ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.211
Committed: Fri Mar 25 15:30:02 2022 UTC (2 years, 2 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.210: +2 -0 lines
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 /*
2 root 1.17 * libecb - http://software.schmorp.de/pkg/libecb
3 root 1.1 *
4 root 1.189 * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de>
5 root 1.7 * Copyright (©) 2011 Emanuele Giaquinta
6 root 1.1 * All rights reserved.
7     *
8     * Redistribution and use in source and binary forms, with or without modifica-
9     * tion, are permitted provided that the following conditions are met:
10     *
11     * 1. Redistributions of source code must retain the above copyright notice,
12     * this list of conditions and the following disclaimer.
13     *
14     * 2. Redistributions in binary form must reproduce the above copyright
15     * notice, this list of conditions and the following disclaimer in the
16     * documentation and/or other materials provided with the distribution.
17     *
18     * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
19     * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
20     * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
21     * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
22     * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
23     * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24     * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25     * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
26     * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27     * OF THE POSSIBILITY OF SUCH DAMAGE.
28 root 1.133 *
29     * Alternatively, the contents of this file may be used under the terms of
30     * the GNU General Public License ("GPL") version 2 or any later version,
31     * in which case the provisions of the GPL are applicable instead of
32     * the above. If you wish to allow the use of your version of this file
33     * only under the terms of the GPL and not to allow others to use your
34     * version of this file under the BSD license, indicate your decision
35     * by deleting the provisions above and replace them with the notice
36     * and other provisions required by the GPL. If you do not delete the
37     * provisions above, a recipient may use your version of this file under
38     * either the BSD or the GPL.
39 root 1.1 */
40    
41     #ifndef ECB_H
42     #define ECB_H
43    
44 root 1.87 /* 16 bits major, 16 bits minor */
45 root 1.204 #define ECB_VERSION 0x0001000c
46 root 1.87
47 root 1.184 #include <string.h> /* for memcpy */
48    
49 root 1.187 #if defined (_WIN32) && !defined (__MINGW32__)
50 root 1.44 typedef signed char int8_t;
51     typedef unsigned char uint8_t;
52 root 1.180 typedef signed char int_fast8_t;
53     typedef unsigned char uint_fast8_t;
54 root 1.44 typedef signed short int16_t;
55     typedef unsigned short uint16_t;
56 root 1.180 typedef signed int int_fast16_t;
57     typedef unsigned int uint_fast16_t;
58 root 1.44 typedef signed int int32_t;
59     typedef unsigned int uint32_t;
60 root 1.180 typedef signed int int_fast32_t;
61     typedef unsigned int uint_fast32_t;
62 root 1.44 #if __GNUC__
63     typedef signed long long int64_t;
64     typedef unsigned long long uint64_t;
65 root 1.51 #else /* _MSC_VER || __BORLANDC__ */
66 root 1.44 typedef signed __int64 int64_t;
67     typedef unsigned __int64 uint64_t;
68     #endif
69 root 1.180 typedef int64_t int_fast64_t;
70     typedef uint64_t uint_fast64_t;
71 root 1.87 #ifdef _WIN64
72     #define ECB_PTRSIZE 8
73     typedef uint64_t uintptr_t;
74     typedef int64_t intptr_t;
75     #else
76     #define ECB_PTRSIZE 4
77     typedef uint32_t uintptr_t;
78     typedef int32_t intptr_t;
79     #endif
80 root 1.44 #else
81     #include <inttypes.h>
82 root 1.173 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
83 root 1.87 #define ECB_PTRSIZE 8
84     #else
85     #define ECB_PTRSIZE 4
86     #endif
87 root 1.44 #endif
88 root 1.6
89 sf-exg 1.159 #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
90     #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
91    
92 root 1.179 #ifndef ECB_OPTIMIZE_SIZE
93     #if __OPTIMIZE_SIZE__
94     #define ECB_OPTIMIZE_SIZE 1
95     #else
96     #define ECB_OPTIMIZE_SIZE 0
97     #endif
98     #endif
99    
100 root 1.114 /* work around x32 idiocy by defining proper macros */
101 sf-exg 1.159 #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
102 root 1.119 #if _ILP32
103 root 1.115 #define ECB_AMD64_X32 1
104 root 1.114 #else
105 root 1.115 #define ECB_AMD64 1
106 root 1.114 #endif
107     #endif
108    
109 root 1.189 #if ECB_PTRSIZE >= 8 || ECB_AMD64_X32
110     #define ECB_64BIT_NATIVE 1
111     #else
112     #define ECB_64BIT_NATIVE 0
113     #endif
114    
115 root 1.12 /* many compilers define _GNUC_ to some versions but then only implement
116     * what their idiot authors think are the "more important" extensions,
117 sf-exg 1.59 * causing enormous grief in return for some better fake benchmark numbers.
118 root 1.18 * or so.
119 root 1.12 * we try to detect these and simply assume they are not gcc - if they have
120     * an issue with that they should have done it right in the first place.
121     */
122 root 1.137 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
123     #define ECB_GCC_VERSION(major,minor) 0
124     #else
125     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
126 root 1.12 #endif
127 root 1.1
128 sf-exg 1.138 #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
129    
130 root 1.147 #if __clang__ && defined __has_builtin
131     #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
132 sf-exg 1.138 #else
133     #define ECB_CLANG_BUILTIN(x) 0
134     #endif
135    
136 root 1.147 #if __clang__ && defined __has_extension
137     #define ECB_CLANG_EXTENSION(x) __has_extension (x)
138 sf-exg 1.140 #else
139     #define ECB_CLANG_EXTENSION(x) 0
140     #endif
141    
142 root 1.91 #define ECB_CPP (__cplusplus+0)
143     #define ECB_CPP11 (__cplusplus >= 201103L)
144 root 1.177 #define ECB_CPP14 (__cplusplus >= 201402L)
145     #define ECB_CPP17 (__cplusplus >= 201703L)
146 root 1.90
147 root 1.102 #if ECB_CPP
148 root 1.127 #define ECB_C 0
149     #define ECB_STDC_VERSION 0
150     #else
151     #define ECB_C 1
152     #define ECB_STDC_VERSION __STDC_VERSION__
153     #endif
154    
155     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
156     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
157 root 1.177 #define ECB_C17 (ECB_STDC_VERSION >= 201710L)
158 root 1.127
159     #if ECB_CPP
160 root 1.102 #define ECB_EXTERN_C extern "C"
161     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
162     #define ECB_EXTERN_C_END }
163     #else
164     #define ECB_EXTERN_C extern
165     #define ECB_EXTERN_C_BEG
166     #define ECB_EXTERN_C_END
167     #endif
168    
169 root 1.52 /*****************************************************************************/
170    
171 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
172     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
173    
174 root 1.79 #if ECB_NO_THREADS
175 root 1.95 #define ECB_NO_SMP 1
176 root 1.79 #endif
177    
178 root 1.93 #if ECB_NO_SMP
179 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
180 root 1.58 #endif
181    
182 sf-exg 1.165 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
183     #if __xlC__ && ECB_CPP
184     #include <builtins.h>
185     #endif
186    
187 root 1.171 #if 1400 <= _MSC_VER
188     #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189     #endif
190    
191 root 1.52 #ifndef ECB_MEMORY_FENCE
192 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
193 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
194 root 1.73 #if __i386 || __i386__
195 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
196 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
197 root 1.176 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
198 sf-exg 1.159 #elif ECB_GCC_AMD64
199 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
200     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
201 root 1.176 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
202 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
203 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
204 root 1.175 #elif defined __ARM_ARCH_2__ \
205     || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
206     || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
207     || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
208     || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
209     || defined __ARM_ARCH_5TEJ__
210     /* should not need any, unless running old code on newer cpu - arm doesn't support that */
211 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
212 root 1.175 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213     || defined __ARM_ARCH_6T2__
214 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
215 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
216 root 1.175 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
217 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
218 root 1.129 #elif __aarch64__
219     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
220 root 1.166 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
221 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
222     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
223     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
224 root 1.85 #elif defined __s390__ || defined __s390x__
225 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
226 root 1.85 #elif defined __mips__
227 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
228 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
229     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
230 root 1.86 #elif defined __alpha__
231 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
232     #elif defined __hppa__
233     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
234     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
235     #elif defined __ia64__
236     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
237 root 1.117 #elif defined __m68k__
238     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
239     #elif defined __m88k__
240     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
241     #elif defined __sh__
242     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
243 root 1.52 #endif
244     #endif
245     #endif
246    
247     #ifndef ECB_MEMORY_FENCE
248 root 1.93 #if ECB_GCC_VERSION(4,7)
249 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
250 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
251 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
252     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
253 root 1.190 #undef ECB_MEMORY_FENCE_RELAXED
254 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
255 root 1.110
256 sf-exg 1.140 #elif ECB_CLANG_EXTENSION(c_atomic)
257     /* see comment below (stdatomic.h) about the C11 memory model. */
258     #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
259     #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
260     #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
261 root 1.190 #undef ECB_MEMORY_FENCE_RELAXED
262 root 1.178 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
263 root 1.110
264 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
265 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
266 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
267     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
268     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
269     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
270     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
271     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
272 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
273     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
274     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
275     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
276     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
277 root 1.85 #elif defined _WIN32
278 root 1.55 #include <WinNT.h>
279 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
280 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
281     #include <mbarrier.h>
282 root 1.178 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
283     #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
284     #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
285     #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
286 root 1.82 #elif __xlC__
287 root 1.83 #define ECB_MEMORY_FENCE __sync ()
288 root 1.52 #endif
289     #endif
290    
291 root 1.53 #ifndef ECB_MEMORY_FENCE
292 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
293     /* we assume that these memory fences work on all variables/all memory accesses, */
294     /* not just C11 atomics and atomic accesses */
295     #include <stdatomic.h>
296     #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
297 root 1.178 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
298     #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
299 root 1.94 #endif
300     #endif
301    
302     #ifndef ECB_MEMORY_FENCE
303 root 1.62 #if !ECB_AVOID_PTHREADS
304     /*
305     * if you get undefined symbol references to pthread_mutex_lock,
306     * or failure to find pthread.h, then you should implement
307     * the ECB_MEMORY_FENCE operations for your cpu/compiler
308     * OR provide pthread.h and link against the posix thread library
309     * of your system.
310     */
311     #include <pthread.h>
312     #define ECB_NEEDS_PTHREADS 1
313     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
314 root 1.52
315 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
316     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
317     #endif
318     #endif
319    
320 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
321 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
322 root 1.62 #endif
323    
324 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
325 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
326     #endif
327    
328 root 1.178 #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
329     #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
330     #endif
331    
332 root 1.52 /*****************************************************************************/
333    
334 root 1.149 #if ECB_CPP
335 root 1.46 #define ecb_inline static inline
336 root 1.38 #elif ECB_GCC_VERSION(2,5)
337 root 1.46 #define ecb_inline static __inline__
338 root 1.39 #elif ECB_C99
339 root 1.46 #define ecb_inline static inline
340 root 1.29 #else
341 root 1.46 #define ecb_inline static
342 root 1.38 #endif
343    
344     #if ECB_GCC_VERSION(3,3)
345     #define ecb_restrict __restrict__
346 root 1.39 #elif ECB_C99
347 root 1.38 #define ecb_restrict restrict
348     #else
349     #define ecb_restrict
350 root 1.4 #endif
351    
352 root 1.38 typedef int ecb_bool;
353    
354 root 1.8 #define ECB_CONCAT_(a, b) a ## b
355     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
356     #define ECB_STRINGIFY_(a) # a
357     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358 root 1.155 #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
359 root 1.8
360 root 1.210 /* This marks larger functions that do not neccessarily need to be inlined */
361 root 1.211 /* The idea is to possibly compile the header twice, */
362     /* once exposing only the declarations, another time to define external functions */
363 root 1.210 /* TODO: popssibly static would be best for these at the moment? */
364 root 1.46 #define ecb_function_ ecb_inline
365 root 1.3
366 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
367 root 1.142 #define ecb_attribute(attrlist) __attribute__ (attrlist)
368 root 1.37 #else
369     #define ecb_attribute(attrlist)
370 sf-exg 1.138 #endif
371 root 1.127
372 sf-exg 1.138 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
373     #define ecb_is_constant(expr) __builtin_constant_p (expr)
374     #else
375 root 1.127 /* possible C11 impl for integral types
376     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
377     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
378    
379 root 1.37 #define ecb_is_constant(expr) 0
380 sf-exg 1.138 #endif
381    
382     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
383     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
384     #else
385 root 1.37 #define ecb_expect(expr,value) (expr)
386 sf-exg 1.138 #endif
387    
388     #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
389     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
390     #else
391 root 1.37 #define ecb_prefetch(addr,rw,locality)
392 root 1.1 #endif
393    
394 root 1.2 /* no emulation for ecb_decltype */
395 root 1.143 #if ECB_CPP11
396 root 1.144 // older implementations might have problems with decltype(x)::type, work around it
397 root 1.146 template<class T> struct ecb_decltype_t { typedef T type; };
398     #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
399 root 1.143 #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
400     #define ecb_decltype(x) __typeof__ (x)
401 root 1.1 #endif
402    
403 root 1.135 #if _MSC_VER >= 1300
404 root 1.149 #define ecb_deprecated __declspec (deprecated)
405 root 1.135 #else
406     #define ecb_deprecated ecb_attribute ((__deprecated__))
407     #endif
408    
409 sf-exg 1.162 #if _MSC_VER >= 1500
410 root 1.154 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
411     #elif ECB_GCC_VERSION(4,5)
412     #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
413     #else
414     #define ecb_deprecated_message(msg) ecb_deprecated
415     #endif
416    
417     #if _MSC_VER >= 1400
418     #define ecb_noinline __declspec (noinline)
419     #else
420     #define ecb_noinline ecb_attribute ((__noinline__))
421     #endif
422    
423 root 1.24 #define ecb_unused ecb_attribute ((__unused__))
424     #define ecb_const ecb_attribute ((__const__))
425     #define ecb_pure ecb_attribute ((__pure__))
426 root 1.35
427 root 1.145 #if ECB_C11 || __IBMC_NORETURN
428 sf-exg 1.165 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
429 root 1.90 #define ecb_noreturn _Noreturn
430 root 1.153 #elif ECB_CPP11
431     #define ecb_noreturn [[noreturn]]
432     #elif _MSC_VER >= 1200
433 sf-exg 1.156 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
434 root 1.153 #define ecb_noreturn __declspec (noreturn)
435 root 1.90 #else
436     #define ecb_noreturn ecb_attribute ((__noreturn__))
437     #endif
438    
439 root 1.35 #if ECB_GCC_VERSION(4,3)
440 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
441     #define ecb_hot ecb_attribute ((__hot__))
442     #define ecb_cold ecb_attribute ((__cold__))
443 root 1.35 #else
444     #define ecb_artificial
445     #define ecb_hot
446     #define ecb_cold
447     #endif
448 root 1.1
449 root 1.39 /* put around conditional expressions if you are very sure that the */
450     /* expression is mostly true or mostly false. note that these return */
451     /* booleans, not the expression. */
452 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
453     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
454 root 1.36 /* for compatibility to the rest of the world */
455 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
456     #define ecb_unlikely(expr) ecb_expect_false (expr)
457 root 1.1
458 root 1.3 /* count trailing zero bits and count # of one bits */
459 root 1.139 #if ECB_GCC_VERSION(3,4) \
460     || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
461     && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
462     && ECB_CLANG_BUILTIN(__builtin_popcount))
463 root 1.206 #define ecb_ctz32(x) __builtin_ctz (x)
464     #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x))
465     #define ecb_clz32(x) __builtin_clz (x)
466     #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x))
467     #define ecb_ld32(x) (ecb_clz32 (x) ^ 31)
468     #define ecb_ld64(x) (ecb_clz64 (x) ^ 63)
469 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
470 root 1.206 /* ecb_popcount64 is more difficult, see below */
471 root 1.1 #else
472 root 1.151 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
473     ecb_function_ ecb_const int
474 root 1.35 ecb_ctz32 (uint32_t x)
475     {
476 root 1.171 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
477 root 1.172 unsigned long r;
478 root 1.171 _BitScanForward (&r, x);
479     return (int)r;
480     #else
481 root 1.208 int r;
482 root 1.35
483 root 1.208 x &= ~x + 1; /* this isolates the lowest bit */
484 root 1.205
485 root 1.208 #if 1
486     /* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */
487     /* This happens to return 32 for x == 0, but the API does not support this */
488 root 1.35
489 root 1.208 /* -0 marks unused entries */
490     static unsigned char table[64] =
491     {
492     32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14,
493     10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15,
494     31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26,
495     30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0
496     };
497    
498     /* magic constant results in 33 unique values in the upper 6 bits */
499     x *= 0x0450fbafU; /* == 17 * 65 * 65535 */
500    
501     r = table [x >> 26];
502     #elif 0 /* branchless on i386, typically */
503     r = 0;
504 root 1.50 r += !!(x & 0xaaaaaaaa) << 0;
505     r += !!(x & 0xcccccccc) << 1;
506     r += !!(x & 0xf0f0f0f0) << 2;
507     r += !!(x & 0xff00ff00) << 3;
508     r += !!(x & 0xffff0000) << 4;
509 root 1.208 #else /* branchless on modern compilers, typically */
510     r = 0;
511 root 1.35 if (x & 0xaaaaaaaa) r += 1;
512     if (x & 0xcccccccc) r += 2;
513     if (x & 0xf0f0f0f0) r += 4;
514     if (x & 0xff00ff00) r += 8;
515     if (x & 0xffff0000) r += 16;
516 root 1.50 #endif
517 root 1.35
518     return r;
519 root 1.171 #endif
520 root 1.35 }
521    
522 root 1.151 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
523     ecb_function_ ecb_const int
524 root 1.49 ecb_ctz64 (uint64_t x)
525     {
526 root 1.171 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
527 root 1.172 unsigned long r;
528 root 1.171 _BitScanForward64 (&r, x);
529     return (int)r;
530     #else
531 root 1.168 int shift = x & 0xffffffff ? 0 : 32;
532 root 1.50 return ecb_ctz32 (x >> shift) + shift;
533 root 1.171 #endif
534 root 1.49 }
535    
536 root 1.208 ecb_function_ ecb_const int ecb_clz32 (uint32_t x);
537     ecb_function_ ecb_const int
538     ecb_clz32 (uint32_t x)
539     {
540     #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
541     unsigned long r;
542     _BitScanReverse (&r, x);
543     return (int)r;
544     #else
545    
546     /* Robert Harley's algorithm from comp.arch 1996-12-07 */
547     /* This happens to return 32 for x == 0, but the API does not support this */
548    
549     /* -0 marks unused table elements */
550     static unsigned char table[64] =
551     {
552     32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0,
553     -0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0,
554     17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18,
555     5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0
556     };
557    
558     /* propagate leftmost 1 bit to the right */
559     x |= x >> 1;
560     x |= x >> 2;
561     x |= x >> 4;
562     x |= x >> 8;
563     x |= x >> 16;
564    
565     /* magic constant results in 33 unique values in the upper 6 bits */
566     x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */
567    
568     return table [x >> 26];
569     #endif
570     }
571    
572     ecb_function_ ecb_const int ecb_clz64 (uint64_t x);
573     ecb_function_ ecb_const int
574     ecb_clz64 (uint64_t x)
575     {
576     #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
577     unsigned long r;
578     _BitScanReverse64 (&r, x);
579     return (int)r;
580     #else
581     uint32_t l = x >> 32;
582     int shift = l ? 0 : 32;
583     return ecb_clz32 (l ? l : x) + shift;
584     #endif
585     }
586    
587 root 1.151 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
588     ecb_function_ ecb_const int
589 root 1.35 ecb_popcount32 (uint32_t x)
590     {
591     x -= (x >> 1) & 0x55555555;
592     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
593     x = ((x >> 4) + x) & 0x0f0f0f0f;
594     x *= 0x01010101;
595 root 1.1
596 root 1.35 return x >> 24;
597     }
598 root 1.49
599 root 1.151 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
600     ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
601 root 1.49 {
602 root 1.171 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
603 root 1.172 unsigned long r;
604 root 1.171 _BitScanReverse (&r, x);
605     return (int)r;
606     #else
607 root 1.50 int r = 0;
608 root 1.49
609 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
610     if (x >> 8) { x >>= 8; r += 8; }
611     if (x >> 4) { x >>= 4; r += 4; }
612     if (x >> 2) { x >>= 2; r += 2; }
613     if (x >> 1) { r += 1; }
614 root 1.49
615     return r;
616 root 1.171 #endif
617 root 1.49 }
618    
619 root 1.151 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
620     ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
621 root 1.49 {
622 root 1.171 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
623 root 1.172 unsigned long r;
624 root 1.171 _BitScanReverse64 (&r, x);
625     return (int)r;
626     #else
627 root 1.50 int r = 0;
628 root 1.49
629 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
630 root 1.49
631 root 1.50 return r + ecb_ld32 (x);
632 root 1.171 #endif
633 root 1.49 }
634 root 1.1 #endif
635    
636 root 1.151 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
637     ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
638     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
639     ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
640 root 1.88
641 root 1.151 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
642     ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
643 root 1.70 {
644     return ( (x * 0x0802U & 0x22110U)
645 root 1.151 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
646 root 1.70 }
647    
648 root 1.151 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
649     ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
650 root 1.70 {
651     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
652     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
653     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
654     x = ( x >> 8 ) | ( x << 8);
655    
656     return x;
657     }
658    
659 root 1.151 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
660     ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
661 root 1.70 {
662     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
663     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
664     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
665     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
666     x = ( x >> 16 ) | ( x << 16);
667    
668     return x;
669     }
670    
671 root 1.151 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
672     ecb_function_ ecb_const int
673 root 1.49 ecb_popcount64 (uint64_t x)
674     {
675 root 1.205 /* popcount64 is only available on 64 bit cpus as gcc builtin. */
676     /* also, gcc/clang make this surprisingly difficult to use */
677 root 1.207 #if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
678 root 1.205 return __builtin_popcountl (x);
679     #else
680 root 1.49 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
681 root 1.205 #endif
682 root 1.49 }
683    
684 root 1.151 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
685     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
686     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
687     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
688     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
689     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
690     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
691     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
692    
693 root 1.198 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
694     ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
695     ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
696     ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
697     ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
698     ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
699     ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
700     ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
701 root 1.50
702 root 1.182 #if ECB_CPP
703    
704     inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
705     inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
706     inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
707     inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
708    
709     inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
710     inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
711     inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
712     inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
713    
714     inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
715     inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
716     inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
717     inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
718    
719     inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
720     inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
721     inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
722     inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
723    
724     inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
725     inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
726     inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
727    
728 root 1.183 inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
729     inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
730     inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
731     inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
732    
733     inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
734     inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
735     inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
736     inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
737 root 1.182
738     #endif
739    
740 sf-exg 1.138 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
741 root 1.164 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
742     #define ecb_bswap16(x) __builtin_bswap16 (x)
743     #else
744 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
745 root 1.164 #endif
746 root 1.49 #define ecb_bswap32(x) __builtin_bswap32 (x)
747     #define ecb_bswap64(x) __builtin_bswap64 (x)
748 root 1.164 #elif _MSC_VER
749     #include <stdlib.h>
750     #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
751     #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
752     #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
753 root 1.13 #else
754 root 1.151 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
755     ecb_function_ ecb_const uint16_t
756 root 1.50 ecb_bswap16 (uint16_t x)
757 root 1.49 {
758 root 1.50 return ecb_rotl16 (x, 8);
759 root 1.49 }
760    
761 root 1.151 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
762     ecb_function_ ecb_const uint32_t
763 root 1.35 ecb_bswap32 (uint32_t x)
764     {
765 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
766 root 1.35 }
767    
768 root 1.151 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
769     ecb_function_ ecb_const uint64_t
770 root 1.49 ecb_bswap64 (uint64_t x)
771 root 1.35 {
772 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
773 root 1.35 }
774 root 1.13 #endif
775    
776 sf-exg 1.138 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
777 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
778 root 1.13 #else
779 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
780 root 1.151 ecb_inline ecb_noreturn void ecb_unreachable (void);
781     ecb_inline ecb_noreturn void ecb_unreachable (void) { }
782 root 1.13 #endif
783    
784 root 1.41 /* try to tell the compiler that some condition is definitely true */
785 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
786 root 1.41
787 root 1.174 ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
788     ecb_inline ecb_const uint32_t
789 root 1.23 ecb_byteorder_helper (void)
790 root 1.3 {
791 root 1.98 /* the union code still generates code under pressure in gcc, */
792 sf-exg 1.111 /* but less than using pointers, and always seems to */
793 root 1.98 /* successfully return a constant. */
794     /* the reason why we have this horrible preprocessor mess */
795 root 1.99 /* is to avoid it in all cases, at least on common architectures */
796 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
797 root 1.174 #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
798     || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
799     #define ECB_LITTLE_ENDIAN 1
800     return 0x44332211;
801     #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
802     || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
803     #define ECB_BIG_ENDIAN 1
804     return 0x11223344;
805 root 1.98 #else
806     union
807     {
808 root 1.174 uint8_t c[4];
809     uint32_t u;
810     } u = { 0x11, 0x22, 0x33, 0x44 };
811     return u.u;
812 root 1.98 #endif
813 root 1.3 }
814    
815 root 1.151 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
816 root 1.174 ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
817 root 1.151 ecb_inline ecb_const ecb_bool ecb_little_endian (void);
818 root 1.174 ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
819 root 1.3
820 root 1.180 /*****************************************************************************/
821     /* unaligned load/store */
822    
823     ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
824     ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
825     ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
826    
827     ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
828     ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
829     ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
830    
831     ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
832     ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
833     ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
834    
835     ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
836     ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
837     ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
838    
839     ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
840     ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
841     ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
842    
843     ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
844     ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
845     ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
846    
847     ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
848     ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
849     ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
850    
851     ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
852     ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
853     ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
854    
855     ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
856     ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
857     ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
858 sf-exg 1.196
859 root 1.180 ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
860     ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
861     ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
862    
863 root 1.186 #if ECB_CPP
864 root 1.180
865     inline uint8_t ecb_bswap (uint8_t v) { return v; }
866     inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
867     inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
868     inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
869    
870     template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
871     template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
872     template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
873     template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
874     template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
875 root 1.184 template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
876 root 1.180 template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
877     template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
878    
879     template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
880     template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
881     template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
882     template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
883     template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
884 root 1.184 template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
885 root 1.180 template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
886     template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
887    
888     #endif
889    
890     /*****************************************************************************/
891 root 1.199 /* pointer/integer hashing */
892    
893     /* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
894     ecb_function_ uint32_t ecb_mix32 (uint32_t v);
895     ecb_function_ uint32_t ecb_mix32 (uint32_t v)
896     {
897     v ^= v >> 16; v *= 0x7feb352dU;
898     v ^= v >> 15; v *= 0x846ca68bU;
899     v ^= v >> 16;
900     return v;
901     }
902    
903     ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
904     ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
905     {
906     v ^= v >> 16 ; v *= 0x43021123U;
907     v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
908     v ^= v >> 16 ;
909     return v;
910     }
911    
912     /* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
913     ecb_function_ uint64_t ecb_mix64 (uint64_t v);
914     ecb_function_ uint64_t ecb_mix64 (uint64_t v)
915     {
916     v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
917     v ^= v >> 27; v *= 0x94d049bb133111ebU;
918     v ^= v >> 31;
919     return v;
920     }
921    
922     ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
923     ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
924     {
925     v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
926     v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
927     v ^= v >> 30 ^ v >> 60;
928     return v;
929     }
930    
931     ecb_function_ uintptr_t ecb_ptrmix (void *p);
932     ecb_function_ uintptr_t ecb_ptrmix (void *p)
933     {
934     #if ECB_PTRSIZE <= 4
935     return ecb_mix32 ((uint32_t)p);
936     #else
937     return ecb_mix64 ((uint64_t)p);
938     #endif
939     }
940    
941     ecb_function_ void *ecb_ptrunmix (uintptr_t v);
942     ecb_function_ void *ecb_ptrunmix (uintptr_t v)
943     {
944     #if ECB_PTRSIZE <= 4
945     return (void *)ecb_unmix32 (v);
946     #else
947     return (void *)ecb_unmix64 (v);
948     #endif
949     }
950    
951     #if ECB_CPP
952    
953     template<typename T>
954     inline uintptr_t ecb_ptrmix (T *p)
955     {
956     return ecb_ptrmix (static_cast<void *>(p));
957     }
958    
959     template<typename T>
960     inline T *ecb_ptrunmix (uintptr_t v)
961     {
962     return static_cast<T *>(ecb_ptrunmix (v));
963     }
964    
965     #endif
966    
967     /*****************************************************************************/
968 root 1.202 /* gray code */
969    
970 root 1.210 ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); }
971     ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); }
972     ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); }
973     ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); }
974 root 1.202
975 root 1.210 ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g);
976     ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g)
977 root 1.202 {
978 root 1.203 g ^= g >> 1;
979     g ^= g >> 2;
980     g ^= g >> 4;
981    
982 root 1.202 return g;
983     }
984    
985 root 1.210 ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g);
986     ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g)
987 root 1.202 {
988 root 1.203 g ^= g >> 1;
989     g ^= g >> 2;
990     g ^= g >> 4;
991     g ^= g >> 8;
992    
993 root 1.202 return g;
994     }
995    
996 root 1.210 ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g);
997     ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g)
998 root 1.202 {
999 root 1.203 g ^= g >> 1;
1000     g ^= g >> 2;
1001     g ^= g >> 4;
1002     g ^= g >> 8;
1003     g ^= g >> 16;
1004    
1005 root 1.202 return g;
1006     }
1007    
1008 root 1.210 ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g);
1009     ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g)
1010 root 1.202 {
1011 root 1.203 g ^= g >> 1;
1012     g ^= g >> 2;
1013     g ^= g >> 4;
1014     g ^= g >> 8;
1015     g ^= g >> 16;
1016     g ^= g >> 32;
1017    
1018 root 1.202 return g;
1019     }
1020    
1021     #if ECB_CPP
1022    
1023 root 1.210 ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); }
1024     ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); }
1025     ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); }
1026     ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); }
1027    
1028     ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); }
1029     ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); }
1030     ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); }
1031     ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); }
1032 root 1.202
1033     #endif
1034    
1035     /*****************************************************************************/
1036 root 1.204 /* 2d hilbert curves */
1037    
1038     /* algorithm from the book Hacker's Delight, modified to not */
1039     /* run into undefined behaviour for n==16 */
1040     static uint32_t
1041     ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
1042     {
1043     uint32_t comp, swap, cs, t, sr;
1044    
1045     /* pad s on the left (unused) bits with 01 (no change groups) */
1046     s |= 0x55555555U << n << n;
1047     /* "s shift right" */
1048     sr = (s >> 1) & 0x55555555U;
1049     /* compute complement and swap info in two-bit groups */
1050     cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
1051    
1052     /* parallel prefix xor op to propagate both complement
1053     * and swap info together from left to right (there is
1054     * no step "cs ^= cs >> 1", so in effect it computes
1055     * two independent parallel prefix operations on two
1056     * interleaved sets of sixteen bits).
1057     */
1058     cs ^= cs >> 2;
1059     cs ^= cs >> 4;
1060     cs ^= cs >> 8;
1061     cs ^= cs >> 16;
1062    
1063     /* separate swap and complement bits */
1064     swap = cs & 0x55555555U;
1065     comp = (cs >> 1) & 0x55555555U;
1066    
1067     /* calculate coordinates in odd and even bit positions */
1068     t = (s & swap) ^ comp;
1069     s = s ^ sr ^ t ^ (t << 1);
1070    
1071     /* unpad/clear out any junk on the left */
1072     s = s & ((1 << n << n) - 1);
1073    
1074     /* Now "unshuffle" to separate the x and y bits. */
1075     t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
1076     t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
1077     t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
1078     t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
1079    
1080     /* now s contains two 16-bit coordinates */
1081     return s;
1082     }
1083    
1084     /* 64 bit, a straightforward extension to the 32 bit case */
1085     static uint64_t
1086     ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1087     {
1088     uint64_t comp, swap, cs, t, sr;
1089    
1090     /* pad s on the left (unused) bits with 01 (no change groups) */
1091     s |= 0x5555555555555555U << n << n;
1092     /* "s shift right" */
1093     sr = (s >> 1) & 0x5555555555555555U;
1094     /* compute complement and swap info in two-bit groups */
1095     cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1096    
1097     /* parallel prefix xor op to propagate both complement
1098     * and swap info together from left to right (there is
1099     * no step "cs ^= cs >> 1", so in effect it computes
1100     * two independent parallel prefix operations on two
1101     * interleaved sets of thirty-two bits).
1102     */
1103     cs ^= cs >> 2;
1104     cs ^= cs >> 4;
1105     cs ^= cs >> 8;
1106     cs ^= cs >> 16;
1107     cs ^= cs >> 32;
1108    
1109     /* separate swap and complement bits */
1110     swap = cs & 0x5555555555555555U;
1111     comp = (cs >> 1) & 0x5555555555555555U;
1112    
1113     /* calculate coordinates in odd and even bit positions */
1114     t = (s & swap) ^ comp;
1115     s = s ^ sr ^ t ^ (t << 1);
1116    
1117     /* unpad/clear out any junk on the left */
1118     s = s & ((1 << n << n) - 1);
1119    
1120     /* Now "unshuffle" to separate the x and y bits. */
1121     t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1122     t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1123     t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1124     t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1125     t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1126    
1127     /* now s contains two 32-bit coordinates */
1128     return s;
1129     }
1130    
1131     /* algorithm from the book Hacker's Delight, but a similar algorithm*/
1132     /* is given in https://doi.org/10.1002/spe.4380160103 */
1133     /* this has been slightly improved over the original version */
1134     ecb_function_ uint32_t
1135     ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1136     {
1137     uint32_t row;
1138     uint32_t state = 0;
1139     uint32_t s = 0;
1140    
1141     do
1142     {
1143     --n;
1144    
1145     row = 4 * state
1146     | (2 & (xy >> n >> 15))
1147     | (1 & (xy >> n ));
1148    
1149     /* these funky constants are lookup tables for two-bit values */
1150     s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1151     state = (0x8fe65831U >> 2 * row) & 3;
1152     }
1153     while (n > 0);
1154    
1155     return s;
1156     }
1157    
1158     /* 64 bit, essentially the same as 32 bit */
1159     ecb_function_ uint64_t
1160     ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1161     {
1162     uint32_t row;
1163     uint32_t state = 0;
1164     uint64_t s = 0;
1165    
1166     do
1167     {
1168     --n;
1169    
1170     row = 4 * state
1171     | (2 & (xy >> n >> 31))
1172     | (1 & (xy >> n ));
1173    
1174     /* these funky constants are lookup tables for two-bit values */
1175     s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1176     state = (0x8fe65831U >> 2 * row) & 3;
1177     }
1178     while (n > 0);
1179    
1180     return s;
1181     }
1182    
1183     /*****************************************************************************/
1184 root 1.188 /* division */
1185 root 1.180
1186 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
1187 root 1.188 /* C99 tightened the definition of %, so we can use a more efficient version */
1188 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
1189 root 1.31 #else
1190 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
1191 root 1.31 #endif
1192 root 1.21
1193 root 1.149 #if ECB_CPP
1194 sf-exg 1.68 template<typename T>
1195     static inline T ecb_div_rd (T val, T div)
1196     {
1197     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
1198     }
1199     template<typename T>
1200     static inline T ecb_div_ru (T val, T div)
1201     {
1202     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
1203     }
1204     #else
1205     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
1206     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
1207     #endif
1208 sf-exg 1.67
1209 root 1.188 /*****************************************************************************/
1210     /* array length */
1211    
1212 root 1.5 #if ecb_cplusplus_does_not_suck
1213 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
1214 root 1.35 template<typename T, int N>
1215     static inline int ecb_array_length (const T (&arr)[N])
1216     {
1217     return N;
1218     }
1219 root 1.5 #else
1220 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
1221 root 1.5 #endif
1222    
1223 root 1.180 /*****************************************************************************/
1224 root 1.188 /* IEEE 754-2008 half float conversions */
1225 root 1.180
1226 root 1.170 ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
1227 root 1.167 ecb_function_ ecb_const uint32_t
1228 root 1.170 ecb_binary16_to_binary32 (uint32_t x)
1229 root 1.167 {
1230     unsigned int s = (x & 0x8000) << (31 - 15);
1231     int e = (x >> 10) & 0x001f;
1232     unsigned int m = x & 0x03ff;
1233    
1234     if (ecb_expect_false (e == 31))
1235     /* infinity or NaN */
1236     e = 255 - (127 - 15);
1237     else if (ecb_expect_false (!e))
1238     {
1239     if (ecb_expect_true (!m))
1240     /* zero, handled by code below by forcing e to 0 */
1241     e = 0 - (127 - 15);
1242     else
1243     {
1244     /* subnormal, renormalise */
1245     unsigned int s = 10 - ecb_ld32 (m);
1246    
1247     m = (m << s) & 0x3ff; /* mask implicit bit */
1248     e -= s - 1;
1249     }
1250     }
1251    
1252     /* e and m now are normalised, or zero, (or inf or nan) */
1253     e += 127 - 15;
1254    
1255     return s | (e << 23) | (m << (23 - 10));
1256     }
1257    
1258     ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1259     ecb_function_ ecb_const uint16_t
1260     ecb_binary32_to_binary16 (uint32_t x)
1261     {
1262     unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1263 root 1.188 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1264 root 1.167 unsigned int m = x & 0x007fffff;
1265    
1266     x &= 0x7fffffff;
1267    
1268     /* if it's within range of binary16 normals, use fast path */
1269     if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1270     {
1271     /* mantissa round-to-even */
1272     m += 0x00000fff + ((m >> (23 - 10)) & 1);
1273    
1274     /* handle overflow */
1275     if (ecb_expect_false (m >= 0x00800000))
1276     {
1277     m >>= 1;
1278     e += 1;
1279     }
1280    
1281     return s | (e << 10) | (m >> (23 - 10));
1282     }
1283    
1284     /* handle large numbers and infinity */
1285     if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1286     return s | 0x7c00;
1287    
1288 root 1.169 /* handle zero, subnormals and small numbers */
1289 root 1.167 if (ecb_expect_true (x < 0x38800000))
1290     {
1291     /* zero */
1292     if (ecb_expect_true (!x))
1293     return s;
1294    
1295     /* handle subnormals */
1296    
1297 root 1.169 /* too small, will be zero */
1298     if (e < (14 - 24)) /* might not be sharp, but is good enough */
1299     return s;
1300    
1301 root 1.167 m |= 0x00800000; /* make implicit bit explicit */
1302    
1303     /* very tricky - we need to round to the nearest e (+10) bit value */
1304     {
1305     unsigned int bits = 14 - e;
1306     unsigned int half = (1 << (bits - 1)) - 1;
1307     unsigned int even = (m >> bits) & 1;
1308    
1309     /* if this overflows, we will end up with a normalised number */
1310     m = (m + half + even) >> bits;
1311     }
1312    
1313     return s | m;
1314     }
1315    
1316     /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1317     m >>= 13;
1318    
1319     return s | 0x7c00 | m | !m;
1320     }
1321    
1322 root 1.104 /*******************************************************************************/
1323 root 1.191 /* fast integer to ascii */
1324    
1325 root 1.195 /*
1326     * This code is pretty complicated because it is general. The idea behind it,
1327     * however, is pretty simple: first, the number is multiplied with a scaling
1328 root 1.197 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1329 root 1.195 * number with the first digit in the upper bits.
1330     * Then this digit is converted to text and masked out. The resulting number
1331     * is then multiplied by 10, by multiplying the fixed point representation
1332     * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1333     * format becomes 5.27, 6.26 and so on.
1334     * The rest involves only advancing the pointer if we already generated a
1335     * non-zero digit, so leading zeroes are overwritten.
1336     */
1337    
1338 root 1.201 /* simply return a mask with "bits" bits set */
1339 root 1.191 #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1340    
1341 root 1.200 /* oputput a single digit. maskvalue is 10**digitidx */
1342 root 1.191 #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1343     if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1344     { \
1345     char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1346     *ptr = digit + '0'; /* output it */ \
1347     nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1348     ptr += nz; /* output digit only if non-zero digit seen */ \
1349     x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1350     }
1351    
1352 root 1.200 /* convert integer to fixed point format and multiply out digits, highest first */
1353     /* requires magic constants: max. digits and number of bits after the decimal point */
1354 root 1.191 #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1355     ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1356     { \
1357     char nz = lz; /* non-zero digit seen? */ \
1358     /* convert to x.bits fixed-point */ \
1359     type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1360     /* output up to 10 digits */ \
1361     ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1362     ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1363     ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1364     ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1365     ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1366     ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1367     ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1368     ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1369     ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1370     ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1371     return ptr; \
1372     }
1373    
1374 root 1.200 /* predefined versions of the above, for various digits */
1375     /* ecb_i2a_xN = almost N digits, limit defined by macro */
1376     /* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1377     /* ecb_i2a_0N = exactly N digits, including leading zeroes */
1378    
1379     /* non-leading-zero versions, limited range */
1380     #define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1381     #define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1382 root 1.191 ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1383     ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1384    
1385 root 1.200 /* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1386 root 1.194 ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1387     ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1388     ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1389     ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1390     ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1391     ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1392     ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1393     ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1394 root 1.191
1395 root 1.200 /* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1396 root 1.194 ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1397     ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1398     ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1399     ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1400     ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1401     ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1402     ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1403     ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1404 root 1.191
1405 root 1.192 #define ECB_I2A_I32_DIGITS 11
1406     #define ECB_I2A_U32_DIGITS 10
1407     #define ECB_I2A_I64_DIGITS 20
1408 root 1.194 #define ECB_I2A_U64_DIGITS 21
1409 root 1.193 #define ECB_I2A_MAX_DIGITS 21
1410 root 1.192
1411 root 1.191 ecb_inline char *
1412     ecb_i2a_u32 (char *ptr, uint32_t u)
1413     {
1414     #if ECB_64BIT_NATIVE
1415     if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1416     ptr = ecb_i2a_x10 (ptr, u);
1417 root 1.200 else /* x10 almost, but not fully, covers 32 bit */
1418 root 1.191 {
1419     uint32_t u1 = u % 1000000000;
1420     uint32_t u2 = u / 1000000000;
1421    
1422     *ptr++ = u2 + '0';
1423     ptr = ecb_i2a_09 (ptr, u1);
1424     }
1425     #else
1426     if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1427     ecb_i2a_x5 (ptr, u);
1428     else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1429     {
1430     uint32_t u1 = u % 10000;
1431     uint32_t u2 = u / 10000;
1432    
1433     ptr = ecb_i2a_x5 (ptr, u2);
1434     ptr = ecb_i2a_04 (ptr, u1);
1435     }
1436     else
1437     {
1438     uint32_t u1 = u % 10000;
1439     uint32_t ua = u / 10000;
1440     uint32_t u2 = ua % 10000;
1441     uint32_t u3 = ua / 10000;
1442    
1443     ptr = ecb_i2a_2 (ptr, u3);
1444     ptr = ecb_i2a_04 (ptr, u2);
1445     ptr = ecb_i2a_04 (ptr, u1);
1446     }
1447     #endif
1448    
1449     return ptr;
1450     }
1451    
1452     ecb_inline char *
1453     ecb_i2a_i32 (char *ptr, int32_t v)
1454     {
1455     *ptr = '-'; ptr += v < 0;
1456     uint32_t u = v < 0 ? -(uint32_t)v : v;
1457    
1458     #if ECB_64BIT_NATIVE
1459 root 1.200 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1460 root 1.191 #else
1461     ptr = ecb_i2a_u32 (ptr, u);
1462     #endif
1463    
1464     return ptr;
1465     }
1466    
1467     ecb_inline char *
1468     ecb_i2a_u64 (char *ptr, uint64_t u)
1469     {
1470     #if ECB_64BIT_NATIVE
1471     if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1472     ptr = ecb_i2a_x10 (ptr, u);
1473     else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1474     {
1475     uint64_t u1 = u % 1000000000;
1476     uint64_t u2 = u / 1000000000;
1477    
1478     ptr = ecb_i2a_x10 (ptr, u2);
1479     ptr = ecb_i2a_09 (ptr, u1);
1480     }
1481     else
1482     {
1483     uint64_t u1 = u % 1000000000;
1484     uint64_t ua = u / 1000000000;
1485     uint64_t u2 = ua % 1000000000;
1486     uint64_t u3 = ua / 1000000000;
1487    
1488     ptr = ecb_i2a_2 (ptr, u3);
1489     ptr = ecb_i2a_09 (ptr, u2);
1490     ptr = ecb_i2a_09 (ptr, u1);
1491     }
1492     #else
1493     if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1494     ptr = ecb_i2a_x5 (ptr, u);
1495     else
1496     {
1497     uint64_t u1 = u % 10000;
1498     uint64_t u2 = u / 10000;
1499    
1500     ptr = ecb_i2a_u64 (ptr, u2);
1501     ptr = ecb_i2a_04 (ptr, u1);
1502     }
1503     #endif
1504    
1505     return ptr;
1506     }
1507    
1508     ecb_inline char *
1509     ecb_i2a_i64 (char *ptr, int64_t v)
1510     {
1511     *ptr = '-'; ptr += v < 0;
1512     uint64_t u = v < 0 ? -(uint64_t)v : v;
1513    
1514     #if ECB_64BIT_NATIVE
1515     if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1516     ptr = ecb_i2a_x10 (ptr, u);
1517     else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1518     {
1519     uint64_t u1 = u % 1000000000;
1520     uint64_t u2 = u / 1000000000;
1521    
1522     ptr = ecb_i2a_x10 (ptr, u2);
1523     ptr = ecb_i2a_09 (ptr, u1);
1524     }
1525     else
1526     {
1527     uint64_t u1 = u % 1000000000;
1528     uint64_t ua = u / 1000000000;
1529     uint64_t u2 = ua % 1000000000;
1530     uint64_t u3 = ua / 1000000000;
1531    
1532 root 1.200 /* 2**31 is 19 digits, so the top is exactly one digit */
1533 root 1.191 *ptr++ = u3 + '0';
1534     ptr = ecb_i2a_09 (ptr, u2);
1535     ptr = ecb_i2a_09 (ptr, u1);
1536     }
1537     #else
1538     ptr = ecb_i2a_u64 (ptr, u);
1539     #endif
1540    
1541     return ptr;
1542     }
1543    
1544     /*******************************************************************************/
1545 root 1.104 /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
1546    
1547     /* basically, everything uses "ieee pure-endian" floating point numbers */
1548     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
1549     #if 0 \
1550     || __i386 || __i386__ \
1551 sf-exg 1.159 || ECB_GCC_AMD64 \
1552 root 1.104 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
1553     || defined __s390__ || defined __s390x__ \
1554     || defined __mips__ \
1555     || defined __alpha__ \
1556     || defined __hppa__ \
1557     || defined __ia64__ \
1558 root 1.117 || defined __m68k__ \
1559     || defined __m88k__ \
1560     || defined __sh__ \
1561 sf-exg 1.159 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1562 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1563 root 1.132 || defined __aarch64__
1564 root 1.104 #define ECB_STDFP 1
1565 root 1.102 #else
1566 root 1.104 #define ECB_STDFP 0
1567 root 1.102 #endif
1568    
1569 root 1.104 #ifndef ECB_NO_LIBM
1570 root 1.103
1571 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
1572    
1573 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
1574     #ifdef INFINITY
1575     #define ECB_INFINITY INFINITY
1576     #else
1577     #define ECB_INFINITY HUGE_VAL
1578     #endif
1579    
1580     #ifdef NAN
1581 root 1.121 #define ECB_NAN NAN
1582     #else
1583 root 1.122 #define ECB_NAN ECB_INFINITY
1584 root 1.121 #endif
1585 root 1.120
1586 root 1.148 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
1587 root 1.150 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1588 sf-exg 1.163 #define ecb_frexpf(x,e) frexpf ((x), (e))
1589 root 1.148 #else
1590 sf-exg 1.161 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1591 sf-exg 1.163 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1592 root 1.148 #endif
1593    
1594 root 1.104 /* convert a float to ieee single/binary32 */
1595 root 1.151 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1596     ecb_function_ ecb_const uint32_t
1597 root 1.103 ecb_float_to_binary32 (float x)
1598     {
1599     uint32_t r;
1600    
1601     #if ECB_STDFP
1602 root 1.104 memcpy (&r, &x, 4);
1603 root 1.103 #else
1604 root 1.105 /* slow emulation, works for anything but -0 */
1605 root 1.103 uint32_t m;
1606     int e;
1607    
1608 root 1.108 if (x == 0e0f ) return 0x00000000U;
1609 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1610     if (x < -3.40282346638528860e+38f) return 0xff800000U;
1611 root 1.105 if (x != x ) return 0x7fbfffffU;
1612 root 1.103
1613 sf-exg 1.163 m = ecb_frexpf (x, &e) * 0x1000000U;
1614 root 1.103
1615     r = m & 0x80000000U;
1616    
1617     if (r)
1618     m = -m;
1619    
1620 root 1.108 if (e <= -126)
1621 root 1.103 {
1622     m &= 0xffffffU;
1623     m >>= (-125 - e);
1624     e = -126;
1625     }
1626    
1627     r |= (e + 126) << 23;
1628     r |= m & 0x7fffffU;
1629     #endif
1630    
1631     return r;
1632     }
1633    
1634 root 1.104 /* converts an ieee single/binary32 to a float */
1635 root 1.151 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1636     ecb_function_ ecb_const float
1637 root 1.103 ecb_binary32_to_float (uint32_t x)
1638     {
1639     float r;
1640    
1641     #if ECB_STDFP
1642 root 1.104 memcpy (&r, &x, 4);
1643 root 1.103 #else
1644     /* emulation, only works for normals and subnormals and +0 */
1645     int neg = x >> 31;
1646     int e = (x >> 23) & 0xffU;
1647    
1648     x &= 0x7fffffU;
1649    
1650     if (e)
1651     x |= 0x800000U;
1652 root 1.104 else
1653     e = 1;
1654 root 1.103
1655     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1656 root 1.148 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1657 root 1.103
1658     r = neg ? -r : r;
1659     #endif
1660    
1661     return r;
1662     }
1663    
1664 root 1.104 /* convert a double to ieee double/binary64 */
1665 root 1.151 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1666     ecb_function_ ecb_const uint64_t
1667 root 1.103 ecb_double_to_binary64 (double x)
1668     {
1669 root 1.104 uint64_t r;
1670    
1671     #if ECB_STDFP
1672     memcpy (&r, &x, 8);
1673     #else
1674 root 1.105 /* slow emulation, works for anything but -0 */
1675 root 1.104 uint64_t m;
1676     int e;
1677    
1678 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
1679 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1680     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1681 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
1682 root 1.104
1683     m = frexp (x, &e) * 0x20000000000000U;
1684    
1685     r = m & 0x8000000000000000;;
1686    
1687     if (r)
1688     m = -m;
1689    
1690 root 1.108 if (e <= -1022)
1691 root 1.104 {
1692     m &= 0x1fffffffffffffU;
1693     m >>= (-1021 - e);
1694     e = -1022;
1695     }
1696    
1697     r |= ((uint64_t)(e + 1022)) << 52;
1698     r |= m & 0xfffffffffffffU;
1699     #endif
1700    
1701     return r;
1702     }
1703    
1704     /* converts an ieee double/binary64 to a double */
1705 root 1.151 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1706     ecb_function_ ecb_const double
1707 root 1.104 ecb_binary64_to_double (uint64_t x)
1708     {
1709     double r;
1710    
1711     #if ECB_STDFP
1712     memcpy (&r, &x, 8);
1713     #else
1714     /* emulation, only works for normals and subnormals and +0 */
1715     int neg = x >> 63;
1716     int e = (x >> 52) & 0x7ffU;
1717    
1718     x &= 0xfffffffffffffU;
1719    
1720     if (e)
1721     x |= 0x10000000000000U;
1722     else
1723     e = 1;
1724    
1725 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1726 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1727 root 1.104
1728     r = neg ? -r : r;
1729     #endif
1730    
1731     return r;
1732 root 1.103 }
1733    
1734 root 1.167 /* convert a float to ieee half/binary16 */
1735     ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1736     ecb_function_ ecb_const uint16_t
1737     ecb_float_to_binary16 (float x)
1738     {
1739     return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1740     }
1741    
1742     /* convert an ieee half/binary16 to float */
1743     ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1744     ecb_function_ ecb_const float
1745     ecb_binary16_to_float (uint16_t x)
1746     {
1747     return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1748     }
1749    
1750 root 1.103 #endif
1751 root 1.102
1752 root 1.1 #endif
1753