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

Comparing libecb/ecb.h (file contents):
Revision 1.24 by root, Thu May 26 19:49:21 2011 UTC vs.
Revision 1.207 by root, Fri Mar 25 14:34:01 2022 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/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
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
34 114
35/* many compilers define _GNUC_ to some versions but then only implement 115/* many compilers define _GNUC_ to some versions but then only implement
36 * what their idiot authors think are the "more important" extensions, 116 * what their idiot authors think are the "more important" extensions,
37 * causing enourmous grief in return for some better fake benchmark numbers. 117 * causing enormous grief in return for some better fake benchmark numbers.
38 * or so. 118 * or so.
39 * we try to detect these and simply assume they are not gcc - if they have 119 * we try to detect these and simply assume they are not gcc - if they have
40 * an issue with that they should have done it right in the first place. 120 * an issue with that they should have done it right in the first place.
41 */ 121 */
42#ifndef ECB_GCC_VERSION 122#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
43# if defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__llvm__)
44# define ECB_GCC_VERSION(major,minor) 0 123 #define ECB_GCC_VERSION(major,minor) 0
45# else 124#else
46# 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)))
47# endif 126#endif
48#endif
49 127
50#ifndef __cplusplus 128#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
51# if __STDC_VERSION__ >= 199901L 129
52# define ECB_INLINE inline 130#if __clang__ && defined __has_builtin
53typedef _Bool ecb_bool; 131 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
54# else 132#else
55# 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
56typedef int ecb_bool; 352typedef int ecb_bool;
57# endif
58#endif
59 353
60#define ECB_CONCAT_(a, b) a ## b 354#define ECB_CONCAT_(a, b) a ## b
61#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
62#define ECB_STRINGIFY_(a) # a 356#define ECB_STRINGIFY_(a) # a
63#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))
64 359
65#define ECB_HEADER_INLINE static ECB_INLINE 360#define ecb_function_ ecb_inline
66 361
67#if ECB_GCC_VERSION(3,1) 362#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
68# 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)
69# define ecb_is_constant(expr) __builtin_constant_p (expr) 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)
70# 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)
71# define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 385 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
72#else 386#else
73# define ecb_attribute(attrlist)
74# define ecb_is_constant(expr) 0
75# define ecb_expect(expr,value) (expr)
76# define ecb_prefetch(addr,rw,locality) 387 #define ecb_prefetch(addr,rw,locality)
77#endif 388#endif
78 389
79/* 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))
80#if ECB_GCC_VERSION(4,5) 407#elif ECB_GCC_VERSION(4,5)
81# define ecb_decltype(x) __decltype(x) 408 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
82#elif ECB_GCC_VERSION(3,0) 409#else
83# define ecb_decltype(x) typeof(x) 410 #define ecb_deprecated_message(msg) ecb_deprecated
84#endif 411#endif
85 412
86#define ecb_artificial ecb_attribute ((__artificial__)) 413#if _MSC_VER >= 1400
414 #define ecb_noinline __declspec (noinline)
415#else
87#define ecb_noinline ecb_attribute ((__noinline__)) 416 #define ecb_noinline ecb_attribute ((__noinline__))
88#define ecb_noreturn ecb_attribute ((__noreturn__)) 417#endif
418
89#define ecb_unused ecb_attribute ((__unused__)) 419#define ecb_unused ecb_attribute ((__unused__))
90#define ecb_const ecb_attribute ((__const__)) 420#define ecb_const ecb_attribute ((__const__))
91#define ecb_pure ecb_attribute ((__pure__)) 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__))
92#define ecb_hot ecb_attribute ((__hot__)) /* 4.3 */ 437 #define ecb_hot ecb_attribute ((__hot__))
93#define ecb_cold ecb_attribute ((__cold__)) /* 4.3 */ 438 #define ecb_cold ecb_attribute ((__cold__))
439#else
440 #define ecb_artificial
441 #define ecb_hot
442 #define ecb_cold
443#endif
94 444
95/* put into if's if you are very sure that the expression */ 445/* put around conditional expressions if you are very sure that the */
96/* is mostly true or mosty false. note that these return */ 446/* expression is mostly true or mostly false. note that these return */
97/* booleans, not the expression. */ 447/* booleans, not the expression. */
98#define ecb_unlikely(expr) ecb_expect (!!(expr), 0) 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 */
99#define ecb_likely(expr) ecb_expect (!!(expr), 1) 451#define ecb_likely(expr) ecb_expect_true (expr)
452#define ecb_unlikely(expr) ecb_expect_false (expr)
453
454/* count trailing zero bits and count # of one bits */
455#if ECB_GCC_VERSION(3,4) \
456 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
457 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
458 && ECB_CLANG_BUILTIN(__builtin_popcount))
459 #define ecb_ctz32(x) __builtin_ctz (x)
460 #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x))
461 #define ecb_clz32(x) __builtin_clz (x)
462 #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x))
463 #define ecb_ld32(x) (ecb_clz32 (x) ^ 31)
464 #define ecb_ld64(x) (ecb_clz64 (x) ^ 63)
465 #define ecb_popcount32(x) __builtin_popcount (x)
466 /* ecb_popcount64 is more difficult, see below */
467#else
468 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
469 ecb_function_ ecb_const int
470 ecb_ctz32 (uint32_t x)
471 {
472#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
473 unsigned long r;
474 _BitScanForward (&r, x);
475 return (int)r;
476#else
477 int r = 0;
478
479 /* todo: use david seal's algorithm */
480
481 x &= ~x + 1; /* this isolates the lowest bit */
482
483#if ECB_branchless_on_i386
484 r += !!(x & 0xaaaaaaaa) << 0;
485 r += !!(x & 0xcccccccc) << 1;
486 r += !!(x & 0xf0f0f0f0) << 2;
487 r += !!(x & 0xff00ff00) << 3;
488 r += !!(x & 0xffff0000) << 4;
489#else
490 if (x & 0xaaaaaaaa) r += 1;
491 if (x & 0xcccccccc) r += 2;
492 if (x & 0xf0f0f0f0) r += 4;
493 if (x & 0xff00ff00) r += 8;
494 if (x & 0xffff0000) r += 16;
495#endif
496
497 return r;
498#endif
499 }
500
501 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
502 ecb_function_ ecb_const int
503 ecb_ctz64 (uint64_t x)
504 {
505#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
506 unsigned long r;
507 _BitScanForward64 (&r, x);
508 return (int)r;
509#else
510 int shift = x & 0xffffffff ? 0 : 32;
511 return ecb_ctz32 (x >> shift) + shift;
512#endif
513 }
514
515 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
516 ecb_function_ ecb_const int
517 ecb_popcount32 (uint32_t x)
518 {
519 x -= (x >> 1) & 0x55555555;
520 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
521 x = ((x >> 4) + x) & 0x0f0f0f0f;
522 x *= 0x01010101;
523
524 return x >> 24;
525 }
526
527 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
528 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
529 {
530#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
531 unsigned long r;
532 _BitScanReverse (&r, x);
533 return (int)r;
534#else
535 int r = 0;
536
537 if (x >> 16) { x >>= 16; r += 16; }
538 if (x >> 8) { x >>= 8; r += 8; }
539 if (x >> 4) { x >>= 4; r += 4; }
540 if (x >> 2) { x >>= 2; r += 2; }
541 if (x >> 1) { r += 1; }
542
543 return r;
544#endif
545 }
546
547 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
548 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
549 {
550#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
551 unsigned long r;
552 _BitScanReverse64 (&r, x);
553 return (int)r;
554#else
555 int r = 0;
556
557 if (x >> 32) { x >>= 32; r += 32; }
558
559 return r + ecb_ld32 (x);
560#endif
561 }
562#endif
563
564ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
565ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
566ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
567ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
568
569ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
570ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
571{
572 return ( (x * 0x0802U & 0x22110U)
573 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
574}
575
576ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
577ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
578{
579 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
580 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
581 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
582 x = ( x >> 8 ) | ( x << 8);
583
584 return x;
585}
586
587ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
588ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
589{
590 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
591 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
592 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
593 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
594 x = ( x >> 16 ) | ( x << 16);
595
596 return x;
597}
598
599ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
600ecb_function_ ecb_const int
601ecb_popcount64 (uint64_t x)
602{
603 /* popcount64 is only available on 64 bit cpus as gcc builtin. */
604 /* also, gcc/clang make this surprisingly difficult to use */
605#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
606 return __builtin_popcountl (x);
607#else
608 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
609#endif
610}
611
612ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
613ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
614ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
615ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
616ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
617ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
618ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
619ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
620
621ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
622ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
623ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
624ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
625ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
626ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
627ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
628ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
629
630#if ECB_CPP
631
632inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
633inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
634inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
635inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
636
637inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
638inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
639inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
640inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
641
642inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
643inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
644inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
645inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
646
647inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
648inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
649inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
650inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
651
652inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
653inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
654inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
655
656inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
657inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
658inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
659inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
660
661inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
662inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
663inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
664inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
665
666#endif
667
668#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
669 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
670 #define ecb_bswap16(x) __builtin_bswap16 (x)
671 #else
672 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
673 #endif
674 #define ecb_bswap32(x) __builtin_bswap32 (x)
675 #define ecb_bswap64(x) __builtin_bswap64 (x)
676#elif _MSC_VER
677 #include <stdlib.h>
678 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
679 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
680 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
681#else
682 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
683 ecb_function_ ecb_const uint16_t
684 ecb_bswap16 (uint16_t x)
685 {
686 return ecb_rotl16 (x, 8);
687 }
688
689 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
690 ecb_function_ ecb_const uint32_t
691 ecb_bswap32 (uint32_t x)
692 {
693 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
694 }
695
696 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
697 ecb_function_ ecb_const uint64_t
698 ecb_bswap64 (uint64_t x)
699 {
700 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
701 }
702#endif
703
704#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
705 #define ecb_unreachable() __builtin_unreachable ()
706#else
707 /* this seems to work fine, but gcc always emits a warning for it :/ */
708 ecb_inline ecb_noreturn void ecb_unreachable (void);
709 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
710#endif
100 711
101/* try to tell the compiler that some condition is definitely true */ 712/* try to tell the compiler that some condition is definitely true */
102#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 713#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
103 714
104/* count trailing zero bits and count # of one bits */ 715ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
105#if ECB_GCC_VERSION(3,4) 716ecb_inline ecb_const uint32_t
106#define ecb_ctz32 (x) __builtin_ctz (x)
107#define ecb_popcount32 (x) __builtin_popcount (x)
108#else
109ECB_HEADER_INLINE int ecb_ctz32 (uint32_t x) ecb_const;
110ECB_HEADER_INLINE
111ecb_ctz32 (uint32_t x)
112{
113 int r = 0;
114
115 x &= -x; /* this isolates the lowest bit */
116
117 if (x & 0xaaaaaaaa) r += 1;
118 if (x & 0xcccccccc) r += 2;
119 if (x & 0xf0f0f0f0) r += 4;
120 if (x & 0xff00ff00) r += 8;
121 if (x & 0xffff0000) r += 16;
122
123 return r;
124}
125
126ECB_HEADER_INLINE int ecb_popcount32 (uint32_t x) ecb_const;
127ECB_HEADER_INLINE
128ecb_popcount32 (uint32_t x)
129{
130 x -= (x >> 1) & 0x55555555;
131 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
132 x = ((x >> 4) + x) & 0x0f0f0f0f;
133 x *= 0x01010101;
134
135 return x >> 24;
136}
137#endif
138
139#if ECB_GCC_VERSION(4,3)
140# define ecb_bswap32(x) __builtin_bswap32 (x)
141# define ecb_bswap16(x) (__builtin_bswap32(x) >> 16)
142#else
143ECB_HEADER_INLINE uint32_t ecb_bswap32 (uint32_t x) ecb_const;
144ECB_HEADER_INLINE uint32_t
145ecb_bswap32 (uint32_t x)
146{
147 return (x >> 24)
148 | ((x >> 8) & 0x0000ff00)
149 | ((x << 8) & 0x00ff0000)
150 | (x << 24);
151}
152
153ECB_HEADER_INLINE uint32_t ecb_bswap16 (uint32_t x) ecb_const;
154ECB_HEADER_INLINE uint32_t
155ecb_bswap16 (uint32_t x)
156{
157 return ((x >> 8) & 0xff)
158 | ((x << 8) & 0x00ff0000)
159 | (x << 24);
160}
161#endif
162
163#if ECB_GCC_VERSION(4,5)
164# define ecb_unreachable() __builtin_unreachable ()
165#else
166/* this seems to work fine, but gcc always emits a warning for it :/ */
167ECB_HEADER_INLINE void ecb_unreachable (void) ecb_noreturn;
168ECB_HEADER_INLINE void ecb_unreachable (void) { }
169#endif
170
171ECB_HEADER_INLINE unsigned char ecb_byteorder_helper (void) ecb_const;
172ECB_HEADER_INLINE unsigned char
173ecb_byteorder_helper (void) 717ecb_byteorder_helper (void)
174{ 718{
175 const uint32_t u = 0x11223344; 719 /* the union code still generates code under pressure in gcc, */
176 return *(unsigned char *)&u; 720 /* but less than using pointers, and always seems to */
721 /* successfully return a constant. */
722 /* the reason why we have this horrible preprocessor mess */
723 /* is to avoid it in all cases, at least on common architectures */
724 /* or when using a recent enough gcc version (>= 4.6) */
725#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
726 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
727 #define ECB_LITTLE_ENDIAN 1
728 return 0x44332211;
729#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
730 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
731 #define ECB_BIG_ENDIAN 1
732 return 0x11223344;
733#else
734 union
735 {
736 uint8_t c[4];
737 uint32_t u;
738 } u = { 0x11, 0x22, 0x33, 0x44 };
739 return u.u;
740#endif
177} 741}
178 742
179ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) ecb_const; 743ecb_inline ecb_const ecb_bool ecb_big_endian (void);
180ECB_HEADER_INLINE ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }; 744ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
181ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) ecb_const; 745ecb_inline ecb_const ecb_bool ecb_little_endian (void);
182ECB_HEADER_INLINE ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }; 746ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
183 747
748/*****************************************************************************/
749/* unaligned load/store */
750
751ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
752ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
753ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
754
755ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
756ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
757ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
758
759ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
760ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
761ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
762
763ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
764ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
765ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
766
767ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
768ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
769ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
770
771ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
772ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
773ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
774
775ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
776ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
777ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
778
779ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
780ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
781ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
782
783ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
784ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
785ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
786
787ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
788ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
789ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
790
791#if ECB_CPP
792
793inline uint8_t ecb_bswap (uint8_t v) { return v; }
794inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
795inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
796inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
797
798template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
799template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
800template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
801template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
802template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
803template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
804template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
805template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
806
807template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
808template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
809template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
810template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
811template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
812template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
813template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
814template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
815
816#endif
817
818/*****************************************************************************/
819/* pointer/integer hashing */
820
821/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
822ecb_function_ uint32_t ecb_mix32 (uint32_t v);
823ecb_function_ uint32_t ecb_mix32 (uint32_t v)
824{
825 v ^= v >> 16; v *= 0x7feb352dU;
826 v ^= v >> 15; v *= 0x846ca68bU;
827 v ^= v >> 16;
828 return v;
829}
830
831ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
832ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
833{
834 v ^= v >> 16 ; v *= 0x43021123U;
835 v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
836 v ^= v >> 16 ;
837 return v;
838}
839
840/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
841ecb_function_ uint64_t ecb_mix64 (uint64_t v);
842ecb_function_ uint64_t ecb_mix64 (uint64_t v)
843{
844 v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
845 v ^= v >> 27; v *= 0x94d049bb133111ebU;
846 v ^= v >> 31;
847 return v;
848}
849
850ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
851ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
852{
853 v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
854 v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
855 v ^= v >> 30 ^ v >> 60;
856 return v;
857}
858
859ecb_function_ uintptr_t ecb_ptrmix (void *p);
860ecb_function_ uintptr_t ecb_ptrmix (void *p)
861{
862 #if ECB_PTRSIZE <= 4
863 return ecb_mix32 ((uint32_t)p);
864 #else
865 return ecb_mix64 ((uint64_t)p);
866 #endif
867}
868
869ecb_function_ void *ecb_ptrunmix (uintptr_t v);
870ecb_function_ void *ecb_ptrunmix (uintptr_t v)
871{
872 #if ECB_PTRSIZE <= 4
873 return (void *)ecb_unmix32 (v);
874 #else
875 return (void *)ecb_unmix64 (v);
876 #endif
877}
878
879#if ECB_CPP
880
881template<typename T>
882inline uintptr_t ecb_ptrmix (T *p)
883{
884 return ecb_ptrmix (static_cast<void *>(p));
885}
886
887template<typename T>
888inline T *ecb_ptrunmix (uintptr_t v)
889{
890 return static_cast<T *>(ecb_ptrunmix (v));
891}
892
893#endif
894
895/*****************************************************************************/
896/* gray code */
897
898ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
899ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
900ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
901ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
902
903ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
904{
905 g ^= g >> 1;
906 g ^= g >> 2;
907 g ^= g >> 4;
908
909 return g;
910}
911
912ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
913{
914 g ^= g >> 1;
915 g ^= g >> 2;
916 g ^= g >> 4;
917 g ^= g >> 8;
918
919 return g;
920}
921
922ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
923{
924 g ^= g >> 1;
925 g ^= g >> 2;
926 g ^= g >> 4;
927 g ^= g >> 8;
928 g ^= g >> 16;
929
930 return g;
931}
932
933ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
934{
935 g ^= g >> 1;
936 g ^= g >> 2;
937 g ^= g >> 4;
938 g ^= g >> 8;
939 g ^= g >> 16;
940 g ^= g >> 32;
941
942 return g;
943}
944
945#if ECB_CPP
946
947ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
948ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
949ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
950ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
951
952ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
953ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
954ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
955ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
956
957#endif
958
959/*****************************************************************************/
960/* 2d hilbert curves */
961
962/* algorithm from the book Hacker's Delight, modified to not */
963/* run into undefined behaviour for n==16 */
964static uint32_t
965ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
966{
967 uint32_t comp, swap, cs, t, sr;
968
969 /* pad s on the left (unused) bits with 01 (no change groups) */
970 s |= 0x55555555U << n << n;
971 /* "s shift right" */
972 sr = (s >> 1) & 0x55555555U;
973 /* compute complement and swap info in two-bit groups */
974 cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
975
976 /* parallel prefix xor op to propagate both complement
977 * and swap info together from left to right (there is
978 * no step "cs ^= cs >> 1", so in effect it computes
979 * two independent parallel prefix operations on two
980 * interleaved sets of sixteen bits).
981 */
982 cs ^= cs >> 2;
983 cs ^= cs >> 4;
984 cs ^= cs >> 8;
985 cs ^= cs >> 16;
986
987 /* separate swap and complement bits */
988 swap = cs & 0x55555555U;
989 comp = (cs >> 1) & 0x55555555U;
990
991 /* calculate coordinates in odd and even bit positions */
992 t = (s & swap) ^ comp;
993 s = s ^ sr ^ t ^ (t << 1);
994
995 /* unpad/clear out any junk on the left */
996 s = s & ((1 << n << n) - 1);
997
998 /* Now "unshuffle" to separate the x and y bits. */
999 t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
1000 t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
1001 t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
1002 t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
1003
1004 /* now s contains two 16-bit coordinates */
1005 return s;
1006}
1007
1008/* 64 bit, a straightforward extension to the 32 bit case */
1009static uint64_t
1010ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1011{
1012 uint64_t comp, swap, cs, t, sr;
1013
1014 /* pad s on the left (unused) bits with 01 (no change groups) */
1015 s |= 0x5555555555555555U << n << n;
1016 /* "s shift right" */
1017 sr = (s >> 1) & 0x5555555555555555U;
1018 /* compute complement and swap info in two-bit groups */
1019 cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1020
1021 /* parallel prefix xor op to propagate both complement
1022 * and swap info together from left to right (there is
1023 * no step "cs ^= cs >> 1", so in effect it computes
1024 * two independent parallel prefix operations on two
1025 * interleaved sets of thirty-two bits).
1026 */
1027 cs ^= cs >> 2;
1028 cs ^= cs >> 4;
1029 cs ^= cs >> 8;
1030 cs ^= cs >> 16;
1031 cs ^= cs >> 32;
1032
1033 /* separate swap and complement bits */
1034 swap = cs & 0x5555555555555555U;
1035 comp = (cs >> 1) & 0x5555555555555555U;
1036
1037 /* calculate coordinates in odd and even bit positions */
1038 t = (s & swap) ^ comp;
1039 s = s ^ sr ^ t ^ (t << 1);
1040
1041 /* unpad/clear out any junk on the left */
1042 s = s & ((1 << n << n) - 1);
1043
1044 /* Now "unshuffle" to separate the x and y bits. */
1045 t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1046 t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1047 t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1048 t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1049 t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1050
1051 /* now s contains two 32-bit coordinates */
1052 return s;
1053}
1054
1055/* algorithm from the book Hacker's Delight, but a similar algorithm*/
1056/* is given in https://doi.org/10.1002/spe.4380160103 */
1057/* this has been slightly improved over the original version */
1058ecb_function_ uint32_t
1059ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1060{
1061 uint32_t row;
1062 uint32_t state = 0;
1063 uint32_t s = 0;
1064
1065 do
1066 {
1067 --n;
1068
1069 row = 4 * state
1070 | (2 & (xy >> n >> 15))
1071 | (1 & (xy >> n ));
1072
1073 /* these funky constants are lookup tables for two-bit values */
1074 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1075 state = (0x8fe65831U >> 2 * row) & 3;
1076 }
1077 while (n > 0);
1078
1079 return s;
1080}
1081
1082/* 64 bit, essentially the same as 32 bit */
1083ecb_function_ uint64_t
1084ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1085{
1086 uint32_t row;
1087 uint32_t state = 0;
1088 uint64_t s = 0;
1089
1090 do
1091 {
1092 --n;
1093
1094 row = 4 * state
1095 | (2 & (xy >> n >> 31))
1096 | (1 & (xy >> n ));
1097
1098 /* these funky constants are lookup tables for two-bit values */
1099 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1100 state = (0x8fe65831U >> 2 * row) & 3;
1101 }
1102 while (n > 0);
1103
1104 return s;
1105}
1106
1107/*****************************************************************************/
1108/* division */
1109
1110#if ECB_GCC_VERSION(3,0) || ECB_C99
1111 /* C99 tightened the definition of %, so we can use a more efficient version */
1112 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
1113#else
184#define ecb_mod(m, n) ((m) < 0 ? ((n) - (-(m) % (n))) : ((m) % (n))) 1114 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
1115#endif
1116
1117#if ECB_CPP
1118 template<typename T>
1119 static inline T ecb_div_rd (T val, T div)
1120 {
1121 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
1122 }
1123 template<typename T>
1124 static inline T ecb_div_ru (T val, T div)
1125 {
1126 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
1127 }
1128#else
1129 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
1130 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
1131#endif
1132
1133/*****************************************************************************/
1134/* array length */
185 1135
186#if ecb_cplusplus_does_not_suck 1136#if ecb_cplusplus_does_not_suck
187// does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) 1137 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
188template<typename T, int N> 1138 template<typename T, int N>
189static inline int ecb_array_length (const T (&arr)[N]) 1139 static inline int ecb_array_length (const T (&arr)[N])
190{ 1140 {
191 return N; 1141 return N;
192} 1142 }
193#else 1143#else
194#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 1144 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
1145#endif
1146
1147/*****************************************************************************/
1148/* IEEE 754-2008 half float conversions */
1149
1150ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
1151ecb_function_ ecb_const uint32_t
1152ecb_binary16_to_binary32 (uint32_t x)
1153{
1154 unsigned int s = (x & 0x8000) << (31 - 15);
1155 int e = (x >> 10) & 0x001f;
1156 unsigned int m = x & 0x03ff;
1157
1158 if (ecb_expect_false (e == 31))
1159 /* infinity or NaN */
1160 e = 255 - (127 - 15);
1161 else if (ecb_expect_false (!e))
1162 {
1163 if (ecb_expect_true (!m))
1164 /* zero, handled by code below by forcing e to 0 */
1165 e = 0 - (127 - 15);
1166 else
1167 {
1168 /* subnormal, renormalise */
1169 unsigned int s = 10 - ecb_ld32 (m);
1170
1171 m = (m << s) & 0x3ff; /* mask implicit bit */
1172 e -= s - 1;
1173 }
1174 }
1175
1176 /* e and m now are normalised, or zero, (or inf or nan) */
1177 e += 127 - 15;
1178
1179 return s | (e << 23) | (m << (23 - 10));
1180}
1181
1182ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1183ecb_function_ ecb_const uint16_t
1184ecb_binary32_to_binary16 (uint32_t x)
1185{
1186 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1187 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1188 unsigned int m = x & 0x007fffff;
1189
1190 x &= 0x7fffffff;
1191
1192 /* if it's within range of binary16 normals, use fast path */
1193 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1194 {
1195 /* mantissa round-to-even */
1196 m += 0x00000fff + ((m >> (23 - 10)) & 1);
1197
1198 /* handle overflow */
1199 if (ecb_expect_false (m >= 0x00800000))
1200 {
1201 m >>= 1;
1202 e += 1;
1203 }
1204
1205 return s | (e << 10) | (m >> (23 - 10));
1206 }
1207
1208 /* handle large numbers and infinity */
1209 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1210 return s | 0x7c00;
1211
1212 /* handle zero, subnormals and small numbers */
1213 if (ecb_expect_true (x < 0x38800000))
1214 {
1215 /* zero */
1216 if (ecb_expect_true (!x))
1217 return s;
1218
1219 /* handle subnormals */
1220
1221 /* too small, will be zero */
1222 if (e < (14 - 24)) /* might not be sharp, but is good enough */
1223 return s;
1224
1225 m |= 0x00800000; /* make implicit bit explicit */
1226
1227 /* very tricky - we need to round to the nearest e (+10) bit value */
1228 {
1229 unsigned int bits = 14 - e;
1230 unsigned int half = (1 << (bits - 1)) - 1;
1231 unsigned int even = (m >> bits) & 1;
1232
1233 /* if this overflows, we will end up with a normalised number */
1234 m = (m + half + even) >> bits;
1235 }
1236
1237 return s | m;
1238 }
1239
1240 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1241 m >>= 13;
1242
1243 return s | 0x7c00 | m | !m;
1244}
1245
1246/*******************************************************************************/
1247/* fast integer to ascii */
1248
1249/*
1250 * This code is pretty complicated because it is general. The idea behind it,
1251 * however, is pretty simple: first, the number is multiplied with a scaling
1252 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1253 * number with the first digit in the upper bits.
1254 * Then this digit is converted to text and masked out. The resulting number
1255 * is then multiplied by 10, by multiplying the fixed point representation
1256 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1257 * format becomes 5.27, 6.26 and so on.
1258 * The rest involves only advancing the pointer if we already generated a
1259 * non-zero digit, so leading zeroes are overwritten.
1260 */
1261
1262/* simply return a mask with "bits" bits set */
1263#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1264
1265/* oputput a single digit. maskvalue is 10**digitidx */
1266#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1267 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1268 { \
1269 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1270 *ptr = digit + '0'; /* output it */ \
1271 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1272 ptr += nz; /* output digit only if non-zero digit seen */ \
1273 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1274 }
1275
1276/* convert integer to fixed point format and multiply out digits, highest first */
1277/* requires magic constants: max. digits and number of bits after the decimal point */
1278#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1279ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1280{ \
1281 char nz = lz; /* non-zero digit seen? */ \
1282 /* convert to x.bits fixed-point */ \
1283 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1284 /* output up to 10 digits */ \
1285 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1286 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1287 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1288 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1289 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1290 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1291 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1292 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1293 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1294 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1295 return ptr; \
1296}
1297
1298/* predefined versions of the above, for various digits */
1299/* ecb_i2a_xN = almost N digits, limit defined by macro */
1300/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1301/* ecb_i2a_0N = exactly N digits, including leading zeroes */
1302
1303/* non-leading-zero versions, limited range */
1304#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1305#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1306ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1307ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1308
1309/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1310ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1311ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1312ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1313ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1314ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1315ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1316ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1317ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1318
1319/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1320ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1321ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1322ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1323ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1324ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1325ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1326ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1327ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1328
1329#define ECB_I2A_I32_DIGITS 11
1330#define ECB_I2A_U32_DIGITS 10
1331#define ECB_I2A_I64_DIGITS 20
1332#define ECB_I2A_U64_DIGITS 21
1333#define ECB_I2A_MAX_DIGITS 21
1334
1335ecb_inline char *
1336ecb_i2a_u32 (char *ptr, uint32_t u)
1337{
1338 #if ECB_64BIT_NATIVE
1339 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1340 ptr = ecb_i2a_x10 (ptr, u);
1341 else /* x10 almost, but not fully, covers 32 bit */
1342 {
1343 uint32_t u1 = u % 1000000000;
1344 uint32_t u2 = u / 1000000000;
1345
1346 *ptr++ = u2 + '0';
1347 ptr = ecb_i2a_09 (ptr, u1);
1348 }
1349 #else
1350 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1351 ecb_i2a_x5 (ptr, u);
1352 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1353 {
1354 uint32_t u1 = u % 10000;
1355 uint32_t u2 = u / 10000;
1356
1357 ptr = ecb_i2a_x5 (ptr, u2);
1358 ptr = ecb_i2a_04 (ptr, u1);
1359 }
1360 else
1361 {
1362 uint32_t u1 = u % 10000;
1363 uint32_t ua = u / 10000;
1364 uint32_t u2 = ua % 10000;
1365 uint32_t u3 = ua / 10000;
1366
1367 ptr = ecb_i2a_2 (ptr, u3);
1368 ptr = ecb_i2a_04 (ptr, u2);
1369 ptr = ecb_i2a_04 (ptr, u1);
1370 }
195#endif 1371 #endif
196 1372
197ECB_INLINE uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 1373 return ptr;
198ECB_INLINE uint32_t
199ecb_rotr32 (uint32_t x, unsigned int count)
200{
201 return (x << (32 - count)) | (x >> count);
202} 1374}
203 1375
204ECB_INLINE uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 1376ecb_inline char *
205ECB_INLINE uint32_t 1377ecb_i2a_i32 (char *ptr, int32_t v)
206ecb_rotl32 (uint32_t x, unsigned int count)
207{ 1378{
208 return (x >> (32 - count)) | (x << count); 1379 *ptr = '-'; ptr += v < 0;
209} 1380 uint32_t u = v < 0 ? -(uint32_t)v : v;
210 1381
1382 #if ECB_64BIT_NATIVE
1383 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1384 #else
1385 ptr = ecb_i2a_u32 (ptr, u);
211#endif 1386 #endif
212 1387
1388 return ptr;
1389}
1390
1391ecb_inline char *
1392ecb_i2a_u64 (char *ptr, uint64_t u)
1393{
1394 #if ECB_64BIT_NATIVE
1395 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1396 ptr = ecb_i2a_x10 (ptr, u);
1397 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1398 {
1399 uint64_t u1 = u % 1000000000;
1400 uint64_t u2 = u / 1000000000;
1401
1402 ptr = ecb_i2a_x10 (ptr, u2);
1403 ptr = ecb_i2a_09 (ptr, u1);
1404 }
1405 else
1406 {
1407 uint64_t u1 = u % 1000000000;
1408 uint64_t ua = u / 1000000000;
1409 uint64_t u2 = ua % 1000000000;
1410 uint64_t u3 = ua / 1000000000;
1411
1412 ptr = ecb_i2a_2 (ptr, u3);
1413 ptr = ecb_i2a_09 (ptr, u2);
1414 ptr = ecb_i2a_09 (ptr, u1);
1415 }
1416 #else
1417 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1418 ptr = ecb_i2a_x5 (ptr, u);
1419 else
1420 {
1421 uint64_t u1 = u % 10000;
1422 uint64_t u2 = u / 10000;
1423
1424 ptr = ecb_i2a_u64 (ptr, u2);
1425 ptr = ecb_i2a_04 (ptr, u1);
1426 }
1427 #endif
1428
1429 return ptr;
1430}
1431
1432ecb_inline char *
1433ecb_i2a_i64 (char *ptr, int64_t v)
1434{
1435 *ptr = '-'; ptr += v < 0;
1436 uint64_t u = v < 0 ? -(uint64_t)v : v;
1437
1438 #if ECB_64BIT_NATIVE
1439 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1440 ptr = ecb_i2a_x10 (ptr, u);
1441 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1442 {
1443 uint64_t u1 = u % 1000000000;
1444 uint64_t u2 = u / 1000000000;
1445
1446 ptr = ecb_i2a_x10 (ptr, u2);
1447 ptr = ecb_i2a_09 (ptr, u1);
1448 }
1449 else
1450 {
1451 uint64_t u1 = u % 1000000000;
1452 uint64_t ua = u / 1000000000;
1453 uint64_t u2 = ua % 1000000000;
1454 uint64_t u3 = ua / 1000000000;
1455
1456 /* 2**31 is 19 digits, so the top is exactly one digit */
1457 *ptr++ = u3 + '0';
1458 ptr = ecb_i2a_09 (ptr, u2);
1459 ptr = ecb_i2a_09 (ptr, u1);
1460 }
1461 #else
1462 ptr = ecb_i2a_u64 (ptr, u);
1463 #endif
1464
1465 return ptr;
1466}
1467
1468/*******************************************************************************/
1469/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
1470
1471/* basically, everything uses "ieee pure-endian" floating point numbers */
1472/* the only noteworthy exception is ancient armle, which uses order 43218765 */
1473#if 0 \
1474 || __i386 || __i386__ \
1475 || ECB_GCC_AMD64 \
1476 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
1477 || defined __s390__ || defined __s390x__ \
1478 || defined __mips__ \
1479 || defined __alpha__ \
1480 || defined __hppa__ \
1481 || defined __ia64__ \
1482 || defined __m68k__ \
1483 || defined __m88k__ \
1484 || defined __sh__ \
1485 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1486 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1487 || defined __aarch64__
1488 #define ECB_STDFP 1
1489#else
1490 #define ECB_STDFP 0
1491#endif
1492
1493#ifndef ECB_NO_LIBM
1494
1495 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
1496
1497 /* only the oldest of old doesn't have this one. solaris. */
1498 #ifdef INFINITY
1499 #define ECB_INFINITY INFINITY
1500 #else
1501 #define ECB_INFINITY HUGE_VAL
1502 #endif
1503
1504 #ifdef NAN
1505 #define ECB_NAN NAN
1506 #else
1507 #define ECB_NAN ECB_INFINITY
1508 #endif
1509
1510 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
1511 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1512 #define ecb_frexpf(x,e) frexpf ((x), (e))
1513 #else
1514 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1515 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1516 #endif
1517
1518 /* convert a float to ieee single/binary32 */
1519 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1520 ecb_function_ ecb_const uint32_t
1521 ecb_float_to_binary32 (float x)
1522 {
1523 uint32_t r;
1524
1525 #if ECB_STDFP
1526 memcpy (&r, &x, 4);
1527 #else
1528 /* slow emulation, works for anything but -0 */
1529 uint32_t m;
1530 int e;
1531
1532 if (x == 0e0f ) return 0x00000000U;
1533 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1534 if (x < -3.40282346638528860e+38f) return 0xff800000U;
1535 if (x != x ) return 0x7fbfffffU;
1536
1537 m = ecb_frexpf (x, &e) * 0x1000000U;
1538
1539 r = m & 0x80000000U;
1540
1541 if (r)
1542 m = -m;
1543
1544 if (e <= -126)
1545 {
1546 m &= 0xffffffU;
1547 m >>= (-125 - e);
1548 e = -126;
1549 }
1550
1551 r |= (e + 126) << 23;
1552 r |= m & 0x7fffffU;
1553 #endif
1554
1555 return r;
1556 }
1557
1558 /* converts an ieee single/binary32 to a float */
1559 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1560 ecb_function_ ecb_const float
1561 ecb_binary32_to_float (uint32_t x)
1562 {
1563 float r;
1564
1565 #if ECB_STDFP
1566 memcpy (&r, &x, 4);
1567 #else
1568 /* emulation, only works for normals and subnormals and +0 */
1569 int neg = x >> 31;
1570 int e = (x >> 23) & 0xffU;
1571
1572 x &= 0x7fffffU;
1573
1574 if (e)
1575 x |= 0x800000U;
1576 else
1577 e = 1;
1578
1579 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1580 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1581
1582 r = neg ? -r : r;
1583 #endif
1584
1585 return r;
1586 }
1587
1588 /* convert a double to ieee double/binary64 */
1589 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1590 ecb_function_ ecb_const uint64_t
1591 ecb_double_to_binary64 (double x)
1592 {
1593 uint64_t r;
1594
1595 #if ECB_STDFP
1596 memcpy (&r, &x, 8);
1597 #else
1598 /* slow emulation, works for anything but -0 */
1599 uint64_t m;
1600 int e;
1601
1602 if (x == 0e0 ) return 0x0000000000000000U;
1603 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1604 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1605 if (x != x ) return 0X7ff7ffffffffffffU;
1606
1607 m = frexp (x, &e) * 0x20000000000000U;
1608
1609 r = m & 0x8000000000000000;;
1610
1611 if (r)
1612 m = -m;
1613
1614 if (e <= -1022)
1615 {
1616 m &= 0x1fffffffffffffU;
1617 m >>= (-1021 - e);
1618 e = -1022;
1619 }
1620
1621 r |= ((uint64_t)(e + 1022)) << 52;
1622 r |= m & 0xfffffffffffffU;
1623 #endif
1624
1625 return r;
1626 }
1627
1628 /* converts an ieee double/binary64 to a double */
1629 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1630 ecb_function_ ecb_const double
1631 ecb_binary64_to_double (uint64_t x)
1632 {
1633 double r;
1634
1635 #if ECB_STDFP
1636 memcpy (&r, &x, 8);
1637 #else
1638 /* emulation, only works for normals and subnormals and +0 */
1639 int neg = x >> 63;
1640 int e = (x >> 52) & 0x7ffU;
1641
1642 x &= 0xfffffffffffffU;
1643
1644 if (e)
1645 x |= 0x10000000000000U;
1646 else
1647 e = 1;
1648
1649 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1650 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1651
1652 r = neg ? -r : r;
1653 #endif
1654
1655 return r;
1656 }
1657
1658 /* convert a float to ieee half/binary16 */
1659 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1660 ecb_function_ ecb_const uint16_t
1661 ecb_float_to_binary16 (float x)
1662 {
1663 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1664 }
1665
1666 /* convert an ieee half/binary16 to float */
1667 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1668 ecb_function_ ecb_const float
1669 ecb_binary16_to_float (uint16_t x)
1670 {
1671 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1672 }
1673
1674#endif
1675
1676#endif
1677

Diff Legend

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