ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
(Generate patch)

Comparing libecb/ecb.h (file contents):
Revision 1.11 by root, Thu May 26 18:00:42 2011 UTC vs.
Revision 1.205 by root, Fri Mar 25 14:21:14 2022 UTC

1/* 1/*
2 * libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2011 Marc Alexander Lehmann 4 * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de>
5 * Copyright (©) 2011 Emanuele Giaquinta 5 * Copyright (©) 2011 Emanuele Giaquinta
6 * All rights reserved. 6 * All rights reserved.
7 * 7 *
8 * Redistribution and use in source and binary forms, with or without modifica- 8 * Redistribution and use in source and binary forms, with or without modifica-
9 * tion, are permitted provided that the following conditions are met: 9 * tion, are permitted provided that the following conditions are met:
… …
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 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, 24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- 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 26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27 * OF THE POSSIBILITY OF SUCH DAMAGE. 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.
28 */ 39 */
29 40
30#ifndef ECB_H 41#ifndef ECB_H
31#define ECB_H 42#define ECB_H
32 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
33#include <inttypes.h> 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
34 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
35#define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 125 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
126#endif
36 127
37#ifndef __cplusplus 128#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
38# if __STDC_VERSION__ >= 199901L 129
39# define ECB_INLINE inline 130#if __clang__ && defined __has_builtin
40typedef _Bool ecb_bool; 131 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
41# else 132#else
42# define ECB_INLINE inline /* yeah! */ 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
43typedef int ecb_bool; 352typedef int ecb_bool;
44# endif
45#endif
46 353
47#define ECB_CONCAT_(a, b) a ## b 354#define ECB_CONCAT_(a, b) a ## b
48#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
49#define ECB_STRINGIFY_(a) # a 356#define ECB_STRINGIFY_(a) # a
50#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 357#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
51 359
52#define ECB_HEADER_INLINE static ECB_INLINE 360#define ecb_function_ ecb_inline
53 361
54#if ECB_GCC_VERSION(3,1) 362#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
55# define ecb_attribute(attrlist) __attribute__(attrlist) 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)
56# define ecb_is_constant(c) __builtin_constant_p (c) 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)
57# define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 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)
58# define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 385 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
59#else 386#else
60# define ecb_attribute(attrlist)
61# define ecb_is_constant(c) 0
62# define ecb_expect(expr,value) (expr)
63# define ecb_prefetch(addr,rw,locality) 387 #define ecb_prefetch(addr,rw,locality)
64#endif 388#endif
65 389
66/* no emulation for ecb_decltype */ 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))
67#if ECB_GCC_VERSION(4,5) 407#elif ECB_GCC_VERSION(4,5)
68# define ecb_decltype(x) __decltype(x) 408 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
69#elif ECB_GCC_VERSION(3,0) 409#else
70# define ecb_decltype(x) typeof(x) 410 #define ecb_deprecated_message(msg) ecb_deprecated
71#endif 411#endif
72 412
413#if _MSC_VER >= 1400
414 #define ecb_noinline __declspec (noinline)
415#else
73#define ecb_noinline ecb_attribute ((noinline)) 416 #define ecb_noinline ecb_attribute ((__noinline__))
74#define ecb_noreturn ecb_attribute ((noreturn)) 417#endif
418
75#define ecb_unused ecb_attribute ((unused)) 419#define ecb_unused ecb_attribute ((__unused__))
76#define ecb_const ecb_attribute ((const)) 420#define ecb_const ecb_attribute ((__const__))
77#define ecb_pure ecb_attribute ((pure)) 421#define ecb_pure ecb_attribute ((__pure__))
78#define ecb_hot ecb_attribute ((hot)) /* 4.3 */
79#define ecb_cold ecb_attribute ((cold)) /* 4.3 */
80 422
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
81#if ECB_GCC_VERSION(4,5) 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 /* todo: use david seal's algorithm */
479
480 x &= ~x + 1; /* this isolates the lowest bit */
481
482#if ECB_branchless_on_i386
483 r += !!(x & 0xaaaaaaaa) << 0;
484 r += !!(x & 0xcccccccc) << 1;
485 r += !!(x & 0xf0f0f0f0) << 2;
486 r += !!(x & 0xff00ff00) << 3;
487 r += !!(x & 0xffff0000) << 4;
488#else
489 if (x & 0xaaaaaaaa) r += 1;
490 if (x & 0xcccccccc) r += 2;
491 if (x & 0xf0f0f0f0) r += 4;
492 if (x & 0xff00ff00) r += 8;
493 if (x & 0xffff0000) r += 16;
494#endif
495
496 return r;
497#endif
498 }
499
500 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
501 ecb_function_ ecb_const int
502 ecb_ctz64 (uint64_t x)
503 {
504#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
505 unsigned long r;
506 _BitScanForward64 (&r, x);
507 return (int)r;
508#else
509 int shift = x & 0xffffffff ? 0 : 32;
510 return ecb_ctz32 (x >> shift) + shift;
511#endif
512 }
513
514 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
515 ecb_function_ ecb_const int
516 ecb_popcount32 (uint32_t x)
517 {
518 x -= (x >> 1) & 0x55555555;
519 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
520 x = ((x >> 4) + x) & 0x0f0f0f0f;
521 x *= 0x01010101;
522
523 return x >> 24;
524 }
525
526 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
527 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
528 {
529#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
530 unsigned long r;
531 _BitScanReverse (&r, x);
532 return (int)r;
533#else
534 int r = 0;
535
536 if (x >> 16) { x >>= 16; r += 16; }
537 if (x >> 8) { x >>= 8; r += 8; }
538 if (x >> 4) { x >>= 4; r += 4; }
539 if (x >> 2) { x >>= 2; r += 2; }
540 if (x >> 1) { r += 1; }
541
542 return r;
543#endif
544 }
545
546 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
547 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
548 {
549#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
550 unsigned long r;
551 _BitScanReverse64 (&r, x);
552 return (int)r;
553#else
554 int r = 0;
555
556 if (x >> 32) { x >>= 32; r += 32; }
557
558 return r + ecb_ld32 (x);
559#endif
560 }
561#endif
562
563ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
564ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
565ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
566ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
567
568ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
569ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
570{
571 return ( (x * 0x0802U & 0x22110U)
572 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
573}
574
575ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
576ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
577{
578 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
579 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
580 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
581 x = ( x >> 8 ) | ( x << 8);
582
583 return x;
584}
585
586ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
587ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
588{
589 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
590 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
591 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
592 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
593 x = ( x >> 16 ) | ( x << 16);
594
595 return x;
596}
597
598ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
599ecb_function_ ecb_const int
600ecb_popcount64 (uint64_t x)
601{
602 /* popcount64 is only available on 64 bit cpus as gcc builtin. */
603 /* also, gcc/clang make this surprisingly difficult to use */
604#if __LP64__ && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
605 return __builtin_popcountl (x);
606#else
607 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
608#endif
609}
610
611ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
612ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
613ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
614ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
615ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
616ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
617ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
618ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
619
620ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
621ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
622ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
623ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
624ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
625ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
626ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
627ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
628
629#if ECB_CPP
630
631inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
632inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
633inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
634inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
635
636inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
637inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
638inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
639inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
640
641inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
642inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
643inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
644inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
645
646inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
647inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
648inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
649inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
650
651inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
652inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
653inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
654
655inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
656inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
657inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
658inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
659
660inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
661inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
662inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
663inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
664
665#endif
666
667#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
668 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
669 #define ecb_bswap16(x) __builtin_bswap16 (x)
670 #else
671 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
672 #endif
673 #define ecb_bswap32(x) __builtin_bswap32 (x)
674 #define ecb_bswap64(x) __builtin_bswap64 (x)
675#elif _MSC_VER
676 #include <stdlib.h>
677 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
678 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
679 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
680#else
681 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
682 ecb_function_ ecb_const uint16_t
683 ecb_bswap16 (uint16_t x)
684 {
685 return ecb_rotl16 (x, 8);
686 }
687
688 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
689 ecb_function_ ecb_const uint32_t
690 ecb_bswap32 (uint32_t x)
691 {
692 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
693 }
694
695 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
696 ecb_function_ ecb_const uint64_t
697 ecb_bswap64 (uint64_t x)
698 {
699 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
700 }
701#endif
702
703#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
82# define ecb_unreachable() __builtin_unreachable () 704 #define ecb_unreachable() __builtin_unreachable ()
83#else 705#else
84/* this seems to work fine, but gcc always emits a warning for it :/ */ 706 /* this seems to work fine, but gcc always emits a warning for it :/ */
85ECB_HEADER_INLINE void ecb_unreachable () ecb_attribute ((noreturn)); 707 ecb_inline ecb_noreturn void ecb_unreachable (void);
86ECB_HEADER_INLINE void ecb_unreachable () { } 708 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
87#endif 709#endif
88
89/* put into ifs if you are very sure that the expression */
90/* is mostly true or mosty false. note that these return */
91/* booleans, not the expression. */
92#define ecb_unlikely(expr) ecb_expect (!!(expr), 0)
93#define ecb_likely(expr) ecb_expect (!!(expr), 1)
94 710
95/* try to tell the compiler that some condition is definitely true */ 711/* try to tell the compiler that some condition is definitely true */
96#define ecb_assume(cond) do { if (!(cond)) unreachable (); } while (0) 712#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
97 713
98/* count trailing zero bits and count # of one bits */ 714ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
99#if ECB_GCC_VERSION(3,4) 715ecb_inline ecb_const uint32_t
100ECB_HEADER_INLINE int ecb_ctz32 (uint32_t x) { return __builtin_ctz (x); } 716ecb_byteorder_helper (void)
101ECB_HEADER_INLINE int ecb_popcount32 (uint32_t x) { return __builtin_popcount (x); }
102#else
103ECB_HEADER_INLINE
104ecb_ctz32 (uint32_t x) ecb_const
105{ 717{
106 int r = 0; 718 /* the union code still generates code under pressure in gcc, */
719 /* but less than using pointers, and always seems to */
720 /* successfully return a constant. */
721 /* the reason why we have this horrible preprocessor mess */
722 /* is to avoid it in all cases, at least on common architectures */
723 /* or when using a recent enough gcc version (>= 4.6) */
724#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
725 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
726 #define ECB_LITTLE_ENDIAN 1
727 return 0x44332211;
728#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
729 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
730 #define ECB_BIG_ENDIAN 1
731 return 0x11223344;
732#else
733 union
734 {
735 uint8_t c[4];
736 uint32_t u;
737 } u = { 0x11, 0x22, 0x33, 0x44 };
738 return u.u;
739#endif
740}
107 741
108 x &= -x; /* this isolates the lowest bit */ 742ecb_inline ecb_const ecb_bool ecb_big_endian (void);
743ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
744ecb_inline ecb_const ecb_bool ecb_little_endian (void);
745ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
109 746
110 if (x & 0xaaaaaaaa) r += 1; 747/*****************************************************************************/
111 if (x & 0xcccccccc) r += 2; 748/* unaligned load/store */
112 if (x & 0xf0f0f0f0) r += 4;
113 if (x & 0xff00ff00) r += 8;
114 if (x & 0xffff0000) r += 16;
115 749
750ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
751ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
752ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
753
754ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
755ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
756ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
757
758ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
759ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
760ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
761
762ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
763ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
764ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
765
766ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
767ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
768ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
769
770ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
771ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
772ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
773
774ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
775ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
776ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
777
778ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
779ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
780ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
781
782ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
783ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
784ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
785
786ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
787ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
788ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
789
790#if ECB_CPP
791
792inline uint8_t ecb_bswap (uint8_t v) { return v; }
793inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
794inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
795inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
796
797template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
798template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
799template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
800template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
801template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
802template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
803template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
804template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
805
806template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
807template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
808template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
809template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
810template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
811template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
812template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
813template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
814
815#endif
816
817/*****************************************************************************/
818/* pointer/integer hashing */
819
820/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
821ecb_function_ uint32_t ecb_mix32 (uint32_t v);
822ecb_function_ uint32_t ecb_mix32 (uint32_t v)
823{
824 v ^= v >> 16; v *= 0x7feb352dU;
825 v ^= v >> 15; v *= 0x846ca68bU;
826 v ^= v >> 16;
116 return r; 827 return v;
117} 828}
118 829
119ECB_HEADER_INLINE 830ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
120ecb_popcount32 (uint32_t x) ecb_const 831ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
121{ 832{
122 x -= (x >> 1) & 0x55555555; 833 v ^= v >> 16 ; v *= 0x43021123U;
123 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 834 v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
124 x = ((x >> 4) + x) & 0x0f0f0f0f; 835 v ^= v >> 16 ;
125 x *= 0x01010101; 836 return v;
126
127 return x >> 24;
128} 837}
838
839/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
840ecb_function_ uint64_t ecb_mix64 (uint64_t v);
841ecb_function_ uint64_t ecb_mix64 (uint64_t v)
842{
843 v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
844 v ^= v >> 27; v *= 0x94d049bb133111ebU;
845 v ^= v >> 31;
846 return v;
847}
848
849ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
850ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
851{
852 v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
853 v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
854 v ^= v >> 30 ^ v >> 60;
855 return v;
856}
857
858ecb_function_ uintptr_t ecb_ptrmix (void *p);
859ecb_function_ uintptr_t ecb_ptrmix (void *p)
860{
861 #if ECB_PTRSIZE <= 4
862 return ecb_mix32 ((uint32_t)p);
863 #else
864 return ecb_mix64 ((uint64_t)p);
129#endif 865 #endif
130
131ECB_HEADER_INLINE unsigned char
132ecb_byteorder_helper () ecb_const
133{
134 const uint32_t u = 0x11223344;
135 return *(unsigned char *)&u;
136} 866}
137 867
138ECB_HEADER_INLINE ecb_bool ecb_big_endian () ecb_const { return ecb_byteorder_helper () == 0x11; }; 868ecb_function_ void *ecb_ptrunmix (uintptr_t v);
139ECB_HEADER_INLINE ecb_bool ecb_little_endian () ecb_const { return ecb_byteorder_helper () == 0x44; }; 869ecb_function_ void *ecb_ptrunmix (uintptr_t v)
870{
871 #if ECB_PTRSIZE <= 4
872 return (void *)ecb_unmix32 (v);
873 #else
874 return (void *)ecb_unmix64 (v);
875 #endif
876}
877
878#if ECB_CPP
879
880template<typename T>
881inline uintptr_t ecb_ptrmix (T *p)
882{
883 return ecb_ptrmix (static_cast<void *>(p));
884}
885
886template<typename T>
887inline T *ecb_ptrunmix (uintptr_t v)
888{
889 return static_cast<T *>(ecb_ptrunmix (v));
890}
891
892#endif
893
894/*****************************************************************************/
895/* gray code */
896
897ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
898ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
899ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
900ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
901
902ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
903{
904 g ^= g >> 1;
905 g ^= g >> 2;
906 g ^= g >> 4;
907
908 return g;
909}
910
911ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
912{
913 g ^= g >> 1;
914 g ^= g >> 2;
915 g ^= g >> 4;
916 g ^= g >> 8;
917
918 return g;
919}
920
921ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
922{
923 g ^= g >> 1;
924 g ^= g >> 2;
925 g ^= g >> 4;
926 g ^= g >> 8;
927 g ^= g >> 16;
928
929 return g;
930}
931
932ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
933{
934 g ^= g >> 1;
935 g ^= g >> 2;
936 g ^= g >> 4;
937 g ^= g >> 8;
938 g ^= g >> 16;
939 g ^= g >> 32;
940
941 return g;
942}
943
944#if ECB_CPP
945
946ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
947ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
948ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
949ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
950
951ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
952ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
953ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
954ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
955
956#endif
957
958/*****************************************************************************/
959/* 2d hilbert curves */
960
961/* algorithm from the book Hacker's Delight, modified to not */
962/* run into undefined behaviour for n==16 */
963static uint32_t
964ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
965{
966 uint32_t comp, swap, cs, t, sr;
967
968 /* pad s on the left (unused) bits with 01 (no change groups) */
969 s |= 0x55555555U << n << n;
970 /* "s shift right" */
971 sr = (s >> 1) & 0x55555555U;
972 /* compute complement and swap info in two-bit groups */
973 cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
974
975 /* parallel prefix xor op to propagate both complement
976 * and swap info together from left to right (there is
977 * no step "cs ^= cs >> 1", so in effect it computes
978 * two independent parallel prefix operations on two
979 * interleaved sets of sixteen bits).
980 */
981 cs ^= cs >> 2;
982 cs ^= cs >> 4;
983 cs ^= cs >> 8;
984 cs ^= cs >> 16;
985
986 /* separate swap and complement bits */
987 swap = cs & 0x55555555U;
988 comp = (cs >> 1) & 0x55555555U;
989
990 /* calculate coordinates in odd and even bit positions */
991 t = (s & swap) ^ comp;
992 s = s ^ sr ^ t ^ (t << 1);
993
994 /* unpad/clear out any junk on the left */
995 s = s & ((1 << n << n) - 1);
996
997 /* Now "unshuffle" to separate the x and y bits. */
998 t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
999 t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
1000 t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
1001 t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
1002
1003 /* now s contains two 16-bit coordinates */
1004 return s;
1005}
1006
1007/* 64 bit, a straightforward extension to the 32 bit case */
1008static uint64_t
1009ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1010{
1011 uint64_t comp, swap, cs, t, sr;
1012
1013 /* pad s on the left (unused) bits with 01 (no change groups) */
1014 s |= 0x5555555555555555U << n << n;
1015 /* "s shift right" */
1016 sr = (s >> 1) & 0x5555555555555555U;
1017 /* compute complement and swap info in two-bit groups */
1018 cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1019
1020 /* parallel prefix xor op to propagate both complement
1021 * and swap info together from left to right (there is
1022 * no step "cs ^= cs >> 1", so in effect it computes
1023 * two independent parallel prefix operations on two
1024 * interleaved sets of thirty-two bits).
1025 */
1026 cs ^= cs >> 2;
1027 cs ^= cs >> 4;
1028 cs ^= cs >> 8;
1029 cs ^= cs >> 16;
1030 cs ^= cs >> 32;
1031
1032 /* separate swap and complement bits */
1033 swap = cs & 0x5555555555555555U;
1034 comp = (cs >> 1) & 0x5555555555555555U;
1035
1036 /* calculate coordinates in odd and even bit positions */
1037 t = (s & swap) ^ comp;
1038 s = s ^ sr ^ t ^ (t << 1);
1039
1040 /* unpad/clear out any junk on the left */
1041 s = s & ((1 << n << n) - 1);
1042
1043 /* Now "unshuffle" to separate the x and y bits. */
1044 t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1045 t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1046 t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1047 t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1048 t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1049
1050 /* now s contains two 32-bit coordinates */
1051 return s;
1052}
1053
1054/* algorithm from the book Hacker's Delight, but a similar algorithm*/
1055/* is given in https://doi.org/10.1002/spe.4380160103 */
1056/* this has been slightly improved over the original version */
1057ecb_function_ uint32_t
1058ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1059{
1060 uint32_t row;
1061 uint32_t state = 0;
1062 uint32_t s = 0;
1063
1064 do
1065 {
1066 --n;
1067
1068 row = 4 * state
1069 | (2 & (xy >> n >> 15))
1070 | (1 & (xy >> n ));
1071
1072 /* these funky constants are lookup tables for two-bit values */
1073 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1074 state = (0x8fe65831U >> 2 * row) & 3;
1075 }
1076 while (n > 0);
1077
1078 return s;
1079}
1080
1081/* 64 bit, essentially the same as 32 bit */
1082ecb_function_ uint64_t
1083ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1084{
1085 uint32_t row;
1086 uint32_t state = 0;
1087 uint64_t s = 0;
1088
1089 do
1090 {
1091 --n;
1092
1093 row = 4 * state
1094 | (2 & (xy >> n >> 31))
1095 | (1 & (xy >> n ));
1096
1097 /* these funky constants are lookup tables for two-bit values */
1098 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1099 state = (0x8fe65831U >> 2 * row) & 3;
1100 }
1101 while (n > 0);
1102
1103 return s;
1104}
1105
1106/*****************************************************************************/
1107/* division */
1108
1109#if ECB_GCC_VERSION(3,0) || ECB_C99
1110 /* C99 tightened the definition of %, so we can use a more efficient version */
1111 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
1112#else
1113 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
1114#endif
1115
1116#if ECB_CPP
1117 template<typename T>
1118 static inline T ecb_div_rd (T val, T div)
1119 {
1120 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
1121 }
1122 template<typename T>
1123 static inline T ecb_div_ru (T val, T div)
1124 {
1125 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
1126 }
1127#else
1128 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
1129 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
1130#endif
1131
1132/*****************************************************************************/
1133/* array length */
140 1134
141#if ecb_cplusplus_does_not_suck 1135#if ecb_cplusplus_does_not_suck
142// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) 1136 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
143template<typename T, int N> 1137 template<typename T, int N>
144static inline int ecb_array_length (const T (&arr)[N]) 1138 static inline int ecb_array_length (const T (&arr)[N])
145{ 1139 {
146 return N; 1140 return N;
147} 1141 }
148#else 1142#else
149#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 1143 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
1144#endif
1145
1146/*****************************************************************************/
1147/* IEEE 754-2008 half float conversions */
1148
1149ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
1150ecb_function_ ecb_const uint32_t
1151ecb_binary16_to_binary32 (uint32_t x)
1152{
1153 unsigned int s = (x & 0x8000) << (31 - 15);
1154 int e = (x >> 10) & 0x001f;
1155 unsigned int m = x & 0x03ff;
1156
1157 if (ecb_expect_false (e == 31))
1158 /* infinity or NaN */
1159 e = 255 - (127 - 15);
1160 else if (ecb_expect_false (!e))
1161 {
1162 if (ecb_expect_true (!m))
1163 /* zero, handled by code below by forcing e to 0 */
1164 e = 0 - (127 - 15);
1165 else
1166 {
1167 /* subnormal, renormalise */
1168 unsigned int s = 10 - ecb_ld32 (m);
1169
1170 m = (m << s) & 0x3ff; /* mask implicit bit */
1171 e -= s - 1;
1172 }
1173 }
1174
1175 /* e and m now are normalised, or zero, (or inf or nan) */
1176 e += 127 - 15;
1177
1178 return s | (e << 23) | (m << (23 - 10));
1179}
1180
1181ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1182ecb_function_ ecb_const uint16_t
1183ecb_binary32_to_binary16 (uint32_t x)
1184{
1185 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1186 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1187 unsigned int m = x & 0x007fffff;
1188
1189 x &= 0x7fffffff;
1190
1191 /* if it's within range of binary16 normals, use fast path */
1192 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1193 {
1194 /* mantissa round-to-even */
1195 m += 0x00000fff + ((m >> (23 - 10)) & 1);
1196
1197 /* handle overflow */
1198 if (ecb_expect_false (m >= 0x00800000))
1199 {
1200 m >>= 1;
1201 e += 1;
1202 }
1203
1204 return s | (e << 10) | (m >> (23 - 10));
1205 }
1206
1207 /* handle large numbers and infinity */
1208 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1209 return s | 0x7c00;
1210
1211 /* handle zero, subnormals and small numbers */
1212 if (ecb_expect_true (x < 0x38800000))
1213 {
1214 /* zero */
1215 if (ecb_expect_true (!x))
1216 return s;
1217
1218 /* handle subnormals */
1219
1220 /* too small, will be zero */
1221 if (e < (14 - 24)) /* might not be sharp, but is good enough */
1222 return s;
1223
1224 m |= 0x00800000; /* make implicit bit explicit */
1225
1226 /* very tricky - we need to round to the nearest e (+10) bit value */
1227 {
1228 unsigned int bits = 14 - e;
1229 unsigned int half = (1 << (bits - 1)) - 1;
1230 unsigned int even = (m >> bits) & 1;
1231
1232 /* if this overflows, we will end up with a normalised number */
1233 m = (m + half + even) >> bits;
1234 }
1235
1236 return s | m;
1237 }
1238
1239 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1240 m >>= 13;
1241
1242 return s | 0x7c00 | m | !m;
1243}
1244
1245/*******************************************************************************/
1246/* fast integer to ascii */
1247
1248/*
1249 * This code is pretty complicated because it is general. The idea behind it,
1250 * however, is pretty simple: first, the number is multiplied with a scaling
1251 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1252 * number with the first digit in the upper bits.
1253 * Then this digit is converted to text and masked out. The resulting number
1254 * is then multiplied by 10, by multiplying the fixed point representation
1255 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1256 * format becomes 5.27, 6.26 and so on.
1257 * The rest involves only advancing the pointer if we already generated a
1258 * non-zero digit, so leading zeroes are overwritten.
1259 */
1260
1261/* simply return a mask with "bits" bits set */
1262#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1263
1264/* oputput a single digit. maskvalue is 10**digitidx */
1265#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1266 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1267 { \
1268 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1269 *ptr = digit + '0'; /* output it */ \
1270 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1271 ptr += nz; /* output digit only if non-zero digit seen */ \
1272 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1273 }
1274
1275/* convert integer to fixed point format and multiply out digits, highest first */
1276/* requires magic constants: max. digits and number of bits after the decimal point */
1277#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1278ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1279{ \
1280 char nz = lz; /* non-zero digit seen? */ \
1281 /* convert to x.bits fixed-point */ \
1282 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1283 /* output up to 10 digits */ \
1284 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1285 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1286 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1287 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1288 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1289 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1290 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1291 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1292 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1293 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1294 return ptr; \
1295}
1296
1297/* predefined versions of the above, for various digits */
1298/* ecb_i2a_xN = almost N digits, limit defined by macro */
1299/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1300/* ecb_i2a_0N = exactly N digits, including leading zeroes */
1301
1302/* non-leading-zero versions, limited range */
1303#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1304#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1305ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1306ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1307
1308/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1309ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1310ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1311ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1312ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1313ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1314ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1315ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1316ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1317
1318/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1319ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1320ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1321ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1322ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1323ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1324ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1325ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1326ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1327
1328#define ECB_I2A_I32_DIGITS 11
1329#define ECB_I2A_U32_DIGITS 10
1330#define ECB_I2A_I64_DIGITS 20
1331#define ECB_I2A_U64_DIGITS 21
1332#define ECB_I2A_MAX_DIGITS 21
1333
1334ecb_inline char *
1335ecb_i2a_u32 (char *ptr, uint32_t u)
1336{
1337 #if ECB_64BIT_NATIVE
1338 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1339 ptr = ecb_i2a_x10 (ptr, u);
1340 else /* x10 almost, but not fully, covers 32 bit */
1341 {
1342 uint32_t u1 = u % 1000000000;
1343 uint32_t u2 = u / 1000000000;
1344
1345 *ptr++ = u2 + '0';
1346 ptr = ecb_i2a_09 (ptr, u1);
1347 }
1348 #else
1349 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1350 ecb_i2a_x5 (ptr, u);
1351 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1352 {
1353 uint32_t u1 = u % 10000;
1354 uint32_t u2 = u / 10000;
1355
1356 ptr = ecb_i2a_x5 (ptr, u2);
1357 ptr = ecb_i2a_04 (ptr, u1);
1358 }
1359 else
1360 {
1361 uint32_t u1 = u % 10000;
1362 uint32_t ua = u / 10000;
1363 uint32_t u2 = ua % 10000;
1364 uint32_t u3 = ua / 10000;
1365
1366 ptr = ecb_i2a_2 (ptr, u3);
1367 ptr = ecb_i2a_04 (ptr, u2);
1368 ptr = ecb_i2a_04 (ptr, u1);
1369 }
150#endif 1370 #endif
151 1371
152ECB_INLINE uint32_t 1372 return ptr;
153ecb_rotr32 (uint32_t x, unsigned int count)
154{
155 return (x << (32 - count)) | (x >> count);
156} 1373}
157 1374
158ECB_INLINE uint32_t 1375ecb_inline char *
159ecb_rotl32 (uint32_t x, unsigned int count) 1376ecb_i2a_i32 (char *ptr, int32_t v)
160{ 1377{
161 return (x >> (32 - count)) | (x << count); 1378 *ptr = '-'; ptr += v < 0;
162} 1379 uint32_t u = v < 0 ? -(uint32_t)v : v;
163 1380
164#if ECB_GCC_VERSION(4,3) 1381 #if ECB_64BIT_NATIVE
165# define ecb_bswap32(x) __builtin_bswap32 (x) 1382 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
166#else 1383 #else
167ECB_GCC_VERSION uint32_t 1384 ptr = ecb_i2a_u32 (ptr, u);
168ecb_bswap32 (uint32_t x)
169{
170 return (x >> 24)
171 | ((x >> 8) & 0x0000ff00)
172 | ((x << 8) & 0x00ff0000)
173 | (x << 24);
174}
175#endif 1385 #endif
176 1386
1387 return ptr;
1388}
1389
1390ecb_inline char *
1391ecb_i2a_u64 (char *ptr, uint64_t u)
1392{
1393 #if ECB_64BIT_NATIVE
1394 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1395 ptr = ecb_i2a_x10 (ptr, u);
1396 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1397 {
1398 uint64_t u1 = u % 1000000000;
1399 uint64_t u2 = u / 1000000000;
1400
1401 ptr = ecb_i2a_x10 (ptr, u2);
1402 ptr = ecb_i2a_09 (ptr, u1);
1403 }
1404 else
1405 {
1406 uint64_t u1 = u % 1000000000;
1407 uint64_t ua = u / 1000000000;
1408 uint64_t u2 = ua % 1000000000;
1409 uint64_t u3 = ua / 1000000000;
1410
1411 ptr = ecb_i2a_2 (ptr, u3);
1412 ptr = ecb_i2a_09 (ptr, u2);
1413 ptr = ecb_i2a_09 (ptr, u1);
1414 }
1415 #else
1416 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1417 ptr = ecb_i2a_x5 (ptr, u);
1418 else
1419 {
1420 uint64_t u1 = u % 10000;
1421 uint64_t u2 = u / 10000;
1422
1423 ptr = ecb_i2a_u64 (ptr, u2);
1424 ptr = ecb_i2a_04 (ptr, u1);
1425 }
177#endif 1426 #endif
178 1427
1428 return ptr;
1429}
1430
1431ecb_inline char *
1432ecb_i2a_i64 (char *ptr, int64_t v)
1433{
1434 *ptr = '-'; ptr += v < 0;
1435 uint64_t u = v < 0 ? -(uint64_t)v : v;
1436
1437 #if ECB_64BIT_NATIVE
1438 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1439 ptr = ecb_i2a_x10 (ptr, u);
1440 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1441 {
1442 uint64_t u1 = u % 1000000000;
1443 uint64_t u2 = u / 1000000000;
1444
1445 ptr = ecb_i2a_x10 (ptr, u2);
1446 ptr = ecb_i2a_09 (ptr, u1);
1447 }
1448 else
1449 {
1450 uint64_t u1 = u % 1000000000;
1451 uint64_t ua = u / 1000000000;
1452 uint64_t u2 = ua % 1000000000;
1453 uint64_t u3 = ua / 1000000000;
1454
1455 /* 2**31 is 19 digits, so the top is exactly one digit */
1456 *ptr++ = u3 + '0';
1457 ptr = ecb_i2a_09 (ptr, u2);
1458 ptr = ecb_i2a_09 (ptr, u1);
1459 }
1460 #else
1461 ptr = ecb_i2a_u64 (ptr, u);
1462 #endif
1463
1464 return ptr;
1465}
1466
1467/*******************************************************************************/
1468/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
1469
1470/* basically, everything uses "ieee pure-endian" floating point numbers */
1471/* the only noteworthy exception is ancient armle, which uses order 43218765 */
1472#if 0 \
1473 || __i386 || __i386__ \
1474 || ECB_GCC_AMD64 \
1475 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
1476 || defined __s390__ || defined __s390x__ \
1477 || defined __mips__ \
1478 || defined __alpha__ \
1479 || defined __hppa__ \
1480 || defined __ia64__ \
1481 || defined __m68k__ \
1482 || defined __m88k__ \
1483 || defined __sh__ \
1484 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1485 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1486 || defined __aarch64__
1487 #define ECB_STDFP 1
1488#else
1489 #define ECB_STDFP 0
1490#endif
1491
1492#ifndef ECB_NO_LIBM
1493
1494 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
1495
1496 /* only the oldest of old doesn't have this one. solaris. */
1497 #ifdef INFINITY
1498 #define ECB_INFINITY INFINITY
1499 #else
1500 #define ECB_INFINITY HUGE_VAL
1501 #endif
1502
1503 #ifdef NAN
1504 #define ECB_NAN NAN
1505 #else
1506 #define ECB_NAN ECB_INFINITY
1507 #endif
1508
1509 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
1510 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1511 #define ecb_frexpf(x,e) frexpf ((x), (e))
1512 #else
1513 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1514 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1515 #endif
1516
1517 /* convert a float to ieee single/binary32 */
1518 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1519 ecb_function_ ecb_const uint32_t
1520 ecb_float_to_binary32 (float x)
1521 {
1522 uint32_t r;
1523
1524 #if ECB_STDFP
1525 memcpy (&r, &x, 4);
1526 #else
1527 /* slow emulation, works for anything but -0 */
1528 uint32_t m;
1529 int e;
1530
1531 if (x == 0e0f ) return 0x00000000U;
1532 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1533 if (x < -3.40282346638528860e+38f) return 0xff800000U;
1534 if (x != x ) return 0x7fbfffffU;
1535
1536 m = ecb_frexpf (x, &e) * 0x1000000U;
1537
1538 r = m & 0x80000000U;
1539
1540 if (r)
1541 m = -m;
1542
1543 if (e <= -126)
1544 {
1545 m &= 0xffffffU;
1546 m >>= (-125 - e);
1547 e = -126;
1548 }
1549
1550 r |= (e + 126) << 23;
1551 r |= m & 0x7fffffU;
1552 #endif
1553
1554 return r;
1555 }
1556
1557 /* converts an ieee single/binary32 to a float */
1558 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1559 ecb_function_ ecb_const float
1560 ecb_binary32_to_float (uint32_t x)
1561 {
1562 float r;
1563
1564 #if ECB_STDFP
1565 memcpy (&r, &x, 4);
1566 #else
1567 /* emulation, only works for normals and subnormals and +0 */
1568 int neg = x >> 31;
1569 int e = (x >> 23) & 0xffU;
1570
1571 x &= 0x7fffffU;
1572
1573 if (e)
1574 x |= 0x800000U;
1575 else
1576 e = 1;
1577
1578 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1579 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1580
1581 r = neg ? -r : r;
1582 #endif
1583
1584 return r;
1585 }
1586
1587 /* convert a double to ieee double/binary64 */
1588 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1589 ecb_function_ ecb_const uint64_t
1590 ecb_double_to_binary64 (double x)
1591 {
1592 uint64_t r;
1593
1594 #if ECB_STDFP
1595 memcpy (&r, &x, 8);
1596 #else
1597 /* slow emulation, works for anything but -0 */
1598 uint64_t m;
1599 int e;
1600
1601 if (x == 0e0 ) return 0x0000000000000000U;
1602 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1603 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1604 if (x != x ) return 0X7ff7ffffffffffffU;
1605
1606 m = frexp (x, &e) * 0x20000000000000U;
1607
1608 r = m & 0x8000000000000000;;
1609
1610 if (r)
1611 m = -m;
1612
1613 if (e <= -1022)
1614 {
1615 m &= 0x1fffffffffffffU;
1616 m >>= (-1021 - e);
1617 e = -1022;
1618 }
1619
1620 r |= ((uint64_t)(e + 1022)) << 52;
1621 r |= m & 0xfffffffffffffU;
1622 #endif
1623
1624 return r;
1625 }
1626
1627 /* converts an ieee double/binary64 to a double */
1628 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1629 ecb_function_ ecb_const double
1630 ecb_binary64_to_double (uint64_t x)
1631 {
1632 double r;
1633
1634 #if ECB_STDFP
1635 memcpy (&r, &x, 8);
1636 #else
1637 /* emulation, only works for normals and subnormals and +0 */
1638 int neg = x >> 63;
1639 int e = (x >> 52) & 0x7ffU;
1640
1641 x &= 0xfffffffffffffU;
1642
1643 if (e)
1644 x |= 0x10000000000000U;
1645 else
1646 e = 1;
1647
1648 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1649 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1650
1651 r = neg ? -r : r;
1652 #endif
1653
1654 return r;
1655 }
1656
1657 /* convert a float to ieee half/binary16 */
1658 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1659 ecb_function_ ecb_const uint16_t
1660 ecb_float_to_binary16 (float x)
1661 {
1662 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1663 }
1664
1665 /* convert an ieee half/binary16 to float */
1666 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1667 ecb_function_ ecb_const float
1668 ecb_binary16_to_float (uint16_t x)
1669 {
1670 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1671 }
1672
1673#endif
1674
1675#endif
1676

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines