ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libecb/ecb.h

Comparing libecb/ecb.h (file contents):
Revision 1.90 by root, Tue May 29 14:09:49 2012 UTC vs.
Revision 1.213 by root, Fri Mar 25 15:34:12 2022 UTC

1	/*	1	/*
2	* libecb - http://software.schmorp.de/pkg/libecb	2	* libecb - http://software.schmorp.de/pkg/libecb
3	*	3	*
4	* Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de>	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
33	/* 16 bits major, 16 bits minor */	44	/* 16 bits major, 16 bits minor */
34	#define ECB_VERSION 0x00010001	45	#define ECB_VERSION 0x0001000c
35		46
36	#ifdef _WIN32	47	#include <string.h> /* for memcpy */
		48
		49	#if defined (_WIN32) && !defined (__MINGW32__)
37	typedef signed char int8_t;	50	typedef signed char int8_t;
38	typedef unsigned char uint8_t;	51	typedef unsigned char uint8_t;
		52	typedef signed char int_fast8_t;
		53	typedef unsigned char uint_fast8_t;
39	typedef signed short int16_t;	54	typedef signed short int16_t;
40	typedef unsigned short uint16_t;	55	typedef unsigned short uint16_t;
		56	typedef signed int int_fast16_t;
		57	typedef unsigned int uint_fast16_t;
41	typedef signed int int32_t;	58	typedef signed int int32_t;
42	typedef unsigned int uint32_t;	59	typedef unsigned int uint32_t;
		60	typedef signed int int_fast32_t;
		61	typedef unsigned int uint_fast32_t;
43	#if __GNUC__	62	#if __GNUC__
44	typedef signed long long int64_t;	63	typedef signed long long int64_t;
45	typedef unsigned long long uint64_t;	64	typedef unsigned long long uint64_t;
46	#else /* _MSC_VER || __BORLANDC__ */	65	#else /* _MSC_VER || __BORLANDC__ */
47	typedef signed __int64 int64_t;	66	typedef signed __int64 int64_t;
48	typedef unsigned __int64 uint64_t;	67	typedef unsigned __int64 uint64_t;
49	#endif	68	#endif
		69	typedef int64_t int_fast64_t;
		70	typedef uint64_t uint_fast64_t;
50	#ifdef _WIN64	71	#ifdef _WIN64
51	#define ECB_PTRSIZE 8	72	#define ECB_PTRSIZE 8
52	typedef uint64_t uintptr_t;	73	typedef uint64_t uintptr_t;
53	typedef int64_t intptr_t;	74	typedef int64_t intptr_t;
54	#else	75	#else
55	#define ECB_PTRSIZE 4	76	#define ECB_PTRSIZE 4
56	typedef uint32_t uintptr_t;	77	typedef uint32_t uintptr_t;
57	typedef int32_t intptr_t;	78	typedef int32_t intptr_t;
58	#endif	79	#endif
59	typedef intptr_t ptrdiff_t;
60	#else	80	#else
61	#include <inttypes.h>	81	#include <inttypes.h>
62	#if UINTMAX_MAX > 0xffffffffU	82	#if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
63	#define ECB_PTRSIZE 8	83	#define ECB_PTRSIZE 8
64	#else	84	#else
65	#define ECB_PTRSIZE 4	85	#define ECB_PTRSIZE 4
66	#endif	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
67	#endif	113	#endif
68		114
69	/* many compilers define _GNUC_ to some versions but then only implement	115	/* many compilers define _GNUC_ to some versions but then only implement
70	* what their idiot authors think are the "more important" extensions,	116	* what their idiot authors think are the "more important" extensions,
71	* causing enormous grief in return for some better fake benchmark numbers.	117	* causing enormous grief in return for some better fake benchmark numbers.
72	* or so.	118	* or so.
73	* 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
74	* 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.
75	*/	121	*/
76	#ifndef ECB_GCC_VERSION
77	#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__	122	#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
78	#define ECB_GCC_VERSION(major,minor) 0	123	#define ECB_GCC_VERSION(major,minor) 0
79	#else	124	#else
80	#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)))
81	#endif	126	#endif
82	#endif
83		127
		128	#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
		129
		130	#if __clang__ && defined __has_builtin
		131	#define ECB_CLANG_BUILTIN(x) __has_builtin (x)
		132	#else
		133	#define ECB_CLANG_BUILTIN(x) 0
		134	#endif
		135
		136	#if __clang__ && defined __has_extension
		137	#define ECB_CLANG_EXTENSION(x) __has_extension (x)
		138	#else
		139	#define ECB_CLANG_EXTENSION(x) 0
		140	#endif
		141
		142	#define ECB_CPP (__cplusplus+0)
		143	#define ECB_CPP11 (__cplusplus >= 201103L)
		144	#define ECB_CPP14 (__cplusplus >= 201402L)
		145	#define ECB_CPP17 (__cplusplus >= 201703L)
		146
		147	#if ECB_CPP
		148	#define ECB_C 0
		149	#define ECB_STDC_VERSION 0
		150	#else
		151	#define ECB_C 1
		152	#define ECB_STDC_VERSION __STDC_VERSION__
		153	#endif
		154
84	#define ECB_C99 (__STDC_VERSION__ >= 199901L)	155	#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
85	#define ECB_C11 (__STDC_VERSION__ >= 201112L)	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
86		168
87	/*****************************************************************************/	169	/*****************************************************************************/
88		170
89	/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */	171	/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
90	/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */	172	/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
91		173
92	#if ECB_NO_THREADS	174	#if ECB_NO_THREADS
93	# define ECB_NO_SMP 1	175	#define ECB_NO_SMP 1
94	#endif	176	#endif
95		177
96	#if ECB_NO_THREADS || ECB_NO_SMP	178	#if ECB_NO_SMP
97	#define ECB_MEMORY_FENCE do { } while (0)	179	#define ECB_MEMORY_FENCE do { } while (0)
98	#endif	180	#endif
99		181
100	#ifndef ECB_MEMORY_FENCE	182	/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
101	#if ECB_C11 && !defined __STDC_NO_ATOMICS__	183	#if __xlC__ && ECB_CPP
102	/* we assume that these memory fences work on all variables/all memory accesses, */	184	#include <builtins.h>
103	/* not just C11 atomics and atomic accesses */
104	#include <stdatomic.h>
105	#define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_acq_rel)
106	#define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
107	#define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
108	#endif	185	#endif
		186
		187	#if 1400 <= _MSC_VER
		188	#include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
109	#endif	189	#endif
110		190
111	#ifndef ECB_MEMORY_FENCE	191	#ifndef ECB_MEMORY_FENCE
112	#if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110	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")
113	#if __i386 || __i386__	194	#if __i386 || __i386__
114	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")	195	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
115	#define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */	196	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
116	#define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */	197	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
117	#elif __amd64 || __amd64__ || __x86_64 || __x86_64__	198	#elif ECB_GCC_AMD64
118	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")	199	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
119	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory")	200	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
120	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */	201	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
121	#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__	202	#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
122	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")	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 */
123	#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \	211	#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
124	|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__	212	|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
		213	|| defined __ARM_ARCH_6T2__
125	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")	214	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
126	#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \	215	#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
127	|| defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__	216	|| defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
128	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")	217	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
129	#elif __sparc || __sparc__	218	#elif __aarch64__
		219	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
		220	#elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
130	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory")	221	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
131	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")	222	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
132	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")	223	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
133	#elif defined __s390__ || defined __s390x__	224	#elif defined __s390__ || defined __s390x__
134	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")	225	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
135	#elif defined __mips__	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. */
136	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")	229	#define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
137	#elif defined __alpha__	230	#elif defined __alpha__
138	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")	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")
139	#endif	243	#endif
140	#endif	244	#endif
141	#endif	245	#endif
142		246
143	#ifndef ECB_MEMORY_FENCE	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
144	#if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__	264	#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
145	#define ECB_MEMORY_FENCE __sync_synchronize ()	265	#define ECB_MEMORY_FENCE __sync_synchronize ()
146	/*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */	266	#elif _MSC_VER >= 1500 /* VC++ 2008 */
147	/*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */	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()
148	#elif _MSC_VER >= 1400 /* VC++ 2005 */	272	#elif _MSC_VER >= 1400 /* VC++ 2005 */
149	#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)	273	#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
150	#define ECB_MEMORY_FENCE _ReadWriteBarrier ()	274	#define ECB_MEMORY_FENCE _ReadWriteBarrier ()
151	#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */	275	#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
152	#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()	276	#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
153	#elif defined _WIN32	277	#elif defined _WIN32
154	#include <WinNT.h>	278	#include <WinNT.h>
155	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */	279	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
156	#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110	280	#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
157	#include <mbarrier.h>	281	#include <mbarrier.h>
158	#define ECB_MEMORY_FENCE __machine_rw_barrier ()	282	#define ECB_MEMORY_FENCE __machine_rw_barrier ()
159	#define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()	283	#define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
160	#define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()	284	#define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
		285	#define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
161	#elif __xlC__	286	#elif __xlC__
162	#define ECB_MEMORY_FENCE __sync ()	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)
163	#endif	299	#endif
164	#endif	300	#endif
165		301
166	#ifndef ECB_MEMORY_FENCE	302	#ifndef ECB_MEMORY_FENCE
167	#if !ECB_AVOID_PTHREADS	303	#if !ECB_AVOID_PTHREADS
…		…
187		323
188	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE	324	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
189	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE	325	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
190	#endif	326	#endif
191		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
192	/*****************************************************************************/	332	/*****************************************************************************/
193		333
194	#if __cplusplus	334	#if ECB_CPP
195	#define ecb_inline static inline	335	#define ecb_inline static inline
196	#elif ECB_GCC_VERSION(2,5)	336	#elif ECB_GCC_VERSION(2,5)
197	#define ecb_inline static __inline__	337	#define ecb_inline static __inline__
198	#elif ECB_C99	338	#elif ECB_C99
199	#define ecb_inline static inline	339	#define ecb_inline static inline
…		…
213		353
214	#define ECB_CONCAT_(a, b) a ## b	354	#define ECB_CONCAT_(a, b) a ## b
215	#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)	355	#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
216	#define ECB_STRINGIFY_(a) # a	356	#define ECB_STRINGIFY_(a) # a
217	#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))
218		359
		360	/* This marks larger functions that do not neccessarily need to be inlined */
		361	/* The idea is to possibly compile the header twice, */
		362	/* once exposing only the declarations, another time to define external functions */
		363	/* TODO: possibly static would be best for these at the moment? */
219	#define ecb_function_ ecb_inline	364	#define ecb_function_ ecb_inline
220		365
221	#if ECB_GCC_VERSION(3,1)	366	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
222	#define ecb_attribute(attrlist) __attribute__(attrlist)	367	#define ecb_attribute(attrlist) __attribute__ (attrlist)
		368	#else
		369	#define ecb_attribute(attrlist)
		370	#endif
		371
		372	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
223	#define ecb_is_constant(expr) __builtin_constant_p (expr)	373	#define ecb_is_constant(expr) __builtin_constant_p (expr)
		374	#else
		375	/* possible C11 impl for integral types
		376	typedef struct ecb_is_constant_struct ecb_is_constant_struct;
		377	#define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
		378
		379	#define ecb_is_constant(expr) 0
		380	#endif
		381
		382	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
224	#define ecb_expect(expr,value) __builtin_expect ((expr),(value))	383	#define ecb_expect(expr,value) __builtin_expect ((expr),(value))
		384	#else
		385	#define ecb_expect(expr,value) (expr)
		386	#endif
		387
		388	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
225	#define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)	389	#define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
226	#else	390	#else
227	#define ecb_attribute(attrlist)
228	#define ecb_is_constant(expr) 0
229	#define ecb_expect(expr,value) (expr)
230	#define ecb_prefetch(addr,rw,locality)	391	#define ecb_prefetch(addr,rw,locality)
231	#endif	392	#endif
232		393
233	/* no emulation for ecb_decltype */	394	/* no emulation for ecb_decltype */
234	#if ECB_GCC_VERSION(4,5)	395	#if ECB_CPP11
		396	// older implementations might have problems with decltype(x)::type, work around it
		397	template<class T> struct ecb_decltype_t { typedef T type; };
235	#define ecb_decltype(x) __decltype(x)	398	#define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
236	#elif ECB_GCC_VERSION(3,0)	399	#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
237	#define ecb_decltype(x) __typeof(x)	400	#define ecb_decltype(x) __typeof__ (x)
238	#endif	401	#endif
239		402
		403	#if _MSC_VER >= 1300
		404	#define ecb_deprecated __declspec (deprecated)
		405	#else
		406	#define ecb_deprecated ecb_attribute ((__deprecated__))
		407	#endif
		408
		409	#if _MSC_VER >= 1500
		410	#define ecb_deprecated_message(msg) __declspec (deprecated (msg))
		411	#elif ECB_GCC_VERSION(4,5)
		412	#define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
		413	#else
		414	#define ecb_deprecated_message(msg) ecb_deprecated
		415	#endif
		416
		417	#if _MSC_VER >= 1400
		418	#define ecb_noinline __declspec (noinline)
		419	#else
240	#define ecb_noinline ecb_attribute ((__noinline__))	420	#define ecb_noinline ecb_attribute ((__noinline__))
		421	#endif
		422
241	#define ecb_unused ecb_attribute ((__unused__))	423	#define ecb_unused ecb_attribute ((__unused__))
242	#define ecb_const ecb_attribute ((__const__))	424	#define ecb_const ecb_attribute ((__const__))
243	#define ecb_pure ecb_attribute ((__pure__))	425	#define ecb_pure ecb_attribute ((__pure__))
244		426
245	#if ECB_C11	427	#if ECB_C11 || __IBMC_NORETURN
		428	/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
246	#define ecb_noreturn _Noreturn	429	#define ecb_noreturn _Noreturn
		430	#elif ECB_CPP11
		431	#define ecb_noreturn [[noreturn]]
		432	#elif _MSC_VER >= 1200
		433	/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
		434	#define ecb_noreturn __declspec (noreturn)
247	#else	435	#else
248	#define ecb_noreturn ecb_attribute ((__noreturn__))	436	#define ecb_noreturn ecb_attribute ((__noreturn__))
249	#endif	437	#endif
250		438
251	#if ECB_GCC_VERSION(4,3)	439	#if ECB_GCC_VERSION(4,3)
…		…
266	/* for compatibility to the rest of the world */	454	/* for compatibility to the rest of the world */
267	#define ecb_likely(expr) ecb_expect_true (expr)	455	#define ecb_likely(expr) ecb_expect_true (expr)
268	#define ecb_unlikely(expr) ecb_expect_false (expr)	456	#define ecb_unlikely(expr) ecb_expect_false (expr)
269		457
270	/* count trailing zero bits and count # of one bits */	458	/* count trailing zero bits and count # of one bits */
271	#if ECB_GCC_VERSION(3,4)	459	#if ECB_GCC_VERSION(3,4) \
272	/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */	460	|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
273	#define ecb_ld32(x) (__builtin_clz (x) ^ 31)	461	&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
274	#define ecb_ld64(x) (__builtin_clzll (x) ^ 63)	462	&& ECB_CLANG_BUILTIN(__builtin_popcount))
275	#define ecb_ctz32(x) __builtin_ctz (x)	463	#define ecb_ctz32(x) __builtin_ctz (x)
		464	#define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x))
276	#define ecb_ctz64(x) __builtin_ctzll (x)	465	#define ecb_clz32(x) __builtin_clz (x)
		466	#define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x))
		467	#define ecb_ld32(x) (ecb_clz32 (x) ^ 31)
		468	#define ecb_ld64(x) (ecb_clz64 (x) ^ 63)
277	#define ecb_popcount32(x) __builtin_popcount (x)	469	#define ecb_popcount32(x) __builtin_popcount (x)
278	/* no popcountll */	470	/* ecb_popcount64 is more difficult, see below */
279	#else	471	#else
280	ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;	472	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
281	ecb_function_ int	473	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x)
282	ecb_ctz32 (uint32_t x)
283	{	474	{
		475	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		476	unsigned long r;
		477	_BitScanForward (&r, x);
		478	return (int)r;
		479	#else
284	int r = 0;	480	int r;
285		481
286	x &= ~x + 1; /* this isolates the lowest bit */	482	x &= ~x + 1; /* this isolates the lowest bit */
287		483
288	#if ECB_branchless_on_i386	484	#if 1
		485	/* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */
		486	/* This happens to return 32 for x == 0, but the API does not support this */
		487
		488	/* -0 marks unused entries */
		489	static unsigned char table[64] =
		490	{
		491	32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14,
		492	10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15,
		493	31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26,
		494	30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0
		495	};
		496
		497	/* magic constant results in 33 unique values in the upper 6 bits */
		498	x *= 0x0450fbafU; /* == 17 * 65 * 65535 */
		499
		500	r = table [x >> 26];
		501	#elif 0 /* branchless on i386, typically */
		502	r = 0;
289	r += !!(x & 0xaaaaaaaa) << 0;	503	r += !!(x & 0xaaaaaaaa) << 0;
290	r += !!(x & 0xcccccccc) << 1;	504	r += !!(x & 0xcccccccc) << 1;
291	r += !!(x & 0xf0f0f0f0) << 2;	505	r += !!(x & 0xf0f0f0f0) << 2;
292	r += !!(x & 0xff00ff00) << 3;	506	r += !!(x & 0xff00ff00) << 3;
293	r += !!(x & 0xffff0000) << 4;	507	r += !!(x & 0xffff0000) << 4;
294	#else	508	#else /* branchless on modern compilers, typically */
		509	r = 0;
295	if (x & 0xaaaaaaaa) r += 1;	510	if (x & 0xaaaaaaaa) r += 1;
296	if (x & 0xcccccccc) r += 2;	511	if (x & 0xcccccccc) r += 2;
297	if (x & 0xf0f0f0f0) r += 4;	512	if (x & 0xf0f0f0f0) r += 4;
298	if (x & 0xff00ff00) r += 8;	513	if (x & 0xff00ff00) r += 8;
299	if (x & 0xffff0000) r += 16;	514	if (x & 0xffff0000) r += 16;
300	#endif	515	#endif
301		516
302	return r;	517	return r;
		518	#endif
303	}	519	}
304		520
305	ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;	521	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
306	ecb_function_ int	522	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x)
307	ecb_ctz64 (uint64_t x)
308	{	523	{
		524	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		525	unsigned long r;
		526	_BitScanForward64 (&r, x);
		527	return (int)r;
		528	#else
309	int shift = x & 0xffffffffU ? 0 : 32;	529	int shift = x & 0xffffffff ? 0 : 32;
310	return ecb_ctz32 (x >> shift) + shift;	530	return ecb_ctz32 (x >> shift) + shift;
		531	#endif
311	}	532	}
312		533
		534	ecb_function_ ecb_const int ecb_clz32 (uint32_t x);
		535	ecb_function_ ecb_const int ecb_clz32 (uint32_t x)
		536	{
		537	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		538	unsigned long r;
		539	_BitScanReverse (&r, x);
		540	return (int)r;
		541	#else
		542
		543	/* Robert Harley's algorithm from comp.arch 1996-12-07 */
		544	/* This happens to return 32 for x == 0, but the API does not support this */
		545
		546	/* -0 marks unused table elements */
		547	static unsigned char table[64] =
		548	{
		549	32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0,
		550	-0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0,
		551	17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18,
		552	5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0
		553	};
		554
		555	/* propagate leftmost 1 bit to the right */
		556	x |= x >> 1;
		557	x |= x >> 2;
		558	x |= x >> 4;
		559	x |= x >> 8;
		560	x |= x >> 16;
		561
		562	/* magic constant results in 33 unique values in the upper 6 bits */
		563	x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */
		564
		565	return table [x >> 26];
		566	#endif
		567	}
		568
		569	ecb_function_ ecb_const int ecb_clz64 (uint64_t x);
		570	ecb_function_ ecb_const int ecb_clz64 (uint64_t x)
		571	{
		572	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		573	unsigned long r;
		574	_BitScanReverse64 (&r, x);
		575	return (int)r;
		576	#else
		577	uint32_t l = x >> 32;
		578	int shift = l ? 0 : 32;
		579	return ecb_clz32 (l ? l : x) + shift;
		580	#endif
		581	}
		582
313	ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;	583	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
314	ecb_function_ int	584	ecb_function_ ecb_const int
315	ecb_popcount32 (uint32_t x)	585	ecb_popcount32 (uint32_t x)
316	{	586	{
317	x -= (x >> 1) & 0x55555555;	587	x -= (x >> 1) & 0x55555555;
318	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);	588	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
319	x = ((x >> 4) + x) & 0x0f0f0f0f;	589	x = ((x >> 4) + x) & 0x0f0f0f0f;
320	x *= 0x01010101;	590	x *= 0x01010101;
321		591
322	return x >> 24;	592	return x >> 24;
323	}	593	}
324		594
325	ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;	595	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
326	ecb_function_ int ecb_ld32 (uint32_t x)	596	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
327	{	597	{
		598	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		599	unsigned long r;
		600	_BitScanReverse (&r, x);
		601	return (int)r;
		602	#else
328	int r = 0;	603	int r = 0;
329		604
330	if (x >> 16) { x >>= 16; r += 16; }	605	if (x >> 16) { x >>= 16; r += 16; }
331	if (x >> 8) { x >>= 8; r += 8; }	606	if (x >> 8) { x >>= 8; r += 8; }
332	if (x >> 4) { x >>= 4; r += 4; }	607	if (x >> 4) { x >>= 4; r += 4; }
333	if (x >> 2) { x >>= 2; r += 2; }	608	if (x >> 2) { x >>= 2; r += 2; }
334	if (x >> 1) { r += 1; }	609	if (x >> 1) { r += 1; }
335		610
336	return r;	611	return r;
		612	#endif
337	}	613	}
338		614
339	ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;	615	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
340	ecb_function_ int ecb_ld64 (uint64_t x)	616	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
341	{	617	{
		618	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		619	unsigned long r;
		620	_BitScanReverse64 (&r, x);
		621	return (int)r;
		622	#else
342	int r = 0;	623	int r = 0;
343		624
344	if (x >> 32) { x >>= 32; r += 32; }	625	if (x >> 32) { x >>= 32; r += 32; }
345		626
346	return r + ecb_ld32 (x);	627	return r + ecb_ld32 (x);
347	}
348	#endif	628	#endif
		629	}
		630	#endif
349		631
350	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;	632	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
351	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }	633	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
352	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;	634	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
353	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }	635	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
354		636
355	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;	637	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
356	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)	638	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
357	{	639	{
358	return ( (x * 0x0802U & 0x22110U)	640	return ( (x * 0x0802U & 0x22110U)
359	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;	641	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
360	}	642	}
361		643
362	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;	644	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
363	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)	645	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
364	{	646	{
365	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);	647	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
366	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);	648	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
367	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);	649	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
368	x = ( x >> 8 ) | ( x << 8);	650	x = ( x >> 8 ) | ( x << 8);
369		651
370	return x;	652	return x;
371	}	653	}
372		654
373	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;	655	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
374	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)	656	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
375	{	657	{
376	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);	658	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
377	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);	659	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
378	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);	660	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
379	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);	661	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
380	x = ( x >> 16 ) | ( x << 16);	662	x = ( x >> 16 ) | ( x << 16);
381		663
382	return x;	664	return x;
383	}	665	}
384		666
		667	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
		668	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x)
		669	{
385	/* popcount64 is only available on 64 bit cpus as gcc builtin */	670	/* popcount64 is only available on 64 bit cpus as gcc builtin. */
386	/* so for this version we are lazy */	671	/* also, gcc/clang make this surprisingly difficult to use */
387	ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;	672	#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
388	ecb_function_ int	673	return __builtin_popcountl (x);
389	ecb_popcount64 (uint64_t x)	674	#else
390	{
391	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);	675	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
		676	#endif
392	}	677	}
393		678
394	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;	679	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
395	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;	680	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
396	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;	681	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
397	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;	682	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
398	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;	683	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
399	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;	684	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
400	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;	685	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
401	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;	686	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
402		687
403	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }	688	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
404	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }	689	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
405	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }	690	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
406	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }	691	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
407	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }	692	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
408	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }	693	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
409	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }	694	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
410	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }	695	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
411		696
412	#if ECB_GCC_VERSION(4,3)	697	#if ECB_CPP
		698
		699	inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
		700	inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
		701	inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
		702	inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
		703
		704	inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
		705	inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
		706	inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
		707	inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
		708
		709	inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
		710	inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
		711	inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
		712	inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
		713
		714	inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
		715	inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
		716	inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
		717	inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
		718
		719	inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
		720	inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
		721	inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
		722
		723	inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
		724	inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
		725	inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
		726	inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
		727
		728	inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
		729	inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
		730	inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
		731	inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
		732
		733	#endif
		734
		735	#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
		736	#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
		737	#define ecb_bswap16(x) __builtin_bswap16 (x)
		738	#else
413	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)	739	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
		740	#endif
414	#define ecb_bswap32(x) __builtin_bswap32 (x)	741	#define ecb_bswap32(x) __builtin_bswap32 (x)
415	#define ecb_bswap64(x) __builtin_bswap64 (x)	742	#define ecb_bswap64(x) __builtin_bswap64 (x)
		743	#elif _MSC_VER
		744	#include <stdlib.h>
		745	#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
		746	#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
		747	#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
416	#else	748	#else
417	ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;	749	ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
418	ecb_function_ uint16_t	750	ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x)
419	ecb_bswap16 (uint16_t x)
420	{	751	{
421	return ecb_rotl16 (x, 8);	752	return ecb_rotl16 (x, 8);
422	}	753	}
423		754
424	ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;	755	ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
425	ecb_function_ uint32_t	756	ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x)
426	ecb_bswap32 (uint32_t x)
427	{	757	{
428	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);	758	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
429	}	759	}
430		760
431	ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;	761	ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
432	ecb_function_ uint64_t	762	ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x)
433	ecb_bswap64 (uint64_t x)
434	{	763	{
435	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);	764	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
436	}	765	}
437	#endif	766	#endif
438		767
439	#if ECB_GCC_VERSION(4,5)	768	#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
440	#define ecb_unreachable() __builtin_unreachable ()	769	#define ecb_unreachable() __builtin_unreachable ()
441	#else	770	#else
442	/* this seems to work fine, but gcc always emits a warning for it :/ */	771	/* this seems to work fine, but gcc always emits a warning for it :/ */
443	ecb_inline void ecb_unreachable (void) ecb_noreturn;	772	ecb_inline ecb_noreturn void ecb_unreachable (void);
444	ecb_inline void ecb_unreachable (void) { }	773	ecb_inline ecb_noreturn void ecb_unreachable (void) { }
445	#endif	774	#endif
446		775
447	/* try to tell the compiler that some condition is definitely true */	776	/* try to tell the compiler that some condition is definitely true */
448	#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0)	777	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
449		778
450	ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;	779	ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
451	ecb_inline unsigned char	780	ecb_inline ecb_const uint32_t ecb_byteorder_helper (void)
452	ecb_byteorder_helper (void)
453	{	781	{
454	const uint32_t u = 0x11223344;	782	/* the union code still generates code under pressure in gcc, */
455	return *(unsigned char *)&u;	783	/* but less than using pointers, and always seems to */
		784	/* successfully return a constant. */
		785	/* the reason why we have this horrible preprocessor mess */
		786	/* is to avoid it in all cases, at least on common architectures */
		787	/* or when using a recent enough gcc version (>= 4.6) */
		788	#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
		789	|| ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
		790	#define ECB_LITTLE_ENDIAN 1
		791	return 0x44332211;
		792	#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
		793	|| ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
		794	#define ECB_BIG_ENDIAN 1
		795	return 0x11223344;
		796	#else
		797	union
		798	{
		799	uint8_t c[4];
		800	uint32_t u;
		801	} u = { 0x11, 0x22, 0x33, 0x44 };
		802	return u.u;
		803	#endif
456	}	804	}
457		805
458	ecb_inline ecb_bool ecb_big_endian (void) ecb_const;	806	ecb_inline ecb_const ecb_bool ecb_big_endian (void);
459	ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }	807	ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
460	ecb_inline ecb_bool ecb_little_endian (void) ecb_const;	808	ecb_inline ecb_const ecb_bool ecb_little_endian (void);
461	ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }	809	ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
		810
		811	/*****************************************************************************/
		812	/* unaligned load/store */
		813
		814	ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		815	ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		816	ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		817
		818	ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		819	ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		820	ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		821
		822	ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		823	ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		824	ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		825
		826	ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
		827	ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
		828	ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
		829
		830	ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
		831	ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
		832	ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
		833
		834	ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		835	ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		836	ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		837
		838	ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		839	ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		840	ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		841
		842	ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
		843	ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
		844	ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
		845
		846	ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
		847	ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
		848	ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
		849
		850	ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
		851	ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
		852	ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
		853
		854	#if ECB_CPP
		855
		856	inline uint8_t ecb_bswap (uint8_t v) { return v; }
		857	inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
		858	inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
		859	inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
		860
		861	template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		862	template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		863	template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
		864	template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
		865	template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
		866	template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
		867	template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
		868	template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
		869
		870	template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		871	template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		872	template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
		873	template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
		874	template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
		875	template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
		876	template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
		877	template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
		878
		879	#endif
		880
		881	/*****************************************************************************/
		882	/* pointer/integer hashing */
		883
		884	/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
		885	ecb_function_ uint32_t ecb_mix32 (uint32_t v);
		886	ecb_function_ uint32_t ecb_mix32 (uint32_t v)
		887	{
		888	v ^= v >> 16; v *= 0x7feb352dU;
		889	v ^= v >> 15; v *= 0x846ca68bU;
		890	v ^= v >> 16;
		891	return v;
		892	}
		893
		894	ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
		895	ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
		896	{
		897	v ^= v >> 16 ; v *= 0x43021123U;
		898	v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
		899	v ^= v >> 16 ;
		900	return v;
		901	}
		902
		903	/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
		904	ecb_function_ uint64_t ecb_mix64 (uint64_t v);
		905	ecb_function_ uint64_t ecb_mix64 (uint64_t v)
		906	{
		907	v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
		908	v ^= v >> 27; v *= 0x94d049bb133111ebU;
		909	v ^= v >> 31;
		910	return v;
		911	}
		912
		913	ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
		914	ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
		915	{
		916	v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
		917	v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
		918	v ^= v >> 30 ^ v >> 60;
		919	return v;
		920	}
		921
		922	ecb_function_ uintptr_t ecb_ptrmix (void *p);
		923	ecb_function_ uintptr_t ecb_ptrmix (void *p)
		924	{
		925	#if ECB_PTRSIZE <= 4
		926	return ecb_mix32 ((uint32_t)p);
		927	#else
		928	return ecb_mix64 ((uint64_t)p);
		929	#endif
		930	}
		931
		932	ecb_function_ void *ecb_ptrunmix (uintptr_t v);
		933	ecb_function_ void *ecb_ptrunmix (uintptr_t v)
		934	{
		935	#if ECB_PTRSIZE <= 4
		936	return (void *)ecb_unmix32 (v);
		937	#else
		938	return (void *)ecb_unmix64 (v);
		939	#endif
		940	}
		941
		942	#if ECB_CPP
		943
		944	template<typename T>
		945	inline uintptr_t ecb_ptrmix (T *p)
		946	{
		947	return ecb_ptrmix (static_cast<void *>(p));
		948	}
		949
		950	template<typename T>
		951	inline T *ecb_ptrunmix (uintptr_t v)
		952	{
		953	return static_cast<T *>(ecb_ptrunmix (v));
		954	}
		955
		956	#endif
		957
		958	/*****************************************************************************/
		959	/* gray code */
		960
		961	ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); }
		962	ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); }
		963	ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); }
		964	ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); }
		965
		966	ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g);
		967	ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g)
		968	{
		969	g ^= g >> 1;
		970	g ^= g >> 2;
		971	g ^= g >> 4;
		972
		973	return g;
		974	}
		975
		976	ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g);
		977	ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g)
		978	{
		979	g ^= g >> 1;
		980	g ^= g >> 2;
		981	g ^= g >> 4;
		982	g ^= g >> 8;
		983
		984	return g;
		985	}
		986
		987	ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g);
		988	ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g)
		989	{
		990	g ^= g >> 1;
		991	g ^= g >> 2;
		992	g ^= g >> 4;
		993	g ^= g >> 8;
		994	g ^= g >> 16;
		995
		996	return g;
		997	}
		998
		999	ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g);
		1000	ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g)
		1001	{
		1002	g ^= g >> 1;
		1003	g ^= g >> 2;
		1004	g ^= g >> 4;
		1005	g ^= g >> 8;
		1006	g ^= g >> 16;
		1007	g ^= g >> 32;
		1008
		1009	return g;
		1010	}
		1011
		1012	#if ECB_CPP
		1013
		1014	ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); }
		1015	ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); }
		1016	ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); }
		1017	ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); }
		1018
		1019	ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); }
		1020	ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); }
		1021	ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); }
		1022	ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); }
		1023
		1024	#endif
		1025
		1026	/*****************************************************************************/
		1027	/* 2d hilbert curves */
		1028
		1029	/* algorithm from the book Hacker's Delight, modified to not */
		1030	/* run into undefined behaviour for n==16 */
		1031	static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s);
		1032	static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
		1033	{
		1034	uint32_t comp, swap, cs, t, sr;
		1035
		1036	/* pad s on the left (unused) bits with 01 (no change groups) */
		1037	s |= 0x55555555U << n << n;
		1038	/* "s shift right" */
		1039	sr = (s >> 1) & 0x55555555U;
		1040	/* compute complement and swap info in two-bit groups */
		1041	cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
		1042
		1043	/* parallel prefix xor op to propagate both complement
		1044	* and swap info together from left to right (there is
		1045	* no step "cs ^= cs >> 1", so in effect it computes
		1046	* two independent parallel prefix operations on two
		1047	* interleaved sets of sixteen bits).
		1048	*/
		1049	cs ^= cs >> 2;
		1050	cs ^= cs >> 4;
		1051	cs ^= cs >> 8;
		1052	cs ^= cs >> 16;
		1053
		1054	/* separate swap and complement bits */
		1055	swap = cs & 0x55555555U;
		1056	comp = (cs >> 1) & 0x55555555U;
		1057
		1058	/* calculate coordinates in odd and even bit positions */
		1059	t = (s & swap) ^ comp;
		1060	s = s ^ sr ^ t ^ (t << 1);
		1061
		1062	/* unpad/clear out any junk on the left */
		1063	s = s & ((1 << n << n) - 1);
		1064
		1065	/* Now "unshuffle" to separate the x and y bits. */
		1066	t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
		1067	t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
		1068	t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
		1069	t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
		1070
		1071	/* now s contains two 16-bit coordinates */
		1072	return s;
		1073	}
		1074
		1075	/* 64 bit, a straightforward extension to the 32 bit case */
		1076	static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s);
		1077	static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
		1078	{
		1079	uint64_t comp, swap, cs, t, sr;
		1080
		1081	/* pad s on the left (unused) bits with 01 (no change groups) */
		1082	s |= 0x5555555555555555U << n << n;
		1083	/* "s shift right" */
		1084	sr = (s >> 1) & 0x5555555555555555U;
		1085	/* compute complement and swap info in two-bit groups */
		1086	cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
		1087
		1088	/* parallel prefix xor op to propagate both complement
		1089	* and swap info together from left to right (there is
		1090	* no step "cs ^= cs >> 1", so in effect it computes
		1091	* two independent parallel prefix operations on two
		1092	* interleaved sets of thirty-two bits).
		1093	*/
		1094	cs ^= cs >> 2;
		1095	cs ^= cs >> 4;
		1096	cs ^= cs >> 8;
		1097	cs ^= cs >> 16;
		1098	cs ^= cs >> 32;
		1099
		1100	/* separate swap and complement bits */
		1101	swap = cs & 0x5555555555555555U;
		1102	comp = (cs >> 1) & 0x5555555555555555U;
		1103
		1104	/* calculate coordinates in odd and even bit positions */
		1105	t = (s & swap) ^ comp;
		1106	s = s ^ sr ^ t ^ (t << 1);
		1107
		1108	/* unpad/clear out any junk on the left */
		1109	s = s & ((1 << n << n) - 1);
		1110
		1111	/* Now "unshuffle" to separate the x and y bits. */
		1112	t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
		1113	t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
		1114	t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
		1115	t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
		1116	t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
		1117
		1118	/* now s contains two 32-bit coordinates */
		1119	return s;
		1120	}
		1121
		1122	/* algorithm from the book Hacker's Delight, but a similar algorithm*/
		1123	/* is given in https://doi.org/10.1002/spe.4380160103 */
		1124	/* this has been slightly improved over the original version */
		1125	ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy);
		1126	ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
		1127	{
		1128	uint32_t row;
		1129	uint32_t state = 0;
		1130	uint32_t s = 0;
		1131
		1132	do
		1133	{
		1134	--n;
		1135
		1136	row = 4 * state
		1137	| (2 & (xy >> n >> 15))
		1138	| (1 & (xy >> n ));
		1139
		1140	/* these funky constants are lookup tables for two-bit values */
		1141	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1142	state = (0x8fe65831U >> 2 * row) & 3;
		1143	}
		1144	while (n > 0);
		1145
		1146	return s;
		1147	}
		1148
		1149	/* 64 bit, essentially the same as 32 bit */
		1150	ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy);
		1151	ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
		1152	{
		1153	uint32_t row;
		1154	uint32_t state = 0;
		1155	uint64_t s = 0;
		1156
		1157	do
		1158	{
		1159	--n;
		1160
		1161	row = 4 * state
		1162	| (2 & (xy >> n >> 31))
		1163	| (1 & (xy >> n ));
		1164
		1165	/* these funky constants are lookup tables for two-bit values */
		1166	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1167	state = (0x8fe65831U >> 2 * row) & 3;
		1168	}
		1169	while (n > 0);
		1170
		1171	return s;
		1172	}
		1173
		1174	/*****************************************************************************/
		1175	/* division */
462		1176
463	#if ECB_GCC_VERSION(3,0) || ECB_C99	1177	#if ECB_GCC_VERSION(3,0) || ECB_C99
		1178	/* C99 tightened the definition of %, so we can use a more efficient version */
464	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))	1179	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
465	#else	1180	#else
466	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))	1181	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
467	#endif	1182	#endif
468		1183
469	#if __cplusplus	1184	#if ECB_CPP
470	template<typename T>	1185	template<typename T>
471	static inline T ecb_div_rd (T val, T div)	1186	static inline T ecb_div_rd (T val, T div)
472	{	1187	{
473	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;	1188	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
474	}	1189	}
…		…
479	}	1194	}
480	#else	1195	#else
481	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))	1196	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
482	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))	1197	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
483	#endif	1198	#endif
		1199
		1200	/*****************************************************************************/
		1201	/* array length */
484		1202
485	#if ecb_cplusplus_does_not_suck	1203	#if ecb_cplusplus_does_not_suck
486	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */	1204	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
487	template<typename T, int N>	1205	template<typename T, int N>
488	static inline int ecb_array_length (const T (&arr)[N])	1206	static inline int ecb_array_length (const T (&arr)[N])
…		…
491	}	1209	}
492	#else	1210	#else
493	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))	1211	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
494	#endif	1212	#endif
495		1213
		1214	/*****************************************************************************/
		1215	/* IEEE 754-2008 half float conversions */
		1216
		1217	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
		1218	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x)
		1219	{
		1220	unsigned int s = (x & 0x8000) << (31 - 15);
		1221	int e = (x >> 10) & 0x001f;
		1222	unsigned int m = x & 0x03ff;
		1223
		1224	if (ecb_expect_false (e == 31))
		1225	/* infinity or NaN */
		1226	e = 255 - (127 - 15);
		1227	else if (ecb_expect_false (!e))
		1228	{
		1229	if (ecb_expect_true (!m))
		1230	/* zero, handled by code below by forcing e to 0 */
		1231	e = 0 - (127 - 15);
		1232	else
		1233	{
		1234	/* subnormal, renormalise */
		1235	unsigned int s = 10 - ecb_ld32 (m);
		1236
		1237	m = (m << s) & 0x3ff; /* mask implicit bit */
		1238	e -= s - 1;
		1239	}
		1240	}
		1241
		1242	/* e and m now are normalised, or zero, (or inf or nan) */
		1243	e += 127 - 15;
		1244
		1245	return s | (e << 23) | (m << (23 - 10));
		1246	}
		1247
		1248	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
		1249	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x)
		1250	{
		1251	unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
		1252	int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
		1253	unsigned int m = x & 0x007fffff;
		1254
		1255	x &= 0x7fffffff;
		1256
		1257	/* if it's within range of binary16 normals, use fast path */
		1258	if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
		1259	{
		1260	/* mantissa round-to-even */
		1261	m += 0x00000fff + ((m >> (23 - 10)) & 1);
		1262
		1263	/* handle overflow */
		1264	if (ecb_expect_false (m >= 0x00800000))
		1265	{
		1266	m >>= 1;
		1267	e += 1;
		1268	}
		1269
		1270	return s | (e << 10) | (m >> (23 - 10));
		1271	}
		1272
		1273	/* handle large numbers and infinity */
		1274	if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
		1275	return s | 0x7c00;
		1276
		1277	/* handle zero, subnormals and small numbers */
		1278	if (ecb_expect_true (x < 0x38800000))
		1279	{
		1280	/* zero */
		1281	if (ecb_expect_true (!x))
		1282	return s;
		1283
		1284	/* handle subnormals */
		1285
		1286	/* too small, will be zero */
		1287	if (e < (14 - 24)) /* might not be sharp, but is good enough */
		1288	return s;
		1289
		1290	m |= 0x00800000; /* make implicit bit explicit */
		1291
		1292	/* very tricky - we need to round to the nearest e (+10) bit value */
		1293	{
		1294	unsigned int bits = 14 - e;
		1295	unsigned int half = (1 << (bits - 1)) - 1;
		1296	unsigned int even = (m >> bits) & 1;
		1297
		1298	/* if this overflows, we will end up with a normalised number */
		1299	m = (m + half + even) >> bits;
		1300	}
		1301
		1302	return s | m;
		1303	}
		1304
		1305	/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
		1306	m >>= 13;
		1307
		1308	return s | 0x7c00 | m | !m;
		1309	}
		1310
		1311	/*******************************************************************************/
		1312	/* fast integer to ascii */
		1313
		1314	/*
		1315	* This code is pretty complicated because it is general. The idea behind it,
		1316	* however, is pretty simple: first, the number is multiplied with a scaling
		1317	* factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
		1318	* number with the first digit in the upper bits.
		1319	* Then this digit is converted to text and masked out. The resulting number
		1320	* is then multiplied by 10, by multiplying the fixed point representation
		1321	* by 5 and shifting the (binary) decimal point one to the right, so a 4.28
		1322	* format becomes 5.27, 6.26 and so on.
		1323	* The rest involves only advancing the pointer if we already generated a
		1324	* non-zero digit, so leading zeroes are overwritten.
		1325	*/
		1326
		1327	/* simply return a mask with "bits" bits set */
		1328	#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
		1329
		1330	/* oputput a single digit. maskvalue is 10**digitidx */
		1331	#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
		1332	if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
		1333	{ \
		1334	char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
		1335	*ptr = digit + '0'; /* output it */ \
		1336	nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
		1337	ptr += nz; /* output digit only if non-zero digit seen */ \
		1338	x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
		1339	}
		1340
		1341	/* convert integer to fixed point format and multiply out digits, highest first */
		1342	/* requires magic constants: max. digits and number of bits after the decimal point */
		1343	#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
		1344	ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
		1345	{ \
		1346	char nz = lz; /* non-zero digit seen? */ \
		1347	/* convert to x.bits fixed-point */ \
		1348	type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
		1349	/* output up to 10 digits */ \
		1350	ecb_i2a_digit (type,bits,digitmask, 1, 0); \
		1351	ecb_i2a_digit (type,bits,digitmask, 10, 1); \
		1352	ecb_i2a_digit (type,bits,digitmask, 100, 2); \
		1353	ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
		1354	ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
		1355	ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
		1356	ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
		1357	ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
		1358	ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
		1359	ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
		1360	return ptr; \
		1361	}
		1362
		1363	/* predefined versions of the above, for various digits */
		1364	/* ecb_i2a_xN = almost N digits, limit defined by macro */
		1365	/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
		1366	/* ecb_i2a_0N = exactly N digits, including leading zeroes */
		1367
		1368	/* non-leading-zero versions, limited range */
		1369	#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
		1370	#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
		1371	ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
		1372	ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
		1373
		1374	/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
		1375	ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
		1376	ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
		1377	ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
		1378	ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
		1379	ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
		1380	ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
		1381	ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
		1382	ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
		1383
		1384	/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
		1385	ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
		1386	ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
		1387	ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
		1388	ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
		1389	ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
		1390	ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
		1391	ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
		1392	ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
		1393
		1394	#define ECB_I2A_I32_DIGITS 11
		1395	#define ECB_I2A_U32_DIGITS 10
		1396	#define ECB_I2A_I64_DIGITS 20
		1397	#define ECB_I2A_U64_DIGITS 21
		1398	#define ECB_I2A_MAX_DIGITS 21
		1399
		1400	ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u)
		1401	ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u)
		1402	{
		1403	#if ECB_64BIT_NATIVE
		1404	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1405	ptr = ecb_i2a_x10 (ptr, u);
		1406	else /* x10 almost, but not fully, covers 32 bit */
		1407	{
		1408	uint32_t u1 = u % 1000000000;
		1409	uint32_t u2 = u / 1000000000;
		1410
		1411	*ptr++ = u2 + '0';
		1412	ptr = ecb_i2a_09 (ptr, u1);
		1413	}
		1414	#else
		1415	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1416	ecb_i2a_x5 (ptr, u);
		1417	else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
		1418	{
		1419	uint32_t u1 = u % 10000;
		1420	uint32_t u2 = u / 10000;
		1421
		1422	ptr = ecb_i2a_x5 (ptr, u2);
		1423	ptr = ecb_i2a_04 (ptr, u1);
		1424	}
		1425	else
		1426	{
		1427	uint32_t u1 = u % 10000;
		1428	uint32_t ua = u / 10000;
		1429	uint32_t u2 = ua % 10000;
		1430	uint32_t u3 = ua / 10000;
		1431
		1432	ptr = ecb_i2a_2 (ptr, u3);
		1433	ptr = ecb_i2a_04 (ptr, u2);
		1434	ptr = ecb_i2a_04 (ptr, u1);
		1435	}
496	#endif	1436	#endif
497		1437
		1438	return ptr;
		1439	}
		1440
		1441	ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v);
		1442	ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v)
		1443	{
		1444	*ptr = '-'; ptr += v < 0;
		1445	uint32_t u = v < 0 ? -(uint32_t)v : v;
		1446
		1447	#if ECB_64BIT_NATIVE
		1448	ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
		1449	#else
		1450	ptr = ecb_i2a_u32 (ptr, u);
		1451	#endif
		1452
		1453	return ptr;
		1454	}
		1455
		1456	ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u);
		1457	ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u)
		1458	{
		1459	#if ECB_64BIT_NATIVE
		1460	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1461	ptr = ecb_i2a_x10 (ptr, u);
		1462	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1463	{
		1464	uint64_t u1 = u % 1000000000;
		1465	uint64_t u2 = u / 1000000000;
		1466
		1467	ptr = ecb_i2a_x10 (ptr, u2);
		1468	ptr = ecb_i2a_09 (ptr, u1);
		1469	}
		1470	else
		1471	{
		1472	uint64_t u1 = u % 1000000000;
		1473	uint64_t ua = u / 1000000000;
		1474	uint64_t u2 = ua % 1000000000;
		1475	uint64_t u3 = ua / 1000000000;
		1476
		1477	ptr = ecb_i2a_2 (ptr, u3);
		1478	ptr = ecb_i2a_09 (ptr, u2);
		1479	ptr = ecb_i2a_09 (ptr, u1);
		1480	}
		1481	#else
		1482	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1483	ptr = ecb_i2a_x5 (ptr, u);
		1484	else
		1485	{
		1486	uint64_t u1 = u % 10000;
		1487	uint64_t u2 = u / 10000;
		1488
		1489	ptr = ecb_i2a_u64 (ptr, u2);
		1490	ptr = ecb_i2a_04 (ptr, u1);
		1491	}
		1492	#endif
		1493
		1494	return ptr;
		1495	}
		1496
		1497	ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v)
		1498	ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v)
		1499	{
		1500	*ptr = '-'; ptr += v < 0;
		1501	uint64_t u = v < 0 ? -(uint64_t)v : v;
		1502
		1503	#if ECB_64BIT_NATIVE
		1504	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1505	ptr = ecb_i2a_x10 (ptr, u);
		1506	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1507	{
		1508	uint64_t u1 = u % 1000000000;
		1509	uint64_t u2 = u / 1000000000;
		1510
		1511	ptr = ecb_i2a_x10 (ptr, u2);
		1512	ptr = ecb_i2a_09 (ptr, u1);
		1513	}
		1514	else
		1515	{
		1516	uint64_t u1 = u % 1000000000;
		1517	uint64_t ua = u / 1000000000;
		1518	uint64_t u2 = ua % 1000000000;
		1519	uint64_t u3 = ua / 1000000000;
		1520
		1521	/* 2**31 is 19 digits, so the top is exactly one digit */
		1522	*ptr++ = u3 + '0';
		1523	ptr = ecb_i2a_09 (ptr, u2);
		1524	ptr = ecb_i2a_09 (ptr, u1);
		1525	}
		1526	#else
		1527	ptr = ecb_i2a_u64 (ptr, u);
		1528	#endif
		1529
		1530	return ptr;
		1531	}
		1532
		1533	/*******************************************************************************/
		1534	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
		1535
		1536	/* basically, everything uses "ieee pure-endian" floating point numbers */
		1537	/* the only noteworthy exception is ancient armle, which uses order 43218765 */
		1538	#if 0 \
		1539	|| __i386 || __i386__ \
		1540	|| ECB_GCC_AMD64 \
		1541	|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
		1542	|| defined __s390__ || defined __s390x__ \
		1543	|| defined __mips__ \
		1544	|| defined __alpha__ \
		1545	|| defined __hppa__ \
		1546	|| defined __ia64__ \
		1547	|| defined __m68k__ \
		1548	|| defined __m88k__ \
		1549	|| defined __sh__ \
		1550	|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
		1551	|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
		1552	|| defined __aarch64__
		1553	#define ECB_STDFP 1
		1554	#else
		1555	#define ECB_STDFP 0
		1556	#endif
		1557
		1558	#ifndef ECB_NO_LIBM
		1559
		1560	#include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
		1561
		1562	/* only the oldest of old doesn't have this one. solaris. */
		1563	#ifdef INFINITY
		1564	#define ECB_INFINITY INFINITY
		1565	#else
		1566	#define ECB_INFINITY HUGE_VAL
		1567	#endif
		1568
		1569	#ifdef NAN
		1570	#define ECB_NAN NAN
		1571	#else
		1572	#define ECB_NAN ECB_INFINITY
		1573	#endif
		1574
		1575	#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
		1576	#define ecb_ldexpf(x,e) ldexpf ((x), (e))
		1577	#define ecb_frexpf(x,e) frexpf ((x), (e))
		1578	#else
		1579	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
		1580	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
		1581	#endif
		1582
		1583	/* convert a float to ieee single/binary32 */
		1584	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
		1585	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x)
		1586	{
		1587	uint32_t r;
		1588
		1589	#if ECB_STDFP
		1590	memcpy (&r, &x, 4);
		1591	#else
		1592	/* slow emulation, works for anything but -0 */
		1593	uint32_t m;
		1594	int e;
		1595
		1596	if (x == 0e0f ) return 0x00000000U;
		1597	if (x > +3.40282346638528860e+38f) return 0x7f800000U;
		1598	if (x < -3.40282346638528860e+38f) return 0xff800000U;
		1599	if (x != x ) return 0x7fbfffffU;
		1600
		1601	m = ecb_frexpf (x, &e) * 0x1000000U;
		1602
		1603	r = m & 0x80000000U;
		1604
		1605	if (r)
		1606	m = -m;
		1607
		1608	if (e <= -126)
		1609	{
		1610	m &= 0xffffffU;
		1611	m >>= (-125 - e);
		1612	e = -126;
		1613	}
		1614
		1615	r |= (e + 126) << 23;
		1616	r |= m & 0x7fffffU;
		1617	#endif
		1618
		1619	return r;
		1620	}
		1621
		1622	/* converts an ieee single/binary32 to a float */
		1623	ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
		1624	ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x)
		1625	{
		1626	float r;
		1627
		1628	#if ECB_STDFP
		1629	memcpy (&r, &x, 4);
		1630	#else
		1631	/* emulation, only works for normals and subnormals and +0 */
		1632	int neg = x >> 31;
		1633	int e = (x >> 23) & 0xffU;
		1634
		1635	x &= 0x7fffffU;
		1636
		1637	if (e)
		1638	x |= 0x800000U;
		1639	else
		1640	e = 1;
		1641
		1642	/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
		1643	r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
		1644
		1645	r = neg ? -r : r;
		1646	#endif
		1647
		1648	return r;
		1649	}
		1650
		1651	/* convert a double to ieee double/binary64 */
		1652	ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
		1653	ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x)
		1654	{
		1655	uint64_t r;
		1656
		1657	#if ECB_STDFP
		1658	memcpy (&r, &x, 8);
		1659	#else
		1660	/* slow emulation, works for anything but -0 */
		1661	uint64_t m;
		1662	int e;
		1663
		1664	if (x == 0e0 ) return 0x0000000000000000U;
		1665	if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
		1666	if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
		1667	if (x != x ) return 0X7ff7ffffffffffffU;
		1668
		1669	m = frexp (x, &e) * 0x20000000000000U;
		1670
		1671	r = m & 0x8000000000000000;;
		1672
		1673	if (r)
		1674	m = -m;
		1675
		1676	if (e <= -1022)
		1677	{
		1678	m &= 0x1fffffffffffffU;
		1679	m >>= (-1021 - e);
		1680	e = -1022;
		1681	}
		1682
		1683	r |= ((uint64_t)(e + 1022)) << 52;
		1684	r |= m & 0xfffffffffffffU;
		1685	#endif
		1686
		1687	return r;
		1688	}
		1689
		1690	/* converts an ieee double/binary64 to a double */
		1691	ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
		1692	ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x)
		1693	{
		1694	double r;
		1695
		1696	#if ECB_STDFP
		1697	memcpy (&r, &x, 8);
		1698	#else
		1699	/* emulation, only works for normals and subnormals and +0 */
		1700	int neg = x >> 63;
		1701	int e = (x >> 52) & 0x7ffU;
		1702
		1703	x &= 0xfffffffffffffU;
		1704
		1705	if (e)
		1706	x |= 0x10000000000000U;
		1707	else
		1708	e = 1;
		1709
		1710	/* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
		1711	r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
		1712
		1713	r = neg ? -r : r;
		1714	#endif
		1715
		1716	return r;
		1717	}
		1718
		1719	/* convert a float to ieee half/binary16 */
		1720	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
		1721	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x)
		1722	{
		1723	return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
		1724	}
		1725
		1726	/* convert an ieee half/binary16 to float */
		1727	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
		1728	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x)
		1729	{
		1730	return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
		1731	}
		1732
		1733	#endif
		1734
		1735	#endif
		1736

Diff Legend

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