >> 8) { x >>= 8; r += 8; }
518     if (x >> 4) { x >>= 4; r += 4; }
519     if (x >> 2) { x >>= 2; r += 2; }
520     if (x >> 1) { r += 1; }
521    
522     return r;
523     #endif
524     }
525    
526     ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
527     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
528     {
529     #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
530     unsigned long r;
531     _BitScanReverse64 (&r, x);
532     return (int)r;
533     #else
534     int r = 0;
535    
536     if (x >> 32) { x >>= 32; r += 32; }
537    
538     return r + ecb_ld32 (x);
539     #endif
540     }
541     #endif
542    
543     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
544     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
545     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
546     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
547    
548     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
549     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
550     {
551     return ( (x * 0x0802U & 0x22110U)
552     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
553     }
554    
555     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
556     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
557     {
558     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
559     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
560     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
561     x = ( x >> 8 ) | ( x << 8);
562    
563     return x;
564     }
565    
566     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
567     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
568     {
569     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
570     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
571     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
572     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
573     x = ( x >> 16 ) | ( x << 16);
574    
575     return x;
576     }
577    
578     /* popcount64 is only available on 64 bit cpus as gcc builtin */
579     /* so for this version we are lazy */
580     ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
581     ecb_function_ ecb_const int
582     ecb_popcount64 (uint64_t x)
583     {
584     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
585     }
586    
587     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
588     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
589     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
590     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
591     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
592     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
593     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
594     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
595    
596     ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
597     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
598     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
599     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
600     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
601     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
602     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
603     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
604    
605     #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
606     #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
607     #define ecb_bswap16(x) __builtin_bswap16 (x)
608     #else
609     #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
610     #endif
611     #define ecb_bswap32(x) __builtin_bswap32 (x)
612     #define ecb_bswap64(x) __builtin_bswap64 (x)
613     #elif _MSC_VER
614     #include <stdlib.h>
615     #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
616     #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
617     #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
618     #else
619     ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
620     ecb_function_ ecb_const uint16_t
621     ecb_bswap16 (uint16_t x)
622     {
623     return ecb_rotl16 (x, 8);
624     }
625    
626     ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
627     ecb_function_ ecb_const uint32_t
628     ecb_bswap32 (uint32_t x)
629     {
630     return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
631     }
632    
633     ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
634     ecb_function_ ecb_const uint64_t
635     ecb_bswap64 (uint64_t x)
636     {
637     return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
638     }
639     #endif
640    
641     #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
642     #define ecb_unreachable() __builtin_unreachable ()
643     #else
644     /* this seems to work fine, but gcc always emits a warning for it :/ */
645     ecb_inline ecb_noreturn void ecb_unreachable (void);
646     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
647     #endif
648    
649     /* try to tell the compiler that some condition is definitely true */
650     #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
651    
652     ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
653     ecb_inline ecb_const uint32_t
654     ecb_byteorder_helper (void)
655     {
656     /* the union code still generates code under pressure in gcc, */
657     /* but less than using pointers, and always seems to */
658     /* successfully return a constant. */
659     /* the reason why we have this horrible preprocessor mess */
660     /* is to avoid it in all cases, at least on common architectures */
661     /* or when using a recent enough gcc version (>= 4.6) */
662     #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
663     || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
664     #define ECB_LITTLE_ENDIAN 1
665     return 0x44332211;
666     #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
667     || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
668     #define ECB_BIG_ENDIAN 1
669     return 0x11223344;
670     #else
671     union
672     {
673     uint8_t c[4];
674     uint32_t u;
675     } u = { 0x11, 0x22, 0x33, 0x44 };
676     return u.u;
677     #endif
678     }
679    
680     ecb_inline ecb_const ecb_bool ecb_big_endian (void);
681     ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
682     ecb_inline ecb_const ecb_bool ecb_little_endian (void);
683     ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
684    
685     #if ECB_GCC_VERSION(3,0) || ECB_C99
686     #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
687     #else
688     #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
689     #endif
690    
691     #if ECB_CPP
692     template<typename T>
693     static inline T ecb_div_rd (T val, T div)
694     {
695     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
696     }
697     template<typename T>
698     static inline T ecb_div_ru (T val, T div)
699     {
700     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
701     }
702     #else
703     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
704     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
705     #endif
706    
707     #if ecb_cplusplus_does_not_suck
708     /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
709     template<typename T, int N>
710     static inline int ecb_array_length (const T (&arr)[N])
711     {
712     return N;
713     }
714     #else
715     #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
716     #endif
717    
718     ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
719     ecb_function_ ecb_const uint32_t
720     ecb_binary16_to_binary32 (uint32_t x)
721     {
722     unsigned int s = (x & 0x8000) << (31 - 15);
723     int e = (x >> 10) & 0x001f;
724     unsigned int m = x & 0x03ff;
725    
726     if (ecb_expect_false (e == 31))
727     /* infinity or NaN */
728     e = 255 - (127 - 15);
729     else if (ecb_expect_false (!e))
730     {
731     if (ecb_expect_true (!m))
732     /* zero, handled by code below by forcing e to 0 */
733     e = 0 - (127 - 15);
734     else
735     {
736     /* subnormal, renormalise */
737     unsigned int s = 10 - ecb_ld32 (m);
738    
739     m = (m << s) & 0x3ff; /* mask implicit bit */
740     e -= s - 1;
741     }
742     }
743    
744     /* e and m now are normalised, or zero, (or inf or nan) */
745     e += 127 - 15;
746    
747     return s | (e << 23) | (m << (23 - 10));
748     }
749    
750     ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
751     ecb_function_ ecb_const uint16_t
752     ecb_binary32_to_binary16 (uint32_t x)
753     {
754     unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
755     unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
756     unsigned int m = x & 0x007fffff;
757    
758     x &= 0x7fffffff;
759    
760     /* if it's within range of binary16 normals, use fast path */
761     if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
762     {
763     /* mantissa round-to-even */
764     m += 0x00000fff + ((m >> (23 - 10)) & 1);
765    
766     /* handle overflow */
767     if (ecb_expect_false (m >= 0x00800000))
768     {
769     m >>= 1;
770     e += 1;
771     }
772    
773     return s | (e << 10) | (m >> (23 - 10));
774     }
775    
776     /* handle large numbers and infinity */
777     if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
778     return s | 0x7c00;
779    
780     /* handle zero, subnormals and small numbers */
781     if (ecb_expect_true (x < 0x38800000))
782     {
783     /* zero */
784     if (ecb_expect_true (!x))
785     return s;
786    
787     /* handle subnormals */
788    
789     /* too small, will be zero */
790     if (e < (14 - 24)) /* might not be sharp, but is good enough */
791     return s;
792    
793     m |= 0x00800000; /* make implicit bit explicit */
794    
795     /* very tricky - we need to round to the nearest e (+10) bit value */
796     {
797     unsigned int bits = 14 - e;
798     unsigned int half = (1 << (bits - 1)) - 1;
799     unsigned int even = (m >> bits) & 1;
800    
801     /* if this overflows, we will end up with a normalised number */
802     m = (m + half + even) >> bits;
803     }
804    
805     return s | m;
806     }
807    
808     /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
809     m >>= 13;
810    
811     return s | 0x7c00 | m | !m;
812     }
813    
814     /*******************************************************************************/
815     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
816    
817     /* basically, everything uses "ieee pure-endian" floating point numbers */
818     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
819     #if 0 \
820     || __i386 || __i386__ \
821     || ECB_GCC_AMD64 \
822     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
823     || defined __s390__ || defined __s390x__ \
824     || defined __mips__ \
825     || defined __alpha__ \
826     || defined __hppa__ \
827     || defined __ia64__ \
828     || defined __m68k__ \
829     || defined __m88k__ \
830     || defined __sh__ \
831     || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
832     || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
833     || defined __aarch64__
834     #define ECB_STDFP 1
835     #include <string.h> /* for memcpy */
836     #else
837     #define ECB_STDFP 0
838     #endif
839    
840     #ifndef ECB_NO_LIBM
841    
842     #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
843    
844     /* only the oldest of old doesn't have this one. solaris. */
845     #ifdef INFINITY
846     #define ECB_INFINITY INFINITY
847     #else
848     #define ECB_INFINITY HUGE_VAL
849     #endif
850    
851     #ifdef NAN
852     #define ECB_NAN NAN
853     #else
854     #define ECB_NAN ECB_INFINITY
855     #endif
856    
857     #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
858     #define ecb_ldexpf(x,e) ldexpf ((x), (e))
859     #define ecb_frexpf(x,e) frexpf ((x), (e))
860     #else
861     #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
862     #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
863     #endif
864    
865     /* convert a float to ieee single/binary32 */
866     ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
867     ecb_function_ ecb_const uint32_t
868     ecb_float_to_binary32 (float x)
869     {
870     uint32_t r;
871    
872     #if ECB_STDFP
873     memcpy (&r, &x, 4);
874     #else
875     /* slow emulation, works for anything but -0 */
876     uint32_t m;
877     int e;
878    
879     if (x == 0e0f ) return 0x00000000U;
880     if (x > +3.40282346638528860e+38f) return 0x7f800000U;
881     if (x < -3.40282346638528860e+38f) return 0xff800000U;
882     if (x != x ) return 0x7fbfffffU;
883    
884     m = ecb_frexpf (x, &e) * 0x1000000U;
885    
886     r = m & 0x80000000U;
887    
888     if (r)
889     m = -m;
890    
891     if (e <= -126)
892     {
893     m &= 0xffffffU;
894     m >>= (-125 - e);
895     e = -126;
896     }
897    
898     r |= (e + 126) << 23;
899     r |= m & 0x7fffffU;
900     #endif
901    
902     return r;
903     }
904    
905     /* converts an ieee single/binary32 to a float */
906     ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
907     ecb_function_ ecb_const float
908     ecb_binary32_to_float (uint32_t x)
909     {
910     float r;
911    
912     #if ECB_STDFP
913     memcpy (&r, &x, 4);
914     #else
915     /* emulation, only works for normals and subnormals and +0 */
916     int neg = x >> 31;
917     int e = (x >> 23) & 0xffU;
918    
919     x &= 0x7fffffU;
920    
921     if (e)
922     x |= 0x800000U;
923     else
924     e = 1;
925    
926     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
927     r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
928    
929     r = neg ? -r : r;
930     #endif
931    
932     return r;
933     }
934    
935     /* convert a double to ieee double/binary64 */
936     ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
937     ecb_function_ ecb_const uint64_t
938     ecb_double_to_binary64 (double x)
939     {
940     uint64_t r;
941    
942     #if ECB_STDFP
943     memcpy (&r, &x, 8);
944     #else
945     /* slow emulation, works for anything but -0 */
946     uint64_t m;
947     int e;
948    
949     if (x == 0e0 ) return 0x0000000000000000U;
950     if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
951     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
952     if (x != x ) return 0X7ff7ffffffffffffU;
953    
954     m = frexp (x, &e) * 0x20000000000000U;
955    
956     r = m & 0x8000000000000000;;
957    
958     if (r)
959     m = -m;
960    
961     if (e <= -1022)
962     {
963     m &= 0x1fffffffffffffU;
964     m >>= (-1021 - e);
965     e = -1022;
966     }
967    
968     r |= ((uint64_t)(e + 1022)) << 52;
969     r |= m & 0xfffffffffffffU;
970     #endif
971    
972     return r;
973     }
974    
975     /* converts an ieee double/binary64 to a double */
976     ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
977     ecb_function_ ecb_const double
978     ecb_binary64_to_double (uint64_t x)
979     {
980     double r;
981    
982     #if ECB_STDFP
983     memcpy (&r, &x, 8);
984     #else
985     /* emulation, only works for normals and subnormals and +0 */
986     int neg = x >> 63;
987     int e = (x >> 52) & 0x7ffU;
988    
989     x &= 0xfffffffffffffU;
990    
991     if (e)
992     x |= 0x10000000000000U;
993     else
994     e = 1;
995    
996     /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
997     r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
998    
999     r = neg ? -r : r;
1000     #endif
1001    
1002     return r;
1003     }
1004    
1005     /* convert a float to ieee half/binary16 */
1006     ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1007     ecb_function_ ecb_const uint16_t
1008     ecb_float_to_binary16 (float x)
1009     {
1010     return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1011     }
1012    
1013     /* convert an ieee half/binary16 to float */
1014     ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1015     ecb_function_ ecb_const float
1016     ecb_binary16_to_float (uint16_t x)
1017     {
1018     return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1019     }
1020    
1021     #endif
1022    
1023     #endif
1024