ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.208
Committed: Fri Mar 25 15:22:17 2022 UTC (2 years, 2 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.207: +76 -5 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 /*
2 * libecb - http://software.schmorp.de/pkg/libecb
3 *
4 * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de>
5 * Copyright (©) 2011 Emanuele Giaquinta
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without modifica-
9 * tion, are permitted provided that the following conditions are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright notice,
12 * this list of conditions and the following disclaimer.
13 *
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
19 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
20 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
21 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
22 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27 * OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * Alternatively, the contents of this file may be used under the terms of
30 * the GNU General Public License ("GPL") version 2 or any later version,
31 * in which case the provisions of the GPL are applicable instead of
32 * the above. If you wish to allow the use of your version of this file
33 * only under the terms of the GPL and not to allow others to use your
34 * version of this file under the BSD license, indicate your decision
35 * by deleting the provisions above and replace them with the notice
36 * and other provisions required by the GPL. If you do not delete the
37 * provisions above, a recipient may use your version of this file under
38 * either the BSD or the GPL.
39 */
40
41 #ifndef ECB_H
42 #define ECB_H
43
44 /* 16 bits major, 16 bits minor */
45 #define ECB_VERSION 0x0001000c
46
47 #include <string.h> /* for memcpy */
48
49 #if defined (_WIN32) && !defined (__MINGW32__)
50 typedef signed char int8_t;
51 typedef unsigned char uint8_t;
52 typedef signed char int_fast8_t;
53 typedef unsigned char uint_fast8_t;
54 typedef signed short int16_t;
55 typedef unsigned short uint16_t;
56 typedef signed int int_fast16_t;
57 typedef unsigned int uint_fast16_t;
58 typedef signed int int32_t;
59 typedef unsigned int uint32_t;
60 typedef signed int int_fast32_t;
61 typedef unsigned int uint_fast32_t;
62 #if __GNUC__
63 typedef signed long long int64_t;
64 typedef unsigned long long uint64_t;
65 #else /* _MSC_VER || __BORLANDC__ */
66 typedef signed __int64 int64_t;
67 typedef unsigned __int64 uint64_t;
68 #endif
69 typedef int64_t int_fast64_t;
70 typedef uint64_t uint_fast64_t;
71 #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 #else
81 #include <inttypes.h>
82 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
83 #define ECB_PTRSIZE 8
84 #else
85 #define ECB_PTRSIZE 4
86 #endif
87 #endif
88
89 #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
90 #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
91
92 #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 /* work around x32 idiocy by defining proper macros */
101 #if ECB_GCC_AMD64 || ECB_MSVC_AMD64
102 #if _ILP32
103 #define ECB_AMD64_X32 1
104 #else
105 #define ECB_AMD64 1
106 #endif
107 #endif
108
109 #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 /* many compilers define _GNUC_ to some versions but then only implement
116 * what their idiot authors think are the "more important" extensions,
117 * causing enormous grief in return for some better fake benchmark numbers.
118 * or so.
119 * 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 #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 #endif
127
128 #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
129
130 #if __clang__ && defined __has_builtin
131 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
132 #else
133 #define ECB_CLANG_BUILTIN(x) 0
134 #endif
135
136 #if __clang__ && defined __has_extension
137 #define ECB_CLANG_EXTENSION(x) __has_extension (x)
138 #else
139 #define ECB_CLANG_EXTENSION(x) 0
140 #endif
141
142 #define ECB_CPP (__cplusplus+0)
143 #define ECB_CPP11 (__cplusplus >= 201103L)
144 #define ECB_CPP14 (__cplusplus >= 201402L)
145 #define ECB_CPP17 (__cplusplus >= 201703L)
146
147 #if ECB_CPP
148 #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 #define ECB_C17 (ECB_STDC_VERSION >= 201710L)
158
159 #if ECB_CPP
160 #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 /*****************************************************************************/
170
171 /* 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 #if ECB_NO_THREADS
175 #define ECB_NO_SMP 1
176 #endif
177
178 #if ECB_NO_SMP
179 #define ECB_MEMORY_FENCE do { } while (0)
180 #endif
181
182 /* 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 #if 1400 <= _MSC_VER
188 #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189 #endif
190
191 #ifndef ECB_MEMORY_FENCE
192 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
193 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
194 #if __i386 || __i386__
195 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
196 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
197 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
198 #elif ECB_GCC_AMD64
199 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
200 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
201 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
202 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
203 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
204 #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 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
212 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213 || defined __ARM_ARCH_6T2__
214 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
215 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
216 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
218 #elif __aarch64__
219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
220 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
221 #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 #elif defined __s390__ || defined __s390x__
225 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
226 #elif defined __mips__
227 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
228 /* 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 #elif defined __alpha__
231 #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 #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 #endif
244 #endif
245 #endif
246
247 #ifndef ECB_MEMORY_FENCE
248 #if ECB_GCC_VERSION(4,7)
249 /* see comment below (stdatomic.h) about the C11 memory model. */
250 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
251 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
252 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
253 #undef ECB_MEMORY_FENCE_RELAXED
254 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
255
256 #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 #undef ECB_MEMORY_FENCE_RELAXED
262 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
263
264 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
265 #define ECB_MEMORY_FENCE __sync_synchronize ()
266 #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 #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 #elif defined _WIN32
278 #include <WinNT.h>
279 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
280 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
281 #include <mbarrier.h>
282 #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 #elif __xlC__
287 #define ECB_MEMORY_FENCE __sync ()
288 #endif
289 #endif
290
291 #ifndef ECB_MEMORY_FENCE
292 #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 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
298 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
299 #endif
300 #endif
301
302 #ifndef ECB_MEMORY_FENCE
303 #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
315 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 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
321 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
322 #endif
323
324 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
325 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
326 #endif
327
328 #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 /*****************************************************************************/
333
334 #if ECB_CPP
335 #define ecb_inline static inline
336 #elif ECB_GCC_VERSION(2,5)
337 #define ecb_inline static __inline__
338 #elif ECB_C99
339 #define ecb_inline static inline
340 #else
341 #define ecb_inline static
342 #endif
343
344 #if ECB_GCC_VERSION(3,3)
345 #define ecb_restrict __restrict__
346 #elif ECB_C99
347 #define ecb_restrict restrict
348 #else
349 #define ecb_restrict
350 #endif
351
352 typedef int ecb_bool;
353
354 #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 #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
359
360 #define ecb_function_ ecb_inline
361
362 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
363 #define ecb_attribute(attrlist) __attribute__ (attrlist)
364 #else
365 #define ecb_attribute(attrlist)
366 #endif
367
368 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
369 #define ecb_is_constant(expr) __builtin_constant_p (expr)
370 #else
371 /* possible C11 impl for integral types
372 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
373 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
374
375 #define ecb_is_constant(expr) 0
376 #endif
377
378 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
379 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
380 #else
381 #define ecb_expect(expr,value) (expr)
382 #endif
383
384 #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
385 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
386 #else
387 #define ecb_prefetch(addr,rw,locality)
388 #endif
389
390 /* no emulation for ecb_decltype */
391 #if ECB_CPP11
392 // older implementations might have problems with decltype(x)::type, work around it
393 template<class T> struct ecb_decltype_t { typedef T type; };
394 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
395 #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
396 #define ecb_decltype(x) __typeof__ (x)
397 #endif
398
399 #if _MSC_VER >= 1300
400 #define ecb_deprecated __declspec (deprecated)
401 #else
402 #define ecb_deprecated ecb_attribute ((__deprecated__))
403 #endif
404
405 #if _MSC_VER >= 1500
406 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
407 #elif ECB_GCC_VERSION(4,5)
408 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
409 #else
410 #define ecb_deprecated_message(msg) ecb_deprecated
411 #endif
412
413 #if _MSC_VER >= 1400
414 #define ecb_noinline __declspec (noinline)
415 #else
416 #define ecb_noinline ecb_attribute ((__noinline__))
417 #endif
418
419 #define ecb_unused ecb_attribute ((__unused__))
420 #define ecb_const ecb_attribute ((__const__))
421 #define ecb_pure ecb_attribute ((__pure__))
422
423 #if ECB_C11 || __IBMC_NORETURN
424 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
425 #define ecb_noreturn _Noreturn
426 #elif ECB_CPP11
427 #define ecb_noreturn [[noreturn]]
428 #elif _MSC_VER >= 1200
429 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
430 #define ecb_noreturn __declspec (noreturn)
431 #else
432 #define ecb_noreturn ecb_attribute ((__noreturn__))
433 #endif
434
435 #if ECB_GCC_VERSION(4,3)
436 #define ecb_artificial ecb_attribute ((__artificial__))
437 #define ecb_hot ecb_attribute ((__hot__))
438 #define ecb_cold ecb_attribute ((__cold__))
439 #else
440 #define ecb_artificial
441 #define ecb_hot
442 #define ecb_cold
443 #endif
444
445 /* put around conditional expressions if you are very sure that the */
446 /* expression is mostly true or mostly false. note that these return */
447 /* booleans, not the expression. */
448 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
449 #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
450 /* for compatibility to the rest of the world */
451 #define ecb_likely(expr) ecb_expect_true (expr)
452 #define ecb_unlikely(expr) ecb_expect_false (expr)
453
454 /* count trailing zero bits and count # of one bits */
455 #if ECB_GCC_VERSION(3,4) \
456 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
457 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
458 && ECB_CLANG_BUILTIN(__builtin_popcount))
459 #define ecb_ctz32(x) __builtin_ctz (x)
460 #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x))
461 #define ecb_clz32(x) __builtin_clz (x)
462 #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x))
463 #define ecb_ld32(x) (ecb_clz32 (x) ^ 31)
464 #define ecb_ld64(x) (ecb_clz64 (x) ^ 63)
465 #define ecb_popcount32(x) __builtin_popcount (x)
466 /* ecb_popcount64 is more difficult, see below */
467 #else
468 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
469 ecb_function_ ecb_const int
470 ecb_ctz32 (uint32_t x)
471 {
472 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
473 unsigned long r;
474 _BitScanForward (&r, x);
475 return (int)r;
476 #else
477 int r;
478
479 x &= ~x + 1; /* this isolates the lowest bit */
480
481 #if 1
482 /* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */
483 /* This happens to return 32 for x == 0, but the API does not support this */
484
485 /* -0 marks unused entries */
486 static unsigned char table[64] =
487 {
488 32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14,
489 10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15,
490 31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26,
491 30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0
492 };
493
494 /* magic constant results in 33 unique values in the upper 6 bits */
495 x *= 0x0450fbafU; /* == 17 * 65 * 65535 */
496
497 r = table [x >> 26];
498 #elif 0 /* branchless on i386, typically */
499 r = 0;
500 r += !!(x & 0xaaaaaaaa) << 0;
501 r += !!(x & 0xcccccccc) << 1;
502 r += !!(x & 0xf0f0f0f0) << 2;
503 r += !!(x & 0xff00ff00) << 3;
504 r += !!(x & 0xffff0000) << 4;
505 #else /* branchless on modern compilers, typically */
506 r = 0;
507 if (x & 0xaaaaaaaa) r += 1;
508 if (x & 0xcccccccc) r += 2;
509 if (x & 0xf0f0f0f0) r += 4;
510 if (x & 0xff00ff00) r += 8;
511 if (x & 0xffff0000) r += 16;
512 #endif
513
514 return r;
515 #endif
516 }
517
518 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
519 ecb_function_ ecb_const int
520 ecb_ctz64 (uint64_t x)
521 {
522 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
523 unsigned long r;
524 _BitScanForward64 (&r, x);
525 return (int)r;
526 #else
527 int shift = x & 0xffffffff ? 0 : 32;
528 return ecb_ctz32 (x >> shift) + shift;
529 #endif
530 }
531
532 ecb_function_ ecb_const int ecb_clz32 (uint32_t x);
533 ecb_function_ ecb_const int
534 ecb_clz32 (uint32_t x)
535 {
536 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
537 unsigned long r;
538 _BitScanReverse (&r, x);
539 return (int)r;
540 #else
541
542 /* Robert Harley's algorithm from comp.arch 1996-12-07 */
543 /* This happens to return 32 for x == 0, but the API does not support this */
544
545 /* -0 marks unused table elements */
546 static unsigned char table[64] =
547 {
548 32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0,
549 -0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0,
550 17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18,
551 5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0
552 };
553
554 /* propagate leftmost 1 bit to the right */
555 x |= x >> 1;
556 x |= x >> 2;
557 x |= x >> 4;
558 x |= x >> 8;
559 x |= x >> 16;
560
561 /* magic constant results in 33 unique values in the upper 6 bits */
562 x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */
563
564 return table [x >> 26];
565 #endif
566 }
567
568 ecb_function_ ecb_const int ecb_clz64 (uint64_t x);
569 ecb_function_ ecb_const int
570 ecb_clz64 (uint64_t x)
571 {
572 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
573 unsigned long r;
574 _BitScanReverse64 (&r, x);
575 return (int)r;
576 #else
577 uint32_t l = x >> 32;
578 int shift = l ? 0 : 32;
579 return ecb_clz32 (l ? l : x) + shift;
580 #endif
581 }
582
583 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
584 ecb_function_ ecb_const int
585 ecb_popcount32 (uint32_t x)
586 {
587 x -= (x >> 1) & 0x55555555;
588 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
589 x = ((x >> 4) + x) & 0x0f0f0f0f;
590 x *= 0x01010101;
591
592 return x >> 24;
593 }
594
595 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
596 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
597 {
598 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
599 unsigned long r;
600 _BitScanReverse (&r, x);
601 return (int)r;
602 #else
603 int r = 0;
604
605 if (x >> 16) { x >>= 16; r += 16; }
606 if (x >> 8) { x >>= 8; r += 8; }
607 if (x >> 4) { x >>= 4; r += 4; }
608 if (x >> 2) { x >>= 2; r += 2; }
609 if (x >> 1) { r += 1; }
610
611 return r;
612 #endif
613 }
614
615 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
616 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
617 {
618 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
619 unsigned long r;
620 _BitScanReverse64 (&r, x);
621 return (int)r;
622 #else
623 int r = 0;
624
625 if (x >> 32) { x >>= 32; r += 32; }
626
627 return r + ecb_ld32 (x);
628 #endif
629 }
630 #endif
631
632 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
633 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
634 ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
635 ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
636
637 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
638 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
639 {
640 return ( (x * 0x0802U & 0x22110U)
641 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
642 }
643
644 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
645 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
646 {
647 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
648 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
649 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
650 x = ( x >> 8 ) | ( x << 8);
651
652 return x;
653 }
654
655 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
656 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
657 {
658 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
659 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
660 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
661 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
662 x = ( x >> 16 ) | ( x << 16);
663
664 return x;
665 }
666
667 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
668 ecb_function_ ecb_const int
669 ecb_popcount64 (uint64_t x)
670 {
671 /* popcount64 is only available on 64 bit cpus as gcc builtin. */
672 /* also, gcc/clang make this surprisingly difficult to use */
673 #if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
674 return __builtin_popcountl (x);
675 #else
676 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
677 #endif
678 }
679
680 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
681 ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
682 ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
683 ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
684 ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
685 ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
686 ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
687 ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
688
689 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
690 ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
691 ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
692 ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
693 ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
694 ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
695 ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
696 ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
697
698 #if ECB_CPP
699
700 inline uint32_t ecb_clz (uint32_t v) { return ecb_clz32 (v); }
701 inline uint64_t ecb_clz (uint64_t v) { return ecb_clz64 (v); }
702
703 inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
704 inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
705 inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
706 inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
707
708 inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
709 inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
710 inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
711 inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
712
713 inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
714 inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
715 inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
716 inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
717
718 inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
719 inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
720 inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
721 inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
722
723 inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
724 inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
725 inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
726
727 inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
728 inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
729 inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
730 inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
731
732 inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
733 inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
734 inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
735 inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
736
737 #endif
738
739 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
740 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
741 #define ecb_bswap16(x) __builtin_bswap16 (x)
742 #else
743 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
744 #endif
745 #define ecb_bswap32(x) __builtin_bswap32 (x)
746 #define ecb_bswap64(x) __builtin_bswap64 (x)
747 #elif _MSC_VER
748 #include <stdlib.h>
749 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
750 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
751 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
752 #else
753 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
754 ecb_function_ ecb_const uint16_t
755 ecb_bswap16 (uint16_t x)
756 {
757 return ecb_rotl16 (x, 8);
758 }
759
760 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
761 ecb_function_ ecb_const uint32_t
762 ecb_bswap32 (uint32_t x)
763 {
764 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
765 }
766
767 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
768 ecb_function_ ecb_const uint64_t
769 ecb_bswap64 (uint64_t x)
770 {
771 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
772 }
773 #endif
774
775 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
776 #define ecb_unreachable() __builtin_unreachable ()
777 #else
778 /* this seems to work fine, but gcc always emits a warning for it :/ */
779 ecb_inline ecb_noreturn void ecb_unreachable (void);
780 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
781 #endif
782
783 /* try to tell the compiler that some condition is definitely true */
784 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
785
786 ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
787 ecb_inline ecb_const uint32_t
788 ecb_byteorder_helper (void)
789 {
790 /* the union code still generates code under pressure in gcc, */
791 /* but less than using pointers, and always seems to */
792 /* successfully return a constant. */
793 /* the reason why we have this horrible preprocessor mess */
794 /* is to avoid it in all cases, at least on common architectures */
795 /* or when using a recent enough gcc version (>= 4.6) */
796 #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
797 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
798 #define ECB_LITTLE_ENDIAN 1
799 return 0x44332211;
800 #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
801 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
802 #define ECB_BIG_ENDIAN 1
803 return 0x11223344;
804 #else
805 union
806 {
807 uint8_t c[4];
808 uint32_t u;
809 } u = { 0x11, 0x22, 0x33, 0x44 };
810 return u.u;
811 #endif
812 }
813
814 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
815 ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
816 ecb_inline ecb_const ecb_bool ecb_little_endian (void);
817 ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
818
819 /*****************************************************************************/
820 /* unaligned load/store */
821
822 ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
823 ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
824 ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
825
826 ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
827 ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
828 ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
829
830 ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
831 ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
832 ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
833
834 ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
835 ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
836 ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
837
838 ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
839 ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
840 ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
841
842 ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
843 ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
844 ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
845
846 ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
847 ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
848 ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
849
850 ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
851 ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
852 ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
853
854 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)); }
855 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)); }
856 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)); }
857
858 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)); }
859 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)); }
860 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)); }
861
862 #if ECB_CPP
863
864 inline uint8_t ecb_bswap (uint8_t v) { return v; }
865 inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
866 inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
867 inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
868
869 template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
870 template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
871 template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
872 template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
873 template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
874 template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
875 template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
876 template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
877
878 template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
879 template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
880 template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
881 template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
882 template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
883 template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
884 template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
885 template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
886
887 #endif
888
889 /*****************************************************************************/
890 /* pointer/integer hashing */
891
892 /* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
893 ecb_function_ uint32_t ecb_mix32 (uint32_t v);
894 ecb_function_ uint32_t ecb_mix32 (uint32_t v)
895 {
896 v ^= v >> 16; v *= 0x7feb352dU;
897 v ^= v >> 15; v *= 0x846ca68bU;
898 v ^= v >> 16;
899 return v;
900 }
901
902 ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
903 ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
904 {
905 v ^= v >> 16 ; v *= 0x43021123U;
906 v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
907 v ^= v >> 16 ;
908 return v;
909 }
910
911 /* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
912 ecb_function_ uint64_t ecb_mix64 (uint64_t v);
913 ecb_function_ uint64_t ecb_mix64 (uint64_t v)
914 {
915 v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
916 v ^= v >> 27; v *= 0x94d049bb133111ebU;
917 v ^= v >> 31;
918 return v;
919 }
920
921 ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
922 ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
923 {
924 v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
925 v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
926 v ^= v >> 30 ^ v >> 60;
927 return v;
928 }
929
930 ecb_function_ uintptr_t ecb_ptrmix (void *p);
931 ecb_function_ uintptr_t ecb_ptrmix (void *p)
932 {
933 #if ECB_PTRSIZE <= 4
934 return ecb_mix32 ((uint32_t)p);
935 #else
936 return ecb_mix64 ((uint64_t)p);
937 #endif
938 }
939
940 ecb_function_ void *ecb_ptrunmix (uintptr_t v);
941 ecb_function_ void *ecb_ptrunmix (uintptr_t v)
942 {
943 #if ECB_PTRSIZE <= 4
944 return (void *)ecb_unmix32 (v);
945 #else
946 return (void *)ecb_unmix64 (v);
947 #endif
948 }
949
950 #if ECB_CPP
951
952 template<typename T>
953 inline uintptr_t ecb_ptrmix (T *p)
954 {
955 return ecb_ptrmix (static_cast<void *>(p));
956 }
957
958 template<typename T>
959 inline T *ecb_ptrunmix (uintptr_t v)
960 {
961 return static_cast<T *>(ecb_ptrunmix (v));
962 }
963
964 #endif
965
966 /*****************************************************************************/
967 /* gray code */
968
969 ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
970 ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
971 ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
972 ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
973
974 ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
975 {
976 g ^= g >> 1;
977 g ^= g >> 2;
978 g ^= g >> 4;
979
980 return g;
981 }
982
983 ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
984 {
985 g ^= g >> 1;
986 g ^= g >> 2;
987 g ^= g >> 4;
988 g ^= g >> 8;
989
990 return g;
991 }
992
993 ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
994 {
995 g ^= g >> 1;
996 g ^= g >> 2;
997 g ^= g >> 4;
998 g ^= g >> 8;
999 g ^= g >> 16;
1000
1001 return g;
1002 }
1003
1004 ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
1005 {
1006 g ^= g >> 1;
1007 g ^= g >> 2;
1008 g ^= g >> 4;
1009 g ^= g >> 8;
1010 g ^= g >> 16;
1011 g ^= g >> 32;
1012
1013 return g;
1014 }
1015
1016 #if ECB_CPP
1017
1018 ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
1019 ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
1020 ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
1021 ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
1022
1023 ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
1024 ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
1025 ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
1026 ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
1027
1028 #endif
1029
1030 /*****************************************************************************/
1031 /* 2d hilbert curves */
1032
1033 /* algorithm from the book Hacker's Delight, modified to not */
1034 /* run into undefined behaviour for n==16 */
1035 static uint32_t
1036 ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
1037 {
1038 uint32_t comp, swap, cs, t, sr;
1039
1040 /* pad s on the left (unused) bits with 01 (no change groups) */
1041 s |= 0x55555555U << n << n;
1042 /* "s shift right" */
1043 sr = (s >> 1) & 0x55555555U;
1044 /* compute complement and swap info in two-bit groups */
1045 cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
1046
1047 /* parallel prefix xor op to propagate both complement
1048 * and swap info together from left to right (there is
1049 * no step "cs ^= cs >> 1", so in effect it computes
1050 * two independent parallel prefix operations on two
1051 * interleaved sets of sixteen bits).
1052 */
1053 cs ^= cs >> 2;
1054 cs ^= cs >> 4;
1055 cs ^= cs >> 8;
1056 cs ^= cs >> 16;
1057
1058 /* separate swap and complement bits */
1059 swap = cs & 0x55555555U;
1060 comp = (cs >> 1) & 0x55555555U;
1061
1062 /* calculate coordinates in odd and even bit positions */
1063 t = (s & swap) ^ comp;
1064 s = s ^ sr ^ t ^ (t << 1);
1065
1066 /* unpad/clear out any junk on the left */
1067 s = s & ((1 << n << n) - 1);
1068
1069 /* Now "unshuffle" to separate the x and y bits. */
1070 t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
1071 t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
1072 t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
1073 t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
1074
1075 /* now s contains two 16-bit coordinates */
1076 return s;
1077 }
1078
1079 /* 64 bit, a straightforward extension to the 32 bit case */
1080 static uint64_t
1081 ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1082 {
1083 uint64_t comp, swap, cs, t, sr;
1084
1085 /* pad s on the left (unused) bits with 01 (no change groups) */
1086 s |= 0x5555555555555555U << n << n;
1087 /* "s shift right" */
1088 sr = (s >> 1) & 0x5555555555555555U;
1089 /* compute complement and swap info in two-bit groups */
1090 cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1091
1092 /* parallel prefix xor op to propagate both complement
1093 * and swap info together from left to right (there is
1094 * no step "cs ^= cs >> 1", so in effect it computes
1095 * two independent parallel prefix operations on two
1096 * interleaved sets of thirty-two bits).
1097 */
1098 cs ^= cs >> 2;
1099 cs ^= cs >> 4;
1100 cs ^= cs >> 8;
1101 cs ^= cs >> 16;
1102 cs ^= cs >> 32;
1103
1104 /* separate swap and complement bits */
1105 swap = cs & 0x5555555555555555U;
1106 comp = (cs >> 1) & 0x5555555555555555U;
1107
1108 /* calculate coordinates in odd and even bit positions */
1109 t = (s & swap) ^ comp;
1110 s = s ^ sr ^ t ^ (t << 1);
1111
1112 /* unpad/clear out any junk on the left */
1113 s = s & ((1 << n << n) - 1);
1114
1115 /* Now "unshuffle" to separate the x and y bits. */
1116 t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1117 t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1118 t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1119 t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1120 t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1121
1122 /* now s contains two 32-bit coordinates */
1123 return s;
1124 }
1125
1126 /* algorithm from the book Hacker's Delight, but a similar algorithm*/
1127 /* is given in https://doi.org/10.1002/spe.4380160103 */
1128 /* this has been slightly improved over the original version */
1129 ecb_function_ uint32_t
1130 ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1131 {
1132 uint32_t row;
1133 uint32_t state = 0;
1134 uint32_t s = 0;
1135
1136 do
1137 {
1138 --n;
1139
1140 row = 4 * state
1141 | (2 & (xy >> n >> 15))
1142 | (1 & (xy >> n ));
1143
1144 /* these funky constants are lookup tables for two-bit values */
1145 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1146 state = (0x8fe65831U >> 2 * row) & 3;
1147 }
1148 while (n > 0);
1149
1150 return s;
1151 }
1152
1153 /* 64 bit, essentially the same as 32 bit */
1154 ecb_function_ uint64_t
1155 ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1156 {
1157 uint32_t row;
1158 uint32_t state = 0;
1159 uint64_t s = 0;
1160
1161 do
1162 {
1163 --n;
1164
1165 row = 4 * state
1166 | (2 & (xy >> n >> 31))
1167 | (1 & (xy >> n ));
1168
1169 /* these funky constants are lookup tables for two-bit values */
1170 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1171 state = (0x8fe65831U >> 2 * row) & 3;
1172 }
1173 while (n > 0);
1174
1175 return s;
1176 }
1177
1178 /*****************************************************************************/
1179 /* division */
1180
1181 #if ECB_GCC_VERSION(3,0) || ECB_C99
1182 /* C99 tightened the definition of %, so we can use a more efficient version */
1183 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
1184 #else
1185 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
1186 #endif
1187
1188 #if ECB_CPP
1189 template<typename T>
1190 static inline T ecb_div_rd (T val, T div)
1191 {
1192 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
1193 }
1194 template<typename T>
1195 static inline T ecb_div_ru (T val, T div)
1196 {
1197 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
1198 }
1199 #else
1200 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
1201 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
1202 #endif
1203
1204 /*****************************************************************************/
1205 /* array length */
1206
1207 #if ecb_cplusplus_does_not_suck
1208 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
1209 template<typename T, int N>
1210 static inline int ecb_array_length (const T (&arr)[N])
1211 {
1212 return N;
1213 }
1214 #else
1215 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
1216 #endif
1217
1218 /*****************************************************************************/
1219 /* IEEE 754-2008 half float conversions */
1220
1221 ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
1222 ecb_function_ ecb_const uint32_t
1223 ecb_binary16_to_binary32 (uint32_t x)
1224 {
1225 unsigned int s = (x & 0x8000) << (31 - 15);
1226 int e = (x >> 10) & 0x001f;
1227 unsigned int m = x & 0x03ff;
1228
1229 if (ecb_expect_false (e == 31))
1230 /* infinity or NaN */
1231 e = 255 - (127 - 15);
1232 else if (ecb_expect_false (!e))
1233 {
1234 if (ecb_expect_true (!m))
1235 /* zero, handled by code below by forcing e to 0 */
1236 e = 0 - (127 - 15);
1237 else
1238 {
1239 /* subnormal, renormalise */
1240 unsigned int s = 10 - ecb_ld32 (m);
1241
1242 m = (m << s) & 0x3ff; /* mask implicit bit */
1243 e -= s - 1;
1244 }
1245 }
1246
1247 /* e and m now are normalised, or zero, (or inf or nan) */
1248 e += 127 - 15;
1249
1250 return s | (e << 23) | (m << (23 - 10));
1251 }
1252
1253 ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1254 ecb_function_ ecb_const uint16_t
1255 ecb_binary32_to_binary16 (uint32_t x)
1256 {
1257 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1258 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1259 unsigned int m = x & 0x007fffff;
1260
1261 x &= 0x7fffffff;
1262
1263 /* if it's within range of binary16 normals, use fast path */
1264 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1265 {
1266 /* mantissa round-to-even */
1267 m += 0x00000fff + ((m >> (23 - 10)) & 1);
1268
1269 /* handle overflow */
1270 if (ecb_expect_false (m >= 0x00800000))
1271 {
1272 m >>= 1;
1273 e += 1;
1274 }
1275
1276 return s | (e << 10) | (m >> (23 - 10));
1277 }
1278
1279 /* handle large numbers and infinity */
1280 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1281 return s | 0x7c00;
1282
1283 /* handle zero, subnormals and small numbers */
1284 if (ecb_expect_true (x < 0x38800000))
1285 {
1286 /* zero */
1287 if (ecb_expect_true (!x))
1288 return s;
1289
1290 /* handle subnormals */
1291
1292 /* too small, will be zero */
1293 if (e < (14 - 24)) /* might not be sharp, but is good enough */
1294 return s;
1295
1296 m |= 0x00800000; /* make implicit bit explicit */
1297
1298 /* very tricky - we need to round to the nearest e (+10) bit value */
1299 {
1300 unsigned int bits = 14 - e;
1301 unsigned int half = (1 << (bits - 1)) - 1;
1302 unsigned int even = (m >> bits) & 1;
1303
1304 /* if this overflows, we will end up with a normalised number */
1305 m = (m + half + even) >> bits;
1306 }
1307
1308 return s | m;
1309 }
1310
1311 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1312 m >>= 13;
1313
1314 return s | 0x7c00 | m | !m;
1315 }
1316
1317 /*******************************************************************************/
1318 /* fast integer to ascii */
1319
1320 /*
1321 * This code is pretty complicated because it is general. The idea behind it,
1322 * however, is pretty simple: first, the number is multiplied with a scaling
1323 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1324 * number with the first digit in the upper bits.
1325 * Then this digit is converted to text and masked out. The resulting number
1326 * is then multiplied by 10, by multiplying the fixed point representation
1327 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1328 * format becomes 5.27, 6.26 and so on.
1329 * The rest involves only advancing the pointer if we already generated a
1330 * non-zero digit, so leading zeroes are overwritten.
1331 */
1332
1333 /* simply return a mask with "bits" bits set */
1334 #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1335
1336 /* oputput a single digit. maskvalue is 10**digitidx */
1337 #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1338 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1339 { \
1340 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1341 *ptr = digit + '0'; /* output it */ \
1342 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1343 ptr += nz; /* output digit only if non-zero digit seen */ \
1344 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1345 }
1346
1347 /* convert integer to fixed point format and multiply out digits, highest first */
1348 /* requires magic constants: max. digits and number of bits after the decimal point */
1349 #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1350 ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1351 { \
1352 char nz = lz; /* non-zero digit seen? */ \
1353 /* convert to x.bits fixed-point */ \
1354 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1355 /* output up to 10 digits */ \
1356 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1357 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1358 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1359 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1360 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1361 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1362 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1363 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1364 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1365 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1366 return ptr; \
1367 }
1368
1369 /* predefined versions of the above, for various digits */
1370 /* ecb_i2a_xN = almost N digits, limit defined by macro */
1371 /* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1372 /* ecb_i2a_0N = exactly N digits, including leading zeroes */
1373
1374 /* non-leading-zero versions, limited range */
1375 #define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1376 #define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1377 ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1378 ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1379
1380 /* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1381 ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1382 ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1383 ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1384 ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1385 ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1386 ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1387 ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1388 ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1389
1390 /* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1391 ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1392 ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1393 ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1394 ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1395 ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1396 ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1397 ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1398 ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1399
1400 #define ECB_I2A_I32_DIGITS 11
1401 #define ECB_I2A_U32_DIGITS 10
1402 #define ECB_I2A_I64_DIGITS 20
1403 #define ECB_I2A_U64_DIGITS 21
1404 #define ECB_I2A_MAX_DIGITS 21
1405
1406 ecb_inline char *
1407 ecb_i2a_u32 (char *ptr, uint32_t u)
1408 {
1409 #if ECB_64BIT_NATIVE
1410 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1411 ptr = ecb_i2a_x10 (ptr, u);
1412 else /* x10 almost, but not fully, covers 32 bit */
1413 {
1414 uint32_t u1 = u % 1000000000;
1415 uint32_t u2 = u / 1000000000;
1416
1417 *ptr++ = u2 + '0';
1418 ptr = ecb_i2a_09 (ptr, u1);
1419 }
1420 #else
1421 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1422 ecb_i2a_x5 (ptr, u);
1423 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1424 {
1425 uint32_t u1 = u % 10000;
1426 uint32_t u2 = u / 10000;
1427
1428 ptr = ecb_i2a_x5 (ptr, u2);
1429 ptr = ecb_i2a_04 (ptr, u1);
1430 }
1431 else
1432 {
1433 uint32_t u1 = u % 10000;
1434 uint32_t ua = u / 10000;
1435 uint32_t u2 = ua % 10000;
1436 uint32_t u3 = ua / 10000;
1437
1438 ptr = ecb_i2a_2 (ptr, u3);
1439 ptr = ecb_i2a_04 (ptr, u2);
1440 ptr = ecb_i2a_04 (ptr, u1);
1441 }
1442 #endif
1443
1444 return ptr;
1445 }
1446
1447 ecb_inline char *
1448 ecb_i2a_i32 (char *ptr, int32_t v)
1449 {
1450 *ptr = '-'; ptr += v < 0;
1451 uint32_t u = v < 0 ? -(uint32_t)v : v;
1452
1453 #if ECB_64BIT_NATIVE
1454 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1455 #else
1456 ptr = ecb_i2a_u32 (ptr, u);
1457 #endif
1458
1459 return ptr;
1460 }
1461
1462 ecb_inline char *
1463 ecb_i2a_u64 (char *ptr, uint64_t u)
1464 {
1465 #if ECB_64BIT_NATIVE
1466 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1467 ptr = ecb_i2a_x10 (ptr, u);
1468 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1469 {
1470 uint64_t u1 = u % 1000000000;
1471 uint64_t u2 = u / 1000000000;
1472
1473 ptr = ecb_i2a_x10 (ptr, u2);
1474 ptr = ecb_i2a_09 (ptr, u1);
1475 }
1476 else
1477 {
1478 uint64_t u1 = u % 1000000000;
1479 uint64_t ua = u / 1000000000;
1480 uint64_t u2 = ua % 1000000000;
1481 uint64_t u3 = ua / 1000000000;
1482
1483 ptr = ecb_i2a_2 (ptr, u3);
1484 ptr = ecb_i2a_09 (ptr, u2);
1485 ptr = ecb_i2a_09 (ptr, u1);
1486 }
1487 #else
1488 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1489 ptr = ecb_i2a_x5 (ptr, u);
1490 else
1491 {
1492 uint64_t u1 = u % 10000;
1493 uint64_t u2 = u / 10000;
1494
1495 ptr = ecb_i2a_u64 (ptr, u2);
1496 ptr = ecb_i2a_04 (ptr, u1);
1497 }
1498 #endif
1499
1500 return ptr;
1501 }
1502
1503 ecb_inline char *
1504 ecb_i2a_i64 (char *ptr, int64_t v)
1505 {
1506 *ptr = '-'; ptr += v < 0;
1507 uint64_t u = v < 0 ? -(uint64_t)v : v;
1508
1509 #if ECB_64BIT_NATIVE
1510 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1511 ptr = ecb_i2a_x10 (ptr, u);
1512 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1513 {
1514 uint64_t u1 = u % 1000000000;
1515 uint64_t u2 = u / 1000000000;
1516
1517 ptr = ecb_i2a_x10 (ptr, u2);
1518 ptr = ecb_i2a_09 (ptr, u1);
1519 }
1520 else
1521 {
1522 uint64_t u1 = u % 1000000000;
1523 uint64_t ua = u / 1000000000;
1524 uint64_t u2 = ua % 1000000000;
1525 uint64_t u3 = ua / 1000000000;
1526
1527 /* 2**31 is 19 digits, so the top is exactly one digit */
1528 *ptr++ = u3 + '0';
1529 ptr = ecb_i2a_09 (ptr, u2);
1530 ptr = ecb_i2a_09 (ptr, u1);
1531 }
1532 #else
1533 ptr = ecb_i2a_u64 (ptr, u);
1534 #endif
1535
1536 return ptr;
1537 }
1538
1539 /*******************************************************************************/
1540 /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
1541
1542 /* basically, everything uses "ieee pure-endian" floating point numbers */
1543 /* the only noteworthy exception is ancient armle, which uses order 43218765 */
1544 #if 0 \
1545 || __i386 || __i386__ \
1546 || ECB_GCC_AMD64 \
1547 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
1548 || defined __s390__ || defined __s390x__ \
1549 || defined __mips__ \
1550 || defined __alpha__ \
1551 || defined __hppa__ \
1552 || defined __ia64__ \
1553 || defined __m68k__ \
1554 || defined __m88k__ \
1555 || defined __sh__ \
1556 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1557 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1558 || defined __aarch64__
1559 #define ECB_STDFP 1
1560 #else
1561 #define ECB_STDFP 0
1562 #endif
1563
1564 #ifndef ECB_NO_LIBM
1565
1566 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
1567
1568 /* only the oldest of old doesn't have this one. solaris. */
1569 #ifdef INFINITY
1570 #define ECB_INFINITY INFINITY
1571 #else
1572 #define ECB_INFINITY HUGE_VAL
1573 #endif
1574
1575 #ifdef NAN
1576 #define ECB_NAN NAN
1577 #else
1578 #define ECB_NAN ECB_INFINITY
1579 #endif
1580
1581 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
1582 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1583 #define ecb_frexpf(x,e) frexpf ((x), (e))
1584 #else
1585 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1586 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1587 #endif
1588
1589 /* convert a float to ieee single/binary32 */
1590 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1591 ecb_function_ ecb_const uint32_t
1592 ecb_float_to_binary32 (float x)
1593 {
1594 uint32_t r;
1595
1596 #if ECB_STDFP
1597 memcpy (&r, &x, 4);
1598 #else
1599 /* slow emulation, works for anything but -0 */
1600 uint32_t m;
1601 int e;
1602
1603 if (x == 0e0f ) return 0x00000000U;
1604 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1605 if (x < -3.40282346638528860e+38f) return 0xff800000U;
1606 if (x != x ) return 0x7fbfffffU;
1607
1608 m = ecb_frexpf (x, &e) * 0x1000000U;
1609
1610 r = m & 0x80000000U;
1611
1612 if (r)
1613 m = -m;
1614
1615 if (e <= -126)
1616 {
1617 m &= 0xffffffU;
1618 m >>= (-125 - e);
1619 e = -126;
1620 }
1621
1622 r |= (e + 126) << 23;
1623 r |= m & 0x7fffffU;
1624 #endif
1625
1626 return r;
1627 }
1628
1629 /* converts an ieee single/binary32 to a float */
1630 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1631 ecb_function_ ecb_const float
1632 ecb_binary32_to_float (uint32_t x)
1633 {
1634 float r;
1635
1636 #if ECB_STDFP
1637 memcpy (&r, &x, 4);
1638 #else
1639 /* emulation, only works for normals and subnormals and +0 */
1640 int neg = x >> 31;
1641 int e = (x >> 23) & 0xffU;
1642
1643 x &= 0x7fffffU;
1644
1645 if (e)
1646 x |= 0x800000U;
1647 else
1648 e = 1;
1649
1650 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1651 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1652
1653 r = neg ? -r : r;
1654 #endif
1655
1656 return r;
1657 }
1658
1659 /* convert a double to ieee double/binary64 */
1660 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1661 ecb_function_ ecb_const uint64_t
1662 ecb_double_to_binary64 (double x)
1663 {
1664 uint64_t r;
1665
1666 #if ECB_STDFP
1667 memcpy (&r, &x, 8);
1668 #else
1669 /* slow emulation, works for anything but -0 */
1670 uint64_t m;
1671 int e;
1672
1673 if (x == 0e0 ) return 0x0000000000000000U;
1674 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1675 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1676 if (x != x ) return 0X7ff7ffffffffffffU;
1677
1678 m = frexp (x, &e) * 0x20000000000000U;
1679
1680 r = m & 0x8000000000000000;;
1681
1682 if (r)
1683 m = -m;
1684
1685 if (e <= -1022)
1686 {
1687 m &= 0x1fffffffffffffU;
1688 m >>= (-1021 - e);
1689 e = -1022;
1690 }
1691
1692 r |= ((uint64_t)(e + 1022)) << 52;
1693 r |= m & 0xfffffffffffffU;
1694 #endif
1695
1696 return r;
1697 }
1698
1699 /* converts an ieee double/binary64 to a double */
1700 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1701 ecb_function_ ecb_const double
1702 ecb_binary64_to_double (uint64_t x)
1703 {
1704 double r;
1705
1706 #if ECB_STDFP
1707 memcpy (&r, &x, 8);
1708 #else
1709 /* emulation, only works for normals and subnormals and +0 */
1710 int neg = x >> 63;
1711 int e = (x >> 52) & 0x7ffU;
1712
1713 x &= 0xfffffffffffffU;
1714
1715 if (e)
1716 x |= 0x10000000000000U;
1717 else
1718 e = 1;
1719
1720 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1721 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1722
1723 r = neg ? -r : r;
1724 #endif
1725
1726 return r;
1727 }
1728
1729 /* convert a float to ieee half/binary16 */
1730 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1731 ecb_function_ ecb_const uint16_t
1732 ecb_float_to_binary16 (float x)
1733 {
1734 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1735 }
1736
1737 /* convert an ieee half/binary16 to float */
1738 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1739 ecb_function_ ecb_const float
1740 ecb_binary16_to_float (uint16_t x)
1741 {
1742 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1743 }
1744
1745 #endif
1746
1747 #endif
1748