ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.202
Committed: Wed Mar 23 09:58:06 2022 UTC (2 years, 3 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.201: +61 -1 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 0x0001000b
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 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
460 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
461 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
462 #define ecb_ctz32(x) __builtin_ctz (x)
463 #define ecb_ctz64(x) __builtin_ctzll (x)
464 #define ecb_popcount32(x) __builtin_popcount (x)
465 /* no popcountll */
466 #else
467 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
468 ecb_function_ ecb_const int
469 ecb_ctz32 (uint32_t x)
470 {
471 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
472 unsigned long r;
473 _BitScanForward (&r, x);
474 return (int)r;
475 #else
476 int r = 0;
477
478 x &= ~x + 1; /* this isolates the lowest bit */
479
480 #if ECB_branchless_on_i386
481 r += !!(x & 0xaaaaaaaa) << 0;
482 r += !!(x & 0xcccccccc) << 1;
483 r += !!(x & 0xf0f0f0f0) << 2;
484 r += !!(x & 0xff00ff00) << 3;
485 r += !!(x & 0xffff0000) << 4;
486 #else
487 if (x & 0xaaaaaaaa) r += 1;
488 if (x & 0xcccccccc) r += 2;
489 if (x & 0xf0f0f0f0) r += 4;
490 if (x & 0xff00ff00) r += 8;
491 if (x & 0xffff0000) r += 16;
492 #endif
493
494 return r;
495 #endif
496 }
497
498 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
499 ecb_function_ ecb_const int
500 ecb_ctz64 (uint64_t x)
501 {
502 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
503 unsigned long r;
504 _BitScanForward64 (&r, x);
505 return (int)r;
506 #else
507 int shift = x & 0xffffffff ? 0 : 32;
508 return ecb_ctz32 (x >> shift) + shift;
509 #endif
510 }
511
512 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
513 ecb_function_ ecb_const int
514 ecb_popcount32 (uint32_t x)
515 {
516 x -= (x >> 1) & 0x55555555;
517 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
518 x = ((x >> 4) + x) & 0x0f0f0f0f;
519 x *= 0x01010101;
520
521 return x >> 24;
522 }
523
524 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
525 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
526 {
527 #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
528 unsigned long r;
529 _BitScanReverse (&r, x);
530 return (int)r;
531 #else
532 int r = 0;
533
534 if (x >> 16) { x >>= 16; r += 16; }
535 if (x >> 8) { x >>= 8; r += 8; }
536 if (x >> 4) { x >>= 4; r += 4; }
537 if (x >> 2) { x >>= 2; r += 2; }
538 if (x >> 1) { r += 1; }
539
540 return r;
541 #endif
542 }
543
544 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
545 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
546 {
547 #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
548 unsigned long r;
549 _BitScanReverse64 (&r, x);
550 return (int)r;
551 #else
552 int r = 0;
553
554 if (x >> 32) { x >>= 32; r += 32; }
555
556 return r + ecb_ld32 (x);
557 #endif
558 }
559 #endif
560
561 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
562 ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
563 ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
564 ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
565
566 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
567 ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
568 {
569 return ( (x * 0x0802U & 0x22110U)
570 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
571 }
572
573 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
574 ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
575 {
576 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
577 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
578 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
579 x = ( x >> 8 ) | ( x << 8);
580
581 return x;
582 }
583
584 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
585 ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
586 {
587 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
588 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
589 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
590 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
591 x = ( x >> 16 ) | ( x << 16);
592
593 return x;
594 }
595
596 /* popcount64 is only available on 64 bit cpus as gcc builtin */
597 /* so for this version we are lazy */
598 ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
599 ecb_function_ ecb_const int
600 ecb_popcount64 (uint64_t x)
601 {
602 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
603 }
604
605 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
606 ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
607 ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
608 ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
609 ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
610 ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
611 ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
612 ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
613
614 ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
615 ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
616 ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
617 ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
618 ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
619 ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
620 ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
621 ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
622
623 #if ECB_CPP
624
625 inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
626 inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
627 inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
628 inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
629
630 inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
631 inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
632 inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
633 inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
634
635 inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
636 inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
637 inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
638 inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
639
640 inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
641 inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
642 inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
643 inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
644
645 inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
646 inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
647 inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
648
649 inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
650 inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
651 inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
652 inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
653
654 inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
655 inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
656 inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
657 inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
658
659 #endif
660
661 #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
662 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
663 #define ecb_bswap16(x) __builtin_bswap16 (x)
664 #else
665 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
666 #endif
667 #define ecb_bswap32(x) __builtin_bswap32 (x)
668 #define ecb_bswap64(x) __builtin_bswap64 (x)
669 #elif _MSC_VER
670 #include <stdlib.h>
671 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
672 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
673 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
674 #else
675 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
676 ecb_function_ ecb_const uint16_t
677 ecb_bswap16 (uint16_t x)
678 {
679 return ecb_rotl16 (x, 8);
680 }
681
682 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
683 ecb_function_ ecb_const uint32_t
684 ecb_bswap32 (uint32_t x)
685 {
686 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
687 }
688
689 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
690 ecb_function_ ecb_const uint64_t
691 ecb_bswap64 (uint64_t x)
692 {
693 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
694 }
695 #endif
696
697 #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
698 #define ecb_unreachable() __builtin_unreachable ()
699 #else
700 /* this seems to work fine, but gcc always emits a warning for it :/ */
701 ecb_inline ecb_noreturn void ecb_unreachable (void);
702 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
703 #endif
704
705 /* try to tell the compiler that some condition is definitely true */
706 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
707
708 ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
709 ecb_inline ecb_const uint32_t
710 ecb_byteorder_helper (void)
711 {
712 /* the union code still generates code under pressure in gcc, */
713 /* but less than using pointers, and always seems to */
714 /* successfully return a constant. */
715 /* the reason why we have this horrible preprocessor mess */
716 /* is to avoid it in all cases, at least on common architectures */
717 /* or when using a recent enough gcc version (>= 4.6) */
718 #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
719 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
720 #define ECB_LITTLE_ENDIAN 1
721 return 0x44332211;
722 #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
723 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
724 #define ECB_BIG_ENDIAN 1
725 return 0x11223344;
726 #else
727 union
728 {
729 uint8_t c[4];
730 uint32_t u;
731 } u = { 0x11, 0x22, 0x33, 0x44 };
732 return u.u;
733 #endif
734 }
735
736 ecb_inline ecb_const ecb_bool ecb_big_endian (void);
737 ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
738 ecb_inline ecb_const ecb_bool ecb_little_endian (void);
739 ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
740
741 /*****************************************************************************/
742 /* unaligned load/store */
743
744 ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
745 ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
746 ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
747
748 ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
749 ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
750 ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
751
752 ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753 ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
754 ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
755
756 ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
757 ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
758 ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
759
760 ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
761 ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
762 ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
763
764 ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
765 ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
766 ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
767
768 ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
769 ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
770 ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
771
772 ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
773 ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
774 ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
775
776 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)); }
777 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)); }
778 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)); }
779
780 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)); }
781 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)); }
782 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)); }
783
784 #if ECB_CPP
785
786 inline uint8_t ecb_bswap (uint8_t v) { return v; }
787 inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
788 inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
789 inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
790
791 template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
792 template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
793 template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
794 template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
795 template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
796 template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
797 template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
798 template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
799
800 template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
801 template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
802 template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
803 template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
804 template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
805 template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
806 template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
807 template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
808
809 #endif
810
811 /*****************************************************************************/
812 /* pointer/integer hashing */
813
814 /* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
815 ecb_function_ uint32_t ecb_mix32 (uint32_t v);
816 ecb_function_ uint32_t ecb_mix32 (uint32_t v)
817 {
818 v ^= v >> 16; v *= 0x7feb352dU;
819 v ^= v >> 15; v *= 0x846ca68bU;
820 v ^= v >> 16;
821 return v;
822 }
823
824 ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
825 ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
826 {
827 v ^= v >> 16 ; v *= 0x43021123U;
828 v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
829 v ^= v >> 16 ;
830 return v;
831 }
832
833 /* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
834 ecb_function_ uint64_t ecb_mix64 (uint64_t v);
835 ecb_function_ uint64_t ecb_mix64 (uint64_t v)
836 {
837 v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
838 v ^= v >> 27; v *= 0x94d049bb133111ebU;
839 v ^= v >> 31;
840 return v;
841 }
842
843 ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
844 ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
845 {
846 v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
847 v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
848 v ^= v >> 30 ^ v >> 60;
849 return v;
850 }
851
852 ecb_function_ uintptr_t ecb_ptrmix (void *p);
853 ecb_function_ uintptr_t ecb_ptrmix (void *p)
854 {
855 #if ECB_PTRSIZE <= 4
856 return ecb_mix32 ((uint32_t)p);
857 #else
858 return ecb_mix64 ((uint64_t)p);
859 #endif
860 }
861
862 ecb_function_ void *ecb_ptrunmix (uintptr_t v);
863 ecb_function_ void *ecb_ptrunmix (uintptr_t v)
864 {
865 #if ECB_PTRSIZE <= 4
866 return (void *)ecb_unmix32 (v);
867 #else
868 return (void *)ecb_unmix64 (v);
869 #endif
870 }
871
872 #if ECB_CPP
873
874 template<typename T>
875 inline uintptr_t ecb_ptrmix (T *p)
876 {
877 return ecb_ptrmix (static_cast<void *>(p));
878 }
879
880 template<typename T>
881 inline T *ecb_ptrunmix (uintptr_t v)
882 {
883 return static_cast<T *>(ecb_ptrunmix (v));
884 }
885
886 #endif
887
888 /*****************************************************************************/
889 /* gray code */
890
891 ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
892 ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
893 ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
894 ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
895
896 ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
897 {
898 g = g ^ (g >> 1);
899 g = g ^ (g >> 2);
900 g = g ^ (g >> 4);
901 return g;
902 }
903
904 ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
905 {
906 g = g ^ (g >> 1);
907 g = g ^ (g >> 2);
908 g = g ^ (g >> 4);
909 g = g ^ (g >> 8);
910 return g;
911 }
912
913 ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
914 {
915 g = g ^ (g >> 1);
916 g = g ^ (g >> 2);
917 g = g ^ (g >> 4);
918 g = g ^ (g >> 8);
919 g = g ^ (g >> 16);
920 return g;
921 }
922
923 ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
924 {
925 g = g ^ (g >> 1);
926 g = g ^ (g >> 2);
927 g = g ^ (g >> 4);
928 g = g ^ (g >> 8);
929 g = g ^ (g >> 16);
930 g = g ^ (g >> 32);
931 return g;
932 }
933
934 #if ECB_CPP
935
936 ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
937 ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
938 ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
939 ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
940
941 ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
942 ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
943 ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
944 ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
945
946 #endif
947
948 /*****************************************************************************/
949 /* division */
950
951 #if ECB_GCC_VERSION(3,0) || ECB_C99
952 /* C99 tightened the definition of %, so we can use a more efficient version */
953 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
954 #else
955 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
956 #endif
957
958 #if ECB_CPP
959 template<typename T>
960 static inline T ecb_div_rd (T val, T div)
961 {
962 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
963 }
964 template<typename T>
965 static inline T ecb_div_ru (T val, T div)
966 {
967 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
968 }
969 #else
970 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
971 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
972 #endif
973
974 /*****************************************************************************/
975 /* array length */
976
977 #if ecb_cplusplus_does_not_suck
978 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
979 template<typename T, int N>
980 static inline int ecb_array_length (const T (&arr)[N])
981 {
982 return N;
983 }
984 #else
985 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
986 #endif
987
988 /*****************************************************************************/
989 /* IEEE 754-2008 half float conversions */
990
991 ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
992 ecb_function_ ecb_const uint32_t
993 ecb_binary16_to_binary32 (uint32_t x)
994 {
995 unsigned int s = (x & 0x8000) << (31 - 15);
996 int e = (x >> 10) & 0x001f;
997 unsigned int m = x & 0x03ff;
998
999 if (ecb_expect_false (e == 31))
1000 /* infinity or NaN */
1001 e = 255 - (127 - 15);
1002 else if (ecb_expect_false (!e))
1003 {
1004 if (ecb_expect_true (!m))
1005 /* zero, handled by code below by forcing e to 0 */
1006 e = 0 - (127 - 15);
1007 else
1008 {
1009 /* subnormal, renormalise */
1010 unsigned int s = 10 - ecb_ld32 (m);
1011
1012 m = (m << s) & 0x3ff; /* mask implicit bit */
1013 e -= s - 1;
1014 }
1015 }
1016
1017 /* e and m now are normalised, or zero, (or inf or nan) */
1018 e += 127 - 15;
1019
1020 return s | (e << 23) | (m << (23 - 10));
1021 }
1022
1023 ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1024 ecb_function_ ecb_const uint16_t
1025 ecb_binary32_to_binary16 (uint32_t x)
1026 {
1027 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1028 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1029 unsigned int m = x & 0x007fffff;
1030
1031 x &= 0x7fffffff;
1032
1033 /* if it's within range of binary16 normals, use fast path */
1034 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1035 {
1036 /* mantissa round-to-even */
1037 m += 0x00000fff + ((m >> (23 - 10)) & 1);
1038
1039 /* handle overflow */
1040 if (ecb_expect_false (m >= 0x00800000))
1041 {
1042 m >>= 1;
1043 e += 1;
1044 }
1045
1046 return s | (e << 10) | (m >> (23 - 10));
1047 }
1048
1049 /* handle large numbers and infinity */
1050 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1051 return s | 0x7c00;
1052
1053 /* handle zero, subnormals and small numbers */
1054 if (ecb_expect_true (x < 0x38800000))
1055 {
1056 /* zero */
1057 if (ecb_expect_true (!x))
1058 return s;
1059
1060 /* handle subnormals */
1061
1062 /* too small, will be zero */
1063 if (e < (14 - 24)) /* might not be sharp, but is good enough */
1064 return s;
1065
1066 m |= 0x00800000; /* make implicit bit explicit */
1067
1068 /* very tricky - we need to round to the nearest e (+10) bit value */
1069 {
1070 unsigned int bits = 14 - e;
1071 unsigned int half = (1 << (bits - 1)) - 1;
1072 unsigned int even = (m >> bits) & 1;
1073
1074 /* if this overflows, we will end up with a normalised number */
1075 m = (m + half + even) >> bits;
1076 }
1077
1078 return s | m;
1079 }
1080
1081 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1082 m >>= 13;
1083
1084 return s | 0x7c00 | m | !m;
1085 }
1086
1087 /*******************************************************************************/
1088 /* fast integer to ascii */
1089
1090 /*
1091 * This code is pretty complicated because it is general. The idea behind it,
1092 * however, is pretty simple: first, the number is multiplied with a scaling
1093 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1094 * number with the first digit in the upper bits.
1095 * Then this digit is converted to text and masked out. The resulting number
1096 * is then multiplied by 10, by multiplying the fixed point representation
1097 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1098 * format becomes 5.27, 6.26 and so on.
1099 * The rest involves only advancing the pointer if we already generated a
1100 * non-zero digit, so leading zeroes are overwritten.
1101 */
1102
1103 /* simply return a mask with "bits" bits set */
1104 #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1105
1106 /* oputput a single digit. maskvalue is 10**digitidx */
1107 #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1108 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1109 { \
1110 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1111 *ptr = digit + '0'; /* output it */ \
1112 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1113 ptr += nz; /* output digit only if non-zero digit seen */ \
1114 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1115 }
1116
1117 /* convert integer to fixed point format and multiply out digits, highest first */
1118 /* requires magic constants: max. digits and number of bits after the decimal point */
1119 #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1120 ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1121 { \
1122 char nz = lz; /* non-zero digit seen? */ \
1123 /* convert to x.bits fixed-point */ \
1124 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1125 /* output up to 10 digits */ \
1126 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1127 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1128 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1129 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1130 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1131 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1132 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1133 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1134 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1135 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1136 return ptr; \
1137 }
1138
1139 /* predefined versions of the above, for various digits */
1140 /* ecb_i2a_xN = almost N digits, limit defined by macro */
1141 /* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1142 /* ecb_i2a_0N = exactly N digits, including leading zeroes */
1143
1144 /* non-leading-zero versions, limited range */
1145 #define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1146 #define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1147 ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1148 ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1149
1150 /* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1151 ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1152 ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1153 ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1154 ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1155 ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1156 ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1157 ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1158 ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1159
1160 /* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1161 ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1162 ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1163 ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1164 ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1165 ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1166 ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1167 ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1168 ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1169
1170 #define ECB_I2A_I32_DIGITS 11
1171 #define ECB_I2A_U32_DIGITS 10
1172 #define ECB_I2A_I64_DIGITS 20
1173 #define ECB_I2A_U64_DIGITS 21
1174 #define ECB_I2A_MAX_DIGITS 21
1175
1176 ecb_inline char *
1177 ecb_i2a_u32 (char *ptr, uint32_t u)
1178 {
1179 #if ECB_64BIT_NATIVE
1180 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1181 ptr = ecb_i2a_x10 (ptr, u);
1182 else /* x10 almost, but not fully, covers 32 bit */
1183 {
1184 uint32_t u1 = u % 1000000000;
1185 uint32_t u2 = u / 1000000000;
1186
1187 *ptr++ = u2 + '0';
1188 ptr = ecb_i2a_09 (ptr, u1);
1189 }
1190 #else
1191 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1192 ecb_i2a_x5 (ptr, u);
1193 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1194 {
1195 uint32_t u1 = u % 10000;
1196 uint32_t u2 = u / 10000;
1197
1198 ptr = ecb_i2a_x5 (ptr, u2);
1199 ptr = ecb_i2a_04 (ptr, u1);
1200 }
1201 else
1202 {
1203 uint32_t u1 = u % 10000;
1204 uint32_t ua = u / 10000;
1205 uint32_t u2 = ua % 10000;
1206 uint32_t u3 = ua / 10000;
1207
1208 ptr = ecb_i2a_2 (ptr, u3);
1209 ptr = ecb_i2a_04 (ptr, u2);
1210 ptr = ecb_i2a_04 (ptr, u1);
1211 }
1212 #endif
1213
1214 return ptr;
1215 }
1216
1217 ecb_inline char *
1218 ecb_i2a_i32 (char *ptr, int32_t v)
1219 {
1220 *ptr = '-'; ptr += v < 0;
1221 uint32_t u = v < 0 ? -(uint32_t)v : v;
1222
1223 #if ECB_64BIT_NATIVE
1224 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1225 #else
1226 ptr = ecb_i2a_u32 (ptr, u);
1227 #endif
1228
1229 return ptr;
1230 }
1231
1232 ecb_inline char *
1233 ecb_i2a_u64 (char *ptr, uint64_t u)
1234 {
1235 #if ECB_64BIT_NATIVE
1236 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1237 ptr = ecb_i2a_x10 (ptr, u);
1238 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1239 {
1240 uint64_t u1 = u % 1000000000;
1241 uint64_t u2 = u / 1000000000;
1242
1243 ptr = ecb_i2a_x10 (ptr, u2);
1244 ptr = ecb_i2a_09 (ptr, u1);
1245 }
1246 else
1247 {
1248 uint64_t u1 = u % 1000000000;
1249 uint64_t ua = u / 1000000000;
1250 uint64_t u2 = ua % 1000000000;
1251 uint64_t u3 = ua / 1000000000;
1252
1253 ptr = ecb_i2a_2 (ptr, u3);
1254 ptr = ecb_i2a_09 (ptr, u2);
1255 ptr = ecb_i2a_09 (ptr, u1);
1256 }
1257 #else
1258 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1259 ptr = ecb_i2a_x5 (ptr, u);
1260 else
1261 {
1262 uint64_t u1 = u % 10000;
1263 uint64_t u2 = u / 10000;
1264
1265 ptr = ecb_i2a_u64 (ptr, u2);
1266 ptr = ecb_i2a_04 (ptr, u1);
1267 }
1268 #endif
1269
1270 return ptr;
1271 }
1272
1273 ecb_inline char *
1274 ecb_i2a_i64 (char *ptr, int64_t v)
1275 {
1276 *ptr = '-'; ptr += v < 0;
1277 uint64_t u = v < 0 ? -(uint64_t)v : v;
1278
1279 #if ECB_64BIT_NATIVE
1280 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1281 ptr = ecb_i2a_x10 (ptr, u);
1282 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1283 {
1284 uint64_t u1 = u % 1000000000;
1285 uint64_t u2 = u / 1000000000;
1286
1287 ptr = ecb_i2a_x10 (ptr, u2);
1288 ptr = ecb_i2a_09 (ptr, u1);
1289 }
1290 else
1291 {
1292 uint64_t u1 = u % 1000000000;
1293 uint64_t ua = u / 1000000000;
1294 uint64_t u2 = ua % 1000000000;
1295 uint64_t u3 = ua / 1000000000;
1296
1297 /* 2**31 is 19 digits, so the top is exactly one digit */
1298 *ptr++ = u3 + '0';
1299 ptr = ecb_i2a_09 (ptr, u2);
1300 ptr = ecb_i2a_09 (ptr, u1);
1301 }
1302 #else
1303 ptr = ecb_i2a_u64 (ptr, u);
1304 #endif
1305
1306 return ptr;
1307 }
1308
1309 /*******************************************************************************/
1310 /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
1311
1312 /* basically, everything uses "ieee pure-endian" floating point numbers */
1313 /* the only noteworthy exception is ancient armle, which uses order 43218765 */
1314 #if 0 \
1315 || __i386 || __i386__ \
1316 || ECB_GCC_AMD64 \
1317 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
1318 || defined __s390__ || defined __s390x__ \
1319 || defined __mips__ \
1320 || defined __alpha__ \
1321 || defined __hppa__ \
1322 || defined __ia64__ \
1323 || defined __m68k__ \
1324 || defined __m88k__ \
1325 || defined __sh__ \
1326 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1327 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1328 || defined __aarch64__
1329 #define ECB_STDFP 1
1330 #else
1331 #define ECB_STDFP 0
1332 #endif
1333
1334 #ifndef ECB_NO_LIBM
1335
1336 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
1337
1338 /* only the oldest of old doesn't have this one. solaris. */
1339 #ifdef INFINITY
1340 #define ECB_INFINITY INFINITY
1341 #else
1342 #define ECB_INFINITY HUGE_VAL
1343 #endif
1344
1345 #ifdef NAN
1346 #define ECB_NAN NAN
1347 #else
1348 #define ECB_NAN ECB_INFINITY
1349 #endif
1350
1351 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
1352 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1353 #define ecb_frexpf(x,e) frexpf ((x), (e))
1354 #else
1355 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1356 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1357 #endif
1358
1359 /* convert a float to ieee single/binary32 */
1360 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1361 ecb_function_ ecb_const uint32_t
1362 ecb_float_to_binary32 (float x)
1363 {
1364 uint32_t r;
1365
1366 #if ECB_STDFP
1367 memcpy (&r, &x, 4);
1368 #else
1369 /* slow emulation, works for anything but -0 */
1370 uint32_t m;
1371 int e;
1372
1373 if (x == 0e0f ) return 0x00000000U;
1374 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1375 if (x < -3.40282346638528860e+38f) return 0xff800000U;
1376 if (x != x ) return 0x7fbfffffU;
1377
1378 m = ecb_frexpf (x, &e) * 0x1000000U;
1379
1380 r = m & 0x80000000U;
1381
1382 if (r)
1383 m = -m;
1384
1385 if (e <= -126)
1386 {
1387 m &= 0xffffffU;
1388 m >>= (-125 - e);
1389 e = -126;
1390 }
1391
1392 r |= (e + 126) << 23;
1393 r |= m & 0x7fffffU;
1394 #endif
1395
1396 return r;
1397 }
1398
1399 /* converts an ieee single/binary32 to a float */
1400 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1401 ecb_function_ ecb_const float
1402 ecb_binary32_to_float (uint32_t x)
1403 {
1404 float r;
1405
1406 #if ECB_STDFP
1407 memcpy (&r, &x, 4);
1408 #else
1409 /* emulation, only works for normals and subnormals and +0 */
1410 int neg = x >> 31;
1411 int e = (x >> 23) & 0xffU;
1412
1413 x &= 0x7fffffU;
1414
1415 if (e)
1416 x |= 0x800000U;
1417 else
1418 e = 1;
1419
1420 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1421 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1422
1423 r = neg ? -r : r;
1424 #endif
1425
1426 return r;
1427 }
1428
1429 /* convert a double to ieee double/binary64 */
1430 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1431 ecb_function_ ecb_const uint64_t
1432 ecb_double_to_binary64 (double x)
1433 {
1434 uint64_t r;
1435
1436 #if ECB_STDFP
1437 memcpy (&r, &x, 8);
1438 #else
1439 /* slow emulation, works for anything but -0 */
1440 uint64_t m;
1441 int e;
1442
1443 if (x == 0e0 ) return 0x0000000000000000U;
1444 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1445 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1446 if (x != x ) return 0X7ff7ffffffffffffU;
1447
1448 m = frexp (x, &e) * 0x20000000000000U;
1449
1450 r = m & 0x8000000000000000;;
1451
1452 if (r)
1453 m = -m;
1454
1455 if (e <= -1022)
1456 {
1457 m &= 0x1fffffffffffffU;
1458 m >>= (-1021 - e);
1459 e = -1022;
1460 }
1461
1462 r |= ((uint64_t)(e + 1022)) << 52;
1463 r |= m & 0xfffffffffffffU;
1464 #endif
1465
1466 return r;
1467 }
1468
1469 /* converts an ieee double/binary64 to a double */
1470 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1471 ecb_function_ ecb_const double
1472 ecb_binary64_to_double (uint64_t x)
1473 {
1474 double r;
1475
1476 #if ECB_STDFP
1477 memcpy (&r, &x, 8);
1478 #else
1479 /* emulation, only works for normals and subnormals and +0 */
1480 int neg = x >> 63;
1481 int e = (x >> 52) & 0x7ffU;
1482
1483 x &= 0xfffffffffffffU;
1484
1485 if (e)
1486 x |= 0x10000000000000U;
1487 else
1488 e = 1;
1489
1490 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1491 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1492
1493 r = neg ? -r : r;
1494 #endif
1495
1496 return r;
1497 }
1498
1499 /* convert a float to ieee half/binary16 */
1500 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1501 ecb_function_ ecb_const uint16_t
1502 ecb_float_to_binary16 (float x)
1503 {
1504 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1505 }
1506
1507 /* convert an ieee half/binary16 to float */
1508 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1509 ecb_function_ ecb_const float
1510 ecb_binary16_to_float (uint16_t x)
1511 {
1512 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1513 }
1514
1515 #endif
1516
1517 #endif
1518