ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libecb/ecb.h

Comparing libecb/ecb.h (file contents):
Revision 1.143 by root, Fri Oct 17 10:56:48 2014 UTC vs.
Revision 1.215 by root, Fri Mar 25 15:51:15 2022 UTC

1	/*	1	/*
2	* libecb - http://software.schmorp.de/pkg/libecb	2	* libecb - http://software.schmorp.de/pkg/libecb
3	*	3	*
4	* Copyright (©) 2009-2014 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:
…		…
40		40
41	#ifndef ECB_H	41	#ifndef ECB_H
42	#define ECB_H	42	#define ECB_H
43		43
44	/* 16 bits major, 16 bits minor */	44	/* 16 bits major, 16 bits minor */
45	#define ECB_VERSION 0x00010004	45	#define ECB_VERSION 0x0001000c
46		46
47	#ifdef _WIN32	47	#include <string.h> /* for memcpy */
		48
		49	#if defined (_WIN32) && !defined (__MINGW32__)
48	typedef signed char int8_t;	50	typedef signed char int8_t;
49	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;
50	typedef signed short int16_t;	54	typedef signed short int16_t;
51	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;
52	typedef signed int int32_t;	58	typedef signed int int32_t;
53	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;
54	#if __GNUC__	62	#if __GNUC__
55	typedef signed long long int64_t;	63	typedef signed long long int64_t;
56	typedef unsigned long long uint64_t;	64	typedef unsigned long long uint64_t;
57	#else /* _MSC_VER || __BORLANDC__ */	65	#else /* _MSC_VER || __BORLANDC__ */
58	typedef signed __int64 int64_t;	66	typedef signed __int64 int64_t;
59	typedef unsigned __int64 uint64_t;	67	typedef unsigned __int64 uint64_t;
60	#endif	68	#endif
		69	typedef int64_t int_fast64_t;
		70	typedef uint64_t uint_fast64_t;
61	#ifdef _WIN64	71	#ifdef _WIN64
62	#define ECB_PTRSIZE 8	72	#define ECB_PTRSIZE 8
63	typedef uint64_t uintptr_t;	73	typedef uint64_t uintptr_t;
64	typedef int64_t intptr_t;	74	typedef int64_t intptr_t;
65	#else	75	#else
…		…
67	typedef uint32_t uintptr_t;	77	typedef uint32_t uintptr_t;
68	typedef int32_t intptr_t;	78	typedef int32_t intptr_t;
69	#endif	79	#endif
70	#else	80	#else
71	#include <inttypes.h>	81	#include <inttypes.h>
72	#if UINTMAX_MAX > 0xffffffffU	82	#if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
73	#define ECB_PTRSIZE 8	83	#define ECB_PTRSIZE 8
74	#else	84	#else
75	#define ECB_PTRSIZE 4	85	#define ECB_PTRSIZE 4
76	#endif	86	#endif
77	#endif	87	#endif
78		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
79	/* work around x32 idiocy by defining proper macros */	100	/* work around x32 idiocy by defining proper macros */
80	#if __amd64 || __x86_64 || _M_AMD64 || _M_X64	101	#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
81	#if _ILP32	102	#if _ILP32
82	#define ECB_AMD64_X32 1	103	#define ECB_AMD64_X32 1
83	#else	104	#else
84	#define ECB_AMD64 1	105	#define ECB_AMD64 1
85	#endif	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
86	#endif	113	#endif
87		114
88	/* many compilers define _GNUC_ to some versions but then only implement	115	/* many compilers define _GNUC_ to some versions but then only implement
89	* what their idiot authors think are the "more important" extensions,	116	* what their idiot authors think are the "more important" extensions,
90	* causing enormous grief in return for some better fake benchmark numbers.	117	* causing enormous grief in return for some better fake benchmark numbers.
…		…
98	#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)))
99	#endif	126	#endif
100		127
101	#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))	128	#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
102		129
103	#if __clang__ && defined(__has_builtin)	130	#if __clang__ && defined __has_builtin
104	#define ECB_CLANG_BUILTIN(x) __has_builtin(x)	131	#define ECB_CLANG_BUILTIN(x) __has_builtin (x)
105	#else	132	#else
106	#define ECB_CLANG_BUILTIN(x) 0	133	#define ECB_CLANG_BUILTIN(x) 0
107	#endif	134	#endif
108		135
109	#if __clang__ && defined(__has_extension)	136	#if __clang__ && defined __has_extension
110	#define ECB_CLANG_EXTENSION(x) __has_extension(x)	137	#define ECB_CLANG_EXTENSION(x) __has_extension (x)
111	#else	138	#else
112	#define ECB_CLANG_EXTENSION(x) 0	139	#define ECB_CLANG_EXTENSION(x) 0
113	#endif	140	#endif
114		141
115	#define ECB_CPP (__cplusplus+0)	142	#define ECB_CPP (__cplusplus+0)
116	#define ECB_CPP11 (__cplusplus >= 201103L)	143	#define ECB_CPP11 (__cplusplus >= 201103L)
		144	#define ECB_CPP14 (__cplusplus >= 201402L)
		145	#define ECB_CPP17 (__cplusplus >= 201703L)
117		146
118	#if ECB_CPP	147	#if ECB_CPP
119	#define ECB_C 0	148	#define ECB_C 0
120	#define ECB_STDC_VERSION 0	149	#define ECB_STDC_VERSION 0
121	#else	150	#else
…		…
123	#define ECB_STDC_VERSION __STDC_VERSION__	152	#define ECB_STDC_VERSION __STDC_VERSION__
124	#endif	153	#endif
125		154
126	#define ECB_C99 (ECB_STDC_VERSION >= 199901L)	155	#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
127	#define ECB_C11 (ECB_STDC_VERSION >= 201112L)	156	#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
		157	#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
128		158
129	#if ECB_CPP	159	#if ECB_CPP
130	#define ECB_EXTERN_C extern "C"	160	#define ECB_EXTERN_C extern "C"
131	#define ECB_EXTERN_C_BEG ECB_EXTERN_C {	161	#define ECB_EXTERN_C_BEG ECB_EXTERN_C {
132	#define ECB_EXTERN_C_END }	162	#define ECB_EXTERN_C_END }
…		…
147		177
148	#if ECB_NO_SMP	178	#if ECB_NO_SMP
149	#define ECB_MEMORY_FENCE do { } while (0)	179	#define ECB_MEMORY_FENCE do { } while (0)
150	#endif	180	#endif
151		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
152	#ifndef ECB_MEMORY_FENCE	191	#ifndef ECB_MEMORY_FENCE
153	#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")
154	#if __i386 || __i386__	194	#if __i386 || __i386__
155	#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")
156	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")	196	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
157	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")	197	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
158	#elif __amd64 || __amd64__ || __x86_64 || __x86_64__	198	#elif ECB_GCC_AMD64
159	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")	199	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
160	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")	200	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
161	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")	201	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
162	#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__	202	#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
163	#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 */
164	#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \	211	#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
165	|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__	212	|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
		213	|| defined __ARM_ARCH_6T2__
166	#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")
167	#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \	215	#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
168	|| defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__	216	|| defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
169	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")	217	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
170	#elif __aarch64__	218	#elif __aarch64__
171	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")	219	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
172	#elif (__sparc || __sparc__) && !__sparcv8	220	#elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
173	#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")
174	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")	222	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
175	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")	223	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
176	#elif defined __s390__ || defined __s390x__	224	#elif defined __s390__ || defined __s390x__
177	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")	225	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
…		…
200	#if ECB_GCC_VERSION(4,7)	248	#if ECB_GCC_VERSION(4,7)
201	/* see comment below (stdatomic.h) about the C11 memory model. */	249	/* see comment below (stdatomic.h) about the C11 memory model. */
202	#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)	250	#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
203	#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)	251	#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
204	#define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)	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)
205		255
206	#elif ECB_CLANG_EXTENSION(c_atomic)	256	#elif ECB_CLANG_EXTENSION(c_atomic)
207	/* see comment below (stdatomic.h) about the C11 memory model. */	257	/* see comment below (stdatomic.h) about the C11 memory model. */
208	#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)	258	#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
209	#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)	259	#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
210	#define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)	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)
211		263
212	#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__	264	#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
213	#define ECB_MEMORY_FENCE __sync_synchronize ()	265	#define ECB_MEMORY_FENCE __sync_synchronize ()
214	#elif _MSC_VER >= 1500 /* VC++ 2008 */	266	#elif _MSC_VER >= 1500 /* VC++ 2008 */
215	/* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */	267	/* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
…		…
225	#elif defined _WIN32	277	#elif defined _WIN32
226	#include <WinNT.h>	278	#include <WinNT.h>
227	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */	279	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
228	#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110	280	#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
229	#include <mbarrier.h>	281	#include <mbarrier.h>
230	#define ECB_MEMORY_FENCE __machine_rw_barrier ()	282	#define ECB_MEMORY_FENCE __machine_rw_barrier ()
231	#define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()	283	#define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
232	#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 ()
233	#elif __xlC__	286	#elif __xlC__
234	#define ECB_MEMORY_FENCE __sync ()	287	#define ECB_MEMORY_FENCE __sync ()
235	#endif	288	#endif
236	#endif	289	#endif
237		290
238	#ifndef ECB_MEMORY_FENCE	291	#ifndef ECB_MEMORY_FENCE
239	#if ECB_C11 && !defined __STDC_NO_ATOMICS__	292	#if ECB_C11 && !defined __STDC_NO_ATOMICS__
240	/* we assume that these memory fences work on all variables/all memory accesses, */	293	/* we assume that these memory fences work on all variables/all memory accesses, */
241	/* not just C11 atomics and atomic accesses */	294	/* not just C11 atomics and atomic accesses */
242	#include <stdatomic.h>	295	#include <stdatomic.h>
243	/* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
244	/* any fence other than seq_cst, which isn't very efficient for us. */
245	/* Why that is, we don't know - either the C11 memory model is quite useless */
246	/* for most usages, or gcc and clang have a bug */
247	/* I *currently* lean towards the latter, and inefficiently implement */
248	/* all three of ecb's fences as a seq_cst fence */
249	/* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
250	/* for all __atomic_thread_fence's except seq_cst */
251	#define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)	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)
252	#endif	299	#endif
253	#endif	300	#endif
254		301
255	#ifndef ECB_MEMORY_FENCE	302	#ifndef ECB_MEMORY_FENCE
256	#if !ECB_AVOID_PTHREADS	303	#if !ECB_AVOID_PTHREADS
…		…
276		323
277	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE	324	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
278	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE	325	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
279	#endif	326	#endif
280		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
281	/*****************************************************************************/	332	/*****************************************************************************/
282		333
283	#if __cplusplus	334	#if ECB_CPP
284	#define ecb_inline static inline	335	#define ecb_inline static inline
285	#elif ECB_GCC_VERSION(2,5)	336	#elif ECB_GCC_VERSION(2,5)
286	#define ecb_inline static __inline__	337	#define ecb_inline static __inline__
287	#elif ECB_C99	338	#elif ECB_C99
288	#define ecb_inline static inline	339	#define ecb_inline static inline
…		…
302		353
303	#define ECB_CONCAT_(a, b) a ## b	354	#define ECB_CONCAT_(a, b) a ## b
304	#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)	355	#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
305	#define ECB_STRINGIFY_(a) # a	356	#define ECB_STRINGIFY_(a) # a
306	#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))
307		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? */
308	#define ecb_function_ ecb_inline	364	#define ecb_function_ ecb_inline
309		365
310	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)	366	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
311	#define ecb_attribute(attrlist) __attribute__ (attrlist)	367	#define ecb_attribute(attrlist) __attribute__ (attrlist)
312	#else	368	#else
…		…
335	#define ecb_prefetch(addr,rw,locality)	391	#define ecb_prefetch(addr,rw,locality)
336	#endif	392	#endif
337		393
338	/* no emulation for ecb_decltype */	394	/* no emulation for ecb_decltype */
339	#if ECB_CPP11	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; };
340	#define ecb_decltype(x) decltype (x)	398	#define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
341	#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)	399	#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
342	#define ecb_decltype(x) __typeof__ (x)	400	#define ecb_decltype(x) __typeof__ (x)
343	#endif	401	#endif
344		402
345	#if _MSC_VER >= 1300	403	#if _MSC_VER >= 1300
346	#define ecb_deprecated __declspec(deprecated)	404	#define ecb_deprecated __declspec (deprecated)
347	#else	405	#else
348	#define ecb_deprecated ecb_attribute ((__deprecated__))	406	#define ecb_deprecated ecb_attribute ((__deprecated__))
349	#endif	407	#endif
350		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
351	#define ecb_noinline ecb_attribute ((__noinline__))	420	#define ecb_noinline ecb_attribute ((__noinline__))
		421	#endif
		422
352	#define ecb_unused ecb_attribute ((__unused__))	423	#define ecb_unused ecb_attribute ((__unused__))
353	#define ecb_const ecb_attribute ((__const__))	424	#define ecb_const ecb_attribute ((__const__))
354	#define ecb_pure ecb_attribute ((__pure__))	425	#define ecb_pure ecb_attribute ((__pure__))
355		426
356	/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */	427	#if ECB_C11 || __IBMC_NORETURN
357	#if ECB_C11	428	/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
358	#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)
359	#else	435	#else
360	#define ecb_noreturn ecb_attribute ((__noreturn__))	436	#define ecb_noreturn ecb_attribute ((__noreturn__))
361	#endif	437	#endif
362		438
363	#if ECB_GCC_VERSION(4,3)	439	#if ECB_GCC_VERSION(4,3)
…		…
382	/* count trailing zero bits and count # of one bits */	458	/* count trailing zero bits and count # of one bits */
383	#if ECB_GCC_VERSION(3,4) \	459	#if ECB_GCC_VERSION(3,4) \
384	|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \	460	|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
385	&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \	461	&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
386	&& ECB_CLANG_BUILTIN(__builtin_popcount))	462	&& ECB_CLANG_BUILTIN(__builtin_popcount))
387	/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
388	#define ecb_ld32(x) (__builtin_clz (x) ^ 31)
389	#define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
390	#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))
391	#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)
392	#define ecb_popcount32(x) __builtin_popcount (x)	469	#define ecb_popcount32(x) __builtin_popcount (x)
393	/* no popcountll */	470	/* ecb_popcount64 is more difficult, see below */
394	#else	471	#else
395	ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;	472	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
396	ecb_function_ int	473	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x)
397	ecb_ctz32 (uint32_t x)
398	{	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
399	int r = 0;	480	int r;
400		481
401	x &= ~x + 1; /* this isolates the lowest bit */	482	x &= ~x + 1; /* this isolates the lowest bit */
402		483
403	#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;
404	r += !!(x & 0xaaaaaaaa) << 0;	503	r += !!(x & 0xaaaaaaaa) << 0;
405	r += !!(x & 0xcccccccc) << 1;	504	r += !!(x & 0xcccccccc) << 1;
406	r += !!(x & 0xf0f0f0f0) << 2;	505	r += !!(x & 0xf0f0f0f0) << 2;
407	r += !!(x & 0xff00ff00) << 3;	506	r += !!(x & 0xff00ff00) << 3;
408	r += !!(x & 0xffff0000) << 4;	507	r += !!(x & 0xffff0000) << 4;
409	#else	508	#else /* branchless on modern compilers, typically */
		509	r = 0;
410	if (x & 0xaaaaaaaa) r += 1;	510	if (x & 0xaaaaaaaa) r += 1;
411	if (x & 0xcccccccc) r += 2;	511	if (x & 0xcccccccc) r += 2;
412	if (x & 0xf0f0f0f0) r += 4;	512	if (x & 0xf0f0f0f0) r += 4;
413	if (x & 0xff00ff00) r += 8;	513	if (x & 0xff00ff00) r += 8;
414	if (x & 0xffff0000) r += 16;	514	if (x & 0xffff0000) r += 16;
415	#endif	515	#endif
416		516
417	return r;	517	return r;
		518	#endif
418	}	519	}
419		520
420	ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;	521	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
421	ecb_function_ int	522	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x)
422	ecb_ctz64 (uint64_t x)
423	{	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
424	int shift = x & 0xffffffffU ? 0 : 32;	529	int shift = x & 0xffffffff ? 0 : 32;
425	return ecb_ctz32 (x >> shift) + shift;	530	return ecb_ctz32 (x >> shift) + shift;
		531	#endif
426	}	532	}
427		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
428	ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;	583	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
429	ecb_function_ int	584	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x)
430	ecb_popcount32 (uint32_t x)
431	{	585	{
432	x -= (x >> 1) & 0x55555555;	586	x -= (x >> 1) & 0x55555555;
433	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);	587	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
434	x = ((x >> 4) + x) & 0x0f0f0f0f;	588	x = ((x >> 4) + x) & 0x0f0f0f0f;
435	x *= 0x01010101;	589	x *= 0x01010101;
436		590
437	return x >> 24;	591	return x >> 24;
438	}	592	}
439		593
440	ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;	594	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
441	ecb_function_ int ecb_ld32 (uint32_t x)	595	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
442	{	596	{
		597	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		598	unsigned long r;
		599	_BitScanReverse (&r, x);
		600	return (int)r;
		601	#else
443	int r = 0;	602	int r = 0;
444		603
445	if (x >> 16) { x >>= 16; r += 16; }	604	if (x >> 16) { x >>= 16; r += 16; }
446	if (x >> 8) { x >>= 8; r += 8; }	605	if (x >> 8) { x >>= 8; r += 8; }
447	if (x >> 4) { x >>= 4; r += 4; }	606	if (x >> 4) { x >>= 4; r += 4; }
448	if (x >> 2) { x >>= 2; r += 2; }	607	if (x >> 2) { x >>= 2; r += 2; }
449	if (x >> 1) { r += 1; }	608	if (x >> 1) { r += 1; }
450		609
451	return r;	610	return r;
		611	#endif
452	}	612	}
453		613
454	ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;	614	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
455	ecb_function_ int ecb_ld64 (uint64_t x)	615	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
456	{	616	{
		617	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		618	unsigned long r;
		619	_BitScanReverse64 (&r, x);
		620	return (int)r;
		621	#else
457	int r = 0;	622	int r = 0;
458		623
459	if (x >> 32) { x >>= 32; r += 32; }	624	if (x >> 32) { x >>= 32; r += 32; }
460		625
461	return r + ecb_ld32 (x);	626	return r + ecb_ld32 (x);
462	}
463	#endif	627	#endif
		628	}
		629	#endif
464		630
465	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;	631	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
466	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }	632	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
467	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;	633	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
468	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }	634	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
469		635
470	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;	636	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
471	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)	637	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
472	{	638	{
473	return ( (x * 0x0802U & 0x22110U)	639	return ( (x * 0x0802U & 0x22110U)
474	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;	640	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
475	}	641	}
476		642
477	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;	643	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
478	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)	644	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
479	{	645	{
480	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);	646	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
481	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);	647	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
482	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);	648	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
483	x = ( x >> 8 ) | ( x << 8);	649	x = ( x >> 8 ) | ( x << 8);
484		650
485	return x;	651	return x;
486	}	652	}
487		653
488	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;	654	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
489	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)	655	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
490	{	656	{
491	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);	657	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
492	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);	658	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
493	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);	659	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
494	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);	660	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
495	x = ( x >> 16 ) | ( x << 16);	661	x = ( x >> 16 ) | ( x << 16);
496		662
497	return x;	663	return x;
498	}	664	}
499		665
		666	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
		667	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x)
		668	{
500	/* popcount64 is only available on 64 bit cpus as gcc builtin */	669	/* popcount64 is only available on 64 bit cpus as gcc builtin. */
501	/* so for this version we are lazy */	670	/* also, gcc/clang make this surprisingly difficult to use */
502	ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;	671	#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
503	ecb_function_ int	672	return __builtin_popcountl (x);
504	ecb_popcount64 (uint64_t x)	673	#else
505	{
506	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);	674	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
		675	#endif
507	}	676	}
508		677
509	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
510	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
511	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
512	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
513	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
514	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
515	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
516	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
517
518	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }	678	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
519	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }	679	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
520	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }	680	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
521	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }	681	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
522	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }	682	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
523	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }	683	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
524	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }	684	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
525	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }	685	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
		686
		687	#if ECB_CPP
		688
		689	inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
		690	inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
		691	inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
		692	inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
		693
		694	inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
		695	inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
		696	inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
		697	inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
		698
		699	inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
		700	inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
		701	inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
		702	inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
		703
		704	inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
		705	inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
		706	inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
		707	inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
		708
		709	inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
		710	inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
		711	inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
		712
		713	inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
		714	inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
		715	inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
		716	inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
		717
		718	inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
		719	inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
		720	inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
		721	inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
		722
		723	#endif
526		724
527	#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))	725	#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
		726	#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
		727	#define ecb_bswap16(x) __builtin_bswap16 (x)
		728	#else
528	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)	729	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
		730	#endif
529	#define ecb_bswap32(x) __builtin_bswap32 (x)	731	#define ecb_bswap32(x) __builtin_bswap32 (x)
530	#define ecb_bswap64(x) __builtin_bswap64 (x)	732	#define ecb_bswap64(x) __builtin_bswap64 (x)
		733	#elif _MSC_VER
		734	#include <stdlib.h>
		735	#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
		736	#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
		737	#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
531	#else	738	#else
532	ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;	739	ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
533	ecb_function_ uint16_t	740	ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x)
534	ecb_bswap16 (uint16_t x)
535	{	741	{
536	return ecb_rotl16 (x, 8);	742	return ecb_rotl16 (x, 8);
537	}	743	}
538		744
539	ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;	745	ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
540	ecb_function_ uint32_t	746	ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x)
541	ecb_bswap32 (uint32_t x)
542	{	747	{
543	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);	748	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
544	}	749	}
545		750
546	ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;	751	ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
547	ecb_function_ uint64_t	752	ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x)
548	ecb_bswap64 (uint64_t x)
549	{	753	{
550	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);	754	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
551	}	755	}
552	#endif	756	#endif
553		757
554	#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)	758	#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
555	#define ecb_unreachable() __builtin_unreachable ()	759	#define ecb_unreachable() __builtin_unreachable ()
556	#else	760	#else
557	/* this seems to work fine, but gcc always emits a warning for it :/ */	761	/* this seems to work fine, but gcc always emits a warning for it :/ */
558	ecb_inline void ecb_unreachable (void) ecb_noreturn;	762	ecb_inline ecb_noreturn void ecb_unreachable (void);
559	ecb_inline void ecb_unreachable (void) { }	763	ecb_inline ecb_noreturn void ecb_unreachable (void) { }
560	#endif	764	#endif
561		765
562	/* try to tell the compiler that some condition is definitely true */	766	/* try to tell the compiler that some condition is definitely true */
563	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0	767	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
564		768
565	ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;	769	ecb_inline uint32_t ecb_byteorder_helper (void);
566	ecb_inline unsigned char	770	ecb_inline uint32_t ecb_byteorder_helper (void)
567	ecb_byteorder_helper (void)
568	{	771	{
569	/* the union code still generates code under pressure in gcc, */	772	/* the union code still generates code under pressure in gcc, */
570	/* but less than using pointers, and always seems to */	773	/* but less than using pointers, and always seems to */
571	/* successfully return a constant. */	774	/* successfully return a constant. */
572	/* the reason why we have this horrible preprocessor mess */	775	/* the reason why we have this horrible preprocessor mess */
573	/* is to avoid it in all cases, at least on common architectures */	776	/* is to avoid it in all cases, at least on common architectures */
574	/* or when using a recent enough gcc version (>= 4.6) */	777	/* or when using a recent enough gcc version (>= 4.6) */
575	#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
576	return 0x44;
577	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__	778	#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
		779	|| ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
		780	#define ECB_LITTLE_ENDIAN 1
578	return 0x44;	781	return 0x44332211;
579	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__	782	#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
		783	|| ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
		784	#define ECB_BIG_ENDIAN 1
580	return 0x11;	785	return 0x11223344;
581	#else	786	#else
582	union	787	union
583	{	788	{
		789	uint8_t c[4];
584	uint32_t i;	790	uint32_t u;
585	uint8_t c;
586	} u = { 0x11223344 };	791	} u = { 0x11, 0x22, 0x33, 0x44 };
587	return u.c;	792	return u.u;
588	#endif	793	#endif
589	}	794	}
590		795
591	ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
592	ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }	796	ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
593	ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
594	ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }	797	ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
		798
		799	/*****************************************************************************/
		800	/* unaligned load/store */
		801
		802	ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		803	ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		804	ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		805
		806	ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		807	ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		808	ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		809
		810	ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		811	ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		812	ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		813
		814	ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
		815	ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
		816	ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
		817
		818	ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
		819	ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
		820	ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
		821
		822	ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		823	ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		824	ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		825
		826	ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		827	ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		828	ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		829
		830	ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
		831	ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
		832	ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
		833
		834	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)); }
		835	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)); }
		836	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)); }
		837
		838	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)); }
		839	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)); }
		840	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)); }
		841
		842	#if ECB_CPP
		843
		844	inline uint8_t ecb_bswap (uint8_t v) { return v; }
		845	inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
		846	inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
		847	inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
		848
		849	template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		850	template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		851	template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
		852	template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
		853	template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
		854	template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
		855	template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
		856	template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
		857
		858	template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		859	template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		860	template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
		861	template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
		862	template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
		863	template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
		864	template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
		865	template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
		866
		867	#endif
		868
		869	/*****************************************************************************/
		870	/* pointer/integer hashing */
		871
		872	/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
		873	ecb_function_ ecb_const uint32_t ecb_mix32 (uint32_t v);
		874	ecb_function_ ecb_const uint32_t ecb_mix32 (uint32_t v)
		875	{
		876	v ^= v >> 16; v *= 0x7feb352dU;
		877	v ^= v >> 15; v *= 0x846ca68bU;
		878	v ^= v >> 16;
		879	return v;
		880	}
		881
		882	ecb_function_ ecb_const uint32_t ecb_unmix32 (uint32_t v);
		883	ecb_function_ ecb_const uint32_t ecb_unmix32 (uint32_t v)
		884	{
		885	v ^= v >> 16 ; v *= 0x43021123U;
		886	v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
		887	v ^= v >> 16 ;
		888	return v;
		889	}
		890
		891	/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
		892	ecb_function_ ecb_const uint64_t ecb_mix64 (uint64_t v);
		893	ecb_function_ ecb_const uint64_t ecb_mix64 (uint64_t v)
		894	{
		895	v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
		896	v ^= v >> 27; v *= 0x94d049bb133111ebU;
		897	v ^= v >> 31;
		898	return v;
		899	}
		900
		901	ecb_function_ ecb_const uint64_t ecb_unmix64 (uint64_t v);
		902	ecb_function_ ecb_const uint64_t ecb_unmix64 (uint64_t v)
		903	{
		904	v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
		905	v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
		906	v ^= v >> 30 ^ v >> 60;
		907	return v;
		908	}
		909
		910	ecb_function_ ecb_const uintptr_t ecb_ptrmix (void *p);
		911	ecb_function_ ecb_const uintptr_t ecb_ptrmix (void *p)
		912	{
		913	#if ECB_PTRSIZE <= 4
		914	return ecb_mix32 ((uint32_t)p);
		915	#else
		916	return ecb_mix64 ((uint64_t)p);
		917	#endif
		918	}
		919
		920	ecb_function_ ecb_const void *ecb_ptrunmix (uintptr_t v);
		921	ecb_function_ ecb_const void *ecb_ptrunmix (uintptr_t v)
		922	{
		923	#if ECB_PTRSIZE <= 4
		924	return (void *)ecb_unmix32 (v);
		925	#else
		926	return (void *)ecb_unmix64 (v);
		927	#endif
		928	}
		929
		930	#if ECB_CPP
		931
		932	template<typename T>
		933	inline uintptr_t ecb_ptrmix (T *p)
		934	{
		935	return ecb_ptrmix (static_cast<void *>(p));
		936	}
		937
		938	template<typename T>
		939	inline T *ecb_ptrunmix (uintptr_t v)
		940	{
		941	return static_cast<T *>(ecb_ptrunmix (v));
		942	}
		943
		944	#endif
		945
		946	/*****************************************************************************/
		947	/* gray code */
		948
		949	ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); }
		950	ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); }
		951	ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); }
		952	ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); }
		953
		954	ecb_function_ ecb_const uint8_t ecb_gray_decode8 (uint8_t g);
		955	ecb_function_ ecb_const uint8_t ecb_gray_decode8 (uint8_t g)
		956	{
		957	g ^= g >> 1;
		958	g ^= g >> 2;
		959	g ^= g >> 4;
		960
		961	return g;
		962	}
		963
		964	ecb_function_ ecb_const uint16_t ecb_gray_decode16 (uint16_t g);
		965	ecb_function_ ecb_const uint16_t ecb_gray_decode16 (uint16_t g)
		966	{
		967	g ^= g >> 1;
		968	g ^= g >> 2;
		969	g ^= g >> 4;
		970	g ^= g >> 8;
		971
		972	return g;
		973	}
		974
		975	ecb_function_ ecb_const uint32_t ecb_gray_decode32 (uint32_t g);
		976	ecb_function_ ecb_const uint32_t ecb_gray_decode32 (uint32_t g)
		977	{
		978	g ^= g >> 1;
		979	g ^= g >> 2;
		980	g ^= g >> 4;
		981	g ^= g >> 8;
		982	g ^= g >> 16;
		983
		984	return g;
		985	}
		986
		987	ecb_function_ ecb_const uint64_t ecb_gray_decode64 (uint64_t g);
		988	ecb_function_ ecb_const uint64_t ecb_gray_decode64 (uint64_t g)
		989	{
		990	g ^= g >> 1;
		991	g ^= g >> 2;
		992	g ^= g >> 4;
		993	g ^= g >> 8;
		994	g ^= g >> 16;
		995	g ^= g >> 32;
		996
		997	return g;
		998	}
		999
		1000	#if ECB_CPP
		1001
		1002	ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); }
		1003	ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); }
		1004	ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); }
		1005	ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); }
		1006
		1007	ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); }
		1008	ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); }
		1009	ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); }
		1010	ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); }
		1011
		1012	#endif
		1013
		1014	/*****************************************************************************/
		1015	/* 2d hilbert curves */
		1016
		1017	/* algorithm from the book Hacker's Delight, modified to not */
		1018	/* run into undefined behaviour for n==16 */
		1019	ecb_function_ ecb_const uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s);
		1020	ecb_function_ ecb_const uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
		1021	{
		1022	uint32_t comp, swap, cs, t, sr;
		1023
		1024	/* pad s on the left (unused) bits with 01 (no change groups) */
		1025	s |= 0x55555555U << n << n;
		1026	/* "s shift right" */
		1027	sr = (s >> 1) & 0x55555555U;
		1028	/* compute complement and swap info in two-bit groups */
		1029	cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
		1030
		1031	/* parallel prefix xor op to propagate both complement
		1032	* and swap info together from left to right (there is
		1033	* no step "cs ^= cs >> 1", so in effect it computes
		1034	* two independent parallel prefix operations on two
		1035	* interleaved sets of sixteen bits).
		1036	*/
		1037	cs ^= cs >> 2;
		1038	cs ^= cs >> 4;
		1039	cs ^= cs >> 8;
		1040	cs ^= cs >> 16;
		1041
		1042	/* separate swap and complement bits */
		1043	swap = cs & 0x55555555U;
		1044	comp = (cs >> 1) & 0x55555555U;
		1045
		1046	/* calculate coordinates in odd and even bit positions */
		1047	t = (s & swap) ^ comp;
		1048	s = s ^ sr ^ t ^ (t << 1);
		1049
		1050	/* unpad/clear out any junk on the left */
		1051	s = s & ((1 << n << n) - 1);
		1052
		1053	/* Now "unshuffle" to separate the x and y bits. */
		1054	t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
		1055	t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
		1056	t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
		1057	t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
		1058
		1059	/* now s contains two 16-bit coordinates */
		1060	return s;
		1061	}
		1062
		1063	/* 64 bit, a straightforward extension to the 32 bit case */
		1064	ecb_function_ ecb_const uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s);
		1065	ecb_function_ ecb_const uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
		1066	{
		1067	uint64_t comp, swap, cs, t, sr;
		1068
		1069	/* pad s on the left (unused) bits with 01 (no change groups) */
		1070	s |= 0x5555555555555555U << n << n;
		1071	/* "s shift right" */
		1072	sr = (s >> 1) & 0x5555555555555555U;
		1073	/* compute complement and swap info in two-bit groups */
		1074	cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
		1075
		1076	/* parallel prefix xor op to propagate both complement
		1077	* and swap info together from left to right (there is
		1078	* no step "cs ^= cs >> 1", so in effect it computes
		1079	* two independent parallel prefix operations on two
		1080	* interleaved sets of thirty-two bits).
		1081	*/
		1082	cs ^= cs >> 2;
		1083	cs ^= cs >> 4;
		1084	cs ^= cs >> 8;
		1085	cs ^= cs >> 16;
		1086	cs ^= cs >> 32;
		1087
		1088	/* separate swap and complement bits */
		1089	swap = cs & 0x5555555555555555U;
		1090	comp = (cs >> 1) & 0x5555555555555555U;
		1091
		1092	/* calculate coordinates in odd and even bit positions */
		1093	t = (s & swap) ^ comp;
		1094	s = s ^ sr ^ t ^ (t << 1);
		1095
		1096	/* unpad/clear out any junk on the left */
		1097	s = s & ((1 << n << n) - 1);
		1098
		1099	/* Now "unshuffle" to separate the x and y bits. */
		1100	t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
		1101	t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
		1102	t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
		1103	t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
		1104	t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
		1105
		1106	/* now s contains two 32-bit coordinates */
		1107	return s;
		1108	}
		1109
		1110	/* algorithm from the book Hacker's Delight, but a similar algorithm*/
		1111	/* is given in https://doi.org/10.1002/spe.4380160103 */
		1112	/* this has been slightly improved over the original version */
		1113	ecb_function_ ecb_const uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy);
		1114	ecb_function_ ecb_const uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
		1115	{
		1116	uint32_t row;
		1117	uint32_t state = 0;
		1118	uint32_t s = 0;
		1119
		1120	do
		1121	{
		1122	--n;
		1123
		1124	row = 4 * state
		1125	| (2 & (xy >> n >> 15))
		1126	| (1 & (xy >> n ));
		1127
		1128	/* these funky constants are lookup tables for two-bit values */
		1129	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1130	state = (0x8fe65831U >> 2 * row) & 3;
		1131	}
		1132	while (n > 0);
		1133
		1134	return s;
		1135	}
		1136
		1137	/* 64 bit, essentially the same as 32 bit */
		1138	ecb_function_ ecb_const uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy);
		1139	ecb_function_ ecb_const uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
		1140	{
		1141	uint32_t row;
		1142	uint32_t state = 0;
		1143	uint64_t s = 0;
		1144
		1145	do
		1146	{
		1147	--n;
		1148
		1149	row = 4 * state
		1150	| (2 & (xy >> n >> 31))
		1151	| (1 & (xy >> n ));
		1152
		1153	/* these funky constants are lookup tables for two-bit values */
		1154	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1155	state = (0x8fe65831U >> 2 * row) & 3;
		1156	}
		1157	while (n > 0);
		1158
		1159	return s;
		1160	}
		1161
		1162	/*****************************************************************************/
		1163	/* division */
595		1164
596	#if ECB_GCC_VERSION(3,0) || ECB_C99	1165	#if ECB_GCC_VERSION(3,0) || ECB_C99
		1166	/* C99 tightened the definition of %, so we can use a more efficient version */
597	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))	1167	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
598	#else	1168	#else
599	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))	1169	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
600	#endif	1170	#endif
601		1171
602	#if __cplusplus	1172	#if ECB_CPP
603	template<typename T>	1173	template<typename T>
604	static inline T ecb_div_rd (T val, T div)	1174	static inline T ecb_div_rd (T val, T div)
605	{	1175	{
606	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;	1176	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
607	}	1177	}
…		…
612	}	1182	}
613	#else	1183	#else
614	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))	1184	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
615	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))	1185	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
616	#endif	1186	#endif
		1187
		1188	/*****************************************************************************/
		1189	/* array length */
617		1190
618	#if ecb_cplusplus_does_not_suck	1191	#if ecb_cplusplus_does_not_suck
619	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */	1192	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
620	template<typename T, int N>	1193	template<typename T, int N>
621	static inline int ecb_array_length (const T (&arr)[N])	1194	static inline int ecb_array_length (const T (&arr)[N])
…		…
624	}	1197	}
625	#else	1198	#else
626	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))	1199	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
627	#endif	1200	#endif
628		1201
		1202	/*****************************************************************************/
		1203	/* IEEE 754-2008 half float conversions */
		1204
		1205	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
		1206	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x)
		1207	{
		1208	unsigned int s = (x & 0x8000) << (31 - 15);
		1209	int e = (x >> 10) & 0x001f;
		1210	unsigned int m = x & 0x03ff;
		1211
		1212	if (ecb_expect_false (e == 31))
		1213	/* infinity or NaN */
		1214	e = 255 - (127 - 15);
		1215	else if (ecb_expect_false (!e))
		1216	{
		1217	if (ecb_expect_true (!m))
		1218	/* zero, handled by code below by forcing e to 0 */
		1219	e = 0 - (127 - 15);
		1220	else
		1221	{
		1222	/* subnormal, renormalise */
		1223	unsigned int s = 10 - ecb_ld32 (m);
		1224
		1225	m = (m << s) & 0x3ff; /* mask implicit bit */
		1226	e -= s - 1;
		1227	}
		1228	}
		1229
		1230	/* e and m now are normalised, or zero, (or inf or nan) */
		1231	e += 127 - 15;
		1232
		1233	return s | (e << 23) | (m << (23 - 10));
		1234	}
		1235
		1236	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
		1237	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x)
		1238	{
		1239	unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
		1240	int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
		1241	unsigned int m = x & 0x007fffff;
		1242
		1243	x &= 0x7fffffff;
		1244
		1245	/* if it's within range of binary16 normals, use fast path */
		1246	if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
		1247	{
		1248	/* mantissa round-to-even */
		1249	m += 0x00000fff + ((m >> (23 - 10)) & 1);
		1250
		1251	/* handle overflow */
		1252	if (ecb_expect_false (m >= 0x00800000))
		1253	{
		1254	m >>= 1;
		1255	e += 1;
		1256	}
		1257
		1258	return s | (e << 10) | (m >> (23 - 10));
		1259	}
		1260
		1261	/* handle large numbers and infinity */
		1262	if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
		1263	return s | 0x7c00;
		1264
		1265	/* handle zero, subnormals and small numbers */
		1266	if (ecb_expect_true (x < 0x38800000))
		1267	{
		1268	/* zero */
		1269	if (ecb_expect_true (!x))
		1270	return s;
		1271
		1272	/* handle subnormals */
		1273
		1274	/* too small, will be zero */
		1275	if (e < (14 - 24)) /* might not be sharp, but is good enough */
		1276	return s;
		1277
		1278	m |= 0x00800000; /* make implicit bit explicit */
		1279
		1280	/* very tricky - we need to round to the nearest e (+10) bit value */
		1281	{
		1282	unsigned int bits = 14 - e;
		1283	unsigned int half = (1 << (bits - 1)) - 1;
		1284	unsigned int even = (m >> bits) & 1;
		1285
		1286	/* if this overflows, we will end up with a normalised number */
		1287	m = (m + half + even) >> bits;
		1288	}
		1289
		1290	return s | m;
		1291	}
		1292
		1293	/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
		1294	m >>= 13;
		1295
		1296	return s | 0x7c00 | m | !m;
		1297	}
		1298
		1299	/*******************************************************************************/
		1300	/* fast integer to ascii */
		1301
		1302	/*
		1303	* This code is pretty complicated because it is general. The idea behind it,
		1304	* however, is pretty simple: first, the number is multiplied with a scaling
		1305	* factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
		1306	* number with the first digit in the upper bits.
		1307	* Then this digit is converted to text and masked out. The resulting number
		1308	* is then multiplied by 10, by multiplying the fixed point representation
		1309	* by 5 and shifting the (binary) decimal point one to the right, so a 4.28
		1310	* format becomes 5.27, 6.26 and so on.
		1311	* The rest involves only advancing the pointer if we already generated a
		1312	* non-zero digit, so leading zeroes are overwritten.
		1313	*/
		1314
		1315	/* simply return a mask with "bits" bits set */
		1316	#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
		1317
		1318	/* oputput a single digit. maskvalue is 10**digitidx */
		1319	#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
		1320	if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
		1321	{ \
		1322	char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
		1323	*ptr = digit + '0'; /* output it */ \
		1324	nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
		1325	ptr += nz; /* output digit only if non-zero digit seen */ \
		1326	x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
		1327	}
		1328
		1329	/* convert integer to fixed point format and multiply out digits, highest first */
		1330	/* requires magic constants: max. digits and number of bits after the decimal point */
		1331	#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
		1332	ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
		1333	{ \
		1334	char nz = lz; /* non-zero digit seen? */ \
		1335	/* convert to x.bits fixed-point */ \
		1336	type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
		1337	/* output up to 10 digits */ \
		1338	ecb_i2a_digit (type,bits,digitmask, 1, 0); \
		1339	ecb_i2a_digit (type,bits,digitmask, 10, 1); \
		1340	ecb_i2a_digit (type,bits,digitmask, 100, 2); \
		1341	ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
		1342	ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
		1343	ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
		1344	ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
		1345	ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
		1346	ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
		1347	ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
		1348	return ptr; \
		1349	}
		1350
		1351	/* predefined versions of the above, for various digits */
		1352	/* ecb_i2a_xN = almost N digits, limit defined by macro */
		1353	/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
		1354	/* ecb_i2a_0N = exactly N digits, including leading zeroes */
		1355
		1356	/* non-leading-zero versions, limited range */
		1357	#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
		1358	#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
		1359	ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
		1360	ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
		1361
		1362	/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
		1363	ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
		1364	ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
		1365	ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
		1366	ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
		1367	ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
		1368	ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
		1369	ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
		1370	ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
		1371
		1372	/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
		1373	ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
		1374	ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
		1375	ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
		1376	ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
		1377	ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
		1378	ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
		1379	ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
		1380	ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
		1381
		1382	#define ECB_I2A_I32_DIGITS 11
		1383	#define ECB_I2A_U32_DIGITS 10
		1384	#define ECB_I2A_I64_DIGITS 20
		1385	#define ECB_I2A_U64_DIGITS 21
		1386	#define ECB_I2A_MAX_DIGITS 21
		1387
		1388	ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u);
		1389	ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u)
		1390	{
		1391	#if ECB_64BIT_NATIVE
		1392	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1393	ptr = ecb_i2a_x10 (ptr, u);
		1394	else /* x10 almost, but not fully, covers 32 bit */
		1395	{
		1396	uint32_t u1 = u % 1000000000;
		1397	uint32_t u2 = u / 1000000000;
		1398
		1399	*ptr++ = u2 + '0';
		1400	ptr = ecb_i2a_09 (ptr, u1);
		1401	}
		1402	#else
		1403	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1404	ecb_i2a_x5 (ptr, u);
		1405	else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
		1406	{
		1407	uint32_t u1 = u % 10000;
		1408	uint32_t u2 = u / 10000;
		1409
		1410	ptr = ecb_i2a_x5 (ptr, u2);
		1411	ptr = ecb_i2a_04 (ptr, u1);
		1412	}
		1413	else
		1414	{
		1415	uint32_t u1 = u % 10000;
		1416	uint32_t ua = u / 10000;
		1417	uint32_t u2 = ua % 10000;
		1418	uint32_t u3 = ua / 10000;
		1419
		1420	ptr = ecb_i2a_2 (ptr, u3);
		1421	ptr = ecb_i2a_04 (ptr, u2);
		1422	ptr = ecb_i2a_04 (ptr, u1);
		1423	}
		1424	#endif
		1425
		1426	return ptr;
		1427	}
		1428
		1429	ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v);
		1430	ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v)
		1431	{
		1432	*ptr = '-'; ptr += v < 0;
		1433	uint32_t u = v < 0 ? -(uint32_t)v : v;
		1434
		1435	#if ECB_64BIT_NATIVE
		1436	ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
		1437	#else
		1438	ptr = ecb_i2a_u32 (ptr, u);
		1439	#endif
		1440
		1441	return ptr;
		1442	}
		1443
		1444	ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u);
		1445	ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u)
		1446	{
		1447	#if ECB_64BIT_NATIVE
		1448	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1449	ptr = ecb_i2a_x10 (ptr, u);
		1450	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1451	{
		1452	uint64_t u1 = u % 1000000000;
		1453	uint64_t u2 = u / 1000000000;
		1454
		1455	ptr = ecb_i2a_x10 (ptr, u2);
		1456	ptr = ecb_i2a_09 (ptr, u1);
		1457	}
		1458	else
		1459	{
		1460	uint64_t u1 = u % 1000000000;
		1461	uint64_t ua = u / 1000000000;
		1462	uint64_t u2 = ua % 1000000000;
		1463	uint64_t u3 = ua / 1000000000;
		1464
		1465	ptr = ecb_i2a_2 (ptr, u3);
		1466	ptr = ecb_i2a_09 (ptr, u2);
		1467	ptr = ecb_i2a_09 (ptr, u1);
		1468	}
		1469	#else
		1470	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1471	ptr = ecb_i2a_x5 (ptr, u);
		1472	else
		1473	{
		1474	uint64_t u1 = u % 10000;
		1475	uint64_t u2 = u / 10000;
		1476
		1477	ptr = ecb_i2a_u64 (ptr, u2);
		1478	ptr = ecb_i2a_04 (ptr, u1);
		1479	}
		1480	#endif
		1481
		1482	return ptr;
		1483	}
		1484
		1485	ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v);
		1486	ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v)
		1487	{
		1488	*ptr = '-'; ptr += v < 0;
		1489	uint64_t u = v < 0 ? -(uint64_t)v : v;
		1490
		1491	#if ECB_64BIT_NATIVE
		1492	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1493	ptr = ecb_i2a_x10 (ptr, u);
		1494	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1495	{
		1496	uint64_t u1 = u % 1000000000;
		1497	uint64_t u2 = u / 1000000000;
		1498
		1499	ptr = ecb_i2a_x10 (ptr, u2);
		1500	ptr = ecb_i2a_09 (ptr, u1);
		1501	}
		1502	else
		1503	{
		1504	uint64_t u1 = u % 1000000000;
		1505	uint64_t ua = u / 1000000000;
		1506	uint64_t u2 = ua % 1000000000;
		1507	uint64_t u3 = ua / 1000000000;
		1508
		1509	/* 2**31 is 19 digits, so the top is exactly one digit */
		1510	*ptr++ = u3 + '0';
		1511	ptr = ecb_i2a_09 (ptr, u2);
		1512	ptr = ecb_i2a_09 (ptr, u1);
		1513	}
		1514	#else
		1515	ptr = ecb_i2a_u64 (ptr, u);
		1516	#endif
		1517
		1518	return ptr;
		1519	}
		1520
629	/*******************************************************************************/	1521	/*******************************************************************************/
630	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */	1522	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
631		1523
632	/* basically, everything uses "ieee pure-endian" floating point numbers */	1524	/* basically, everything uses "ieee pure-endian" floating point numbers */
633	/* the only noteworthy exception is ancient armle, which uses order 43218765 */	1525	/* the only noteworthy exception is ancient armle, which uses order 43218765 */
634	#if 0 \	1526	#if 0 \
635	|| __i386 || __i386__ \	1527	|| __i386 || __i386__ \
636	|| __amd64 || __amd64__ || __x86_64 || __x86_64__ \	1528	|| ECB_GCC_AMD64 \
637	|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \	1529	|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
638	|| defined __s390__ || defined __s390x__ \	1530	|| defined __s390__ || defined __s390x__ \
639	|| defined __mips__ \	1531	|| defined __mips__ \
640	|| defined __alpha__ \	1532	|| defined __alpha__ \
641	|| defined __hppa__ \	1533	|| defined __hppa__ \
642	|| defined __ia64__ \	1534	|| defined __ia64__ \
643	|| defined __m68k__ \	1535	|| defined __m68k__ \
644	|| defined __m88k__ \	1536	|| defined __m88k__ \
645	|| defined __sh__ \	1537	|| defined __sh__ \
646	|| defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \	1538	|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
647	|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \	1539	|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
648	|| defined __aarch64__	1540	|| defined __aarch64__
649	#define ECB_STDFP 1	1541	#define ECB_STDFP 1
650	#include <string.h> /* for memcpy */
651	#else	1542	#else
652	#define ECB_STDFP 0	1543	#define ECB_STDFP 0
653	#endif	1544	#endif
654		1545
655	#ifndef ECB_NO_LIBM	1546	#ifndef ECB_NO_LIBM
…		…
667	#define ECB_NAN NAN	1558	#define ECB_NAN NAN
668	#else	1559	#else
669	#define ECB_NAN ECB_INFINITY	1560	#define ECB_NAN ECB_INFINITY
670	#endif	1561	#endif
671		1562
672	/* converts an ieee half/binary16 to a float */	1563	#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
673	ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;	1564	#define ecb_ldexpf(x,e) ldexpf ((x), (e))
674	ecb_function_ float	1565	#define ecb_frexpf(x,e) frexpf ((x), (e))
675	ecb_binary16_to_float (uint16_t x)	1566	#else
676	{	1567	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
677	int e = (x >> 10) & 0x1f;	1568	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
678	int m = x & 0x3ff;	1569	#endif
679	float r;
680
681	if (!e ) r = ldexpf (m , -24);
682	else if (e != 31) r = ldexpf (m + 0x400, e - 25);
683	else if (m ) r = ECB_NAN;
684	else r = ECB_INFINITY;
685
686	return x & 0x8000 ? -r : r;
687	}
688		1570
689	/* convert a float to ieee single/binary32 */	1571	/* convert a float to ieee single/binary32 */
690	ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;	1572	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
691	ecb_function_ uint32_t	1573	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x)
692	ecb_float_to_binary32 (float x)
693	{	1574	{
694	uint32_t r;	1575	uint32_t r;
695		1576
696	#if ECB_STDFP	1577	#if ECB_STDFP
697	memcpy (&r, &x, 4);	1578	memcpy (&r, &x, 4);
…		…
703	if (x == 0e0f ) return 0x00000000U;	1584	if (x == 0e0f ) return 0x00000000U;
704	if (x > +3.40282346638528860e+38f) return 0x7f800000U;	1585	if (x > +3.40282346638528860e+38f) return 0x7f800000U;
705	if (x < -3.40282346638528860e+38f) return 0xff800000U;	1586	if (x < -3.40282346638528860e+38f) return 0xff800000U;
706	if (x != x ) return 0x7fbfffffU;	1587	if (x != x ) return 0x7fbfffffU;
707		1588
708	m = frexpf (x, &e) * 0x1000000U;	1589	m = ecb_frexpf (x, &e) * 0x1000000U;
709		1590
710	r = m & 0x80000000U;	1591	r = m & 0x80000000U;
711		1592
712	if (r)	1593	if (r)
713	m = -m;	1594	m = -m;
…		…
725		1606
726	return r;	1607	return r;
727	}	1608	}
728		1609
729	/* converts an ieee single/binary32 to a float */	1610	/* converts an ieee single/binary32 to a float */
730	ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;	1611	ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
731	ecb_function_ float	1612	ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x)
732	ecb_binary32_to_float (uint32_t x)
733	{	1613	{
734	float r;	1614	float r;
735		1615
736	#if ECB_STDFP	1616	#if ECB_STDFP
737	memcpy (&r, &x, 4);	1617	memcpy (&r, &x, 4);
…		…
746	x |= 0x800000U;	1626	x |= 0x800000U;
747	else	1627	else
748	e = 1;	1628	e = 1;
749		1629
750	/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */	1630	/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
751	r = ldexpf (x * (0.5f / 0x800000U), e - 126);	1631	r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
752		1632
753	r = neg ? -r : r;	1633	r = neg ? -r : r;
754	#endif	1634	#endif
755		1635
756	return r;	1636	return r;
757	}	1637	}
758		1638
759	/* convert a double to ieee double/binary64 */	1639	/* convert a double to ieee double/binary64 */
760	ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;	1640	ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
761	ecb_function_ uint64_t	1641	ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x)
762	ecb_double_to_binary64 (double x)
763	{	1642	{
764	uint64_t r;	1643	uint64_t r;
765		1644
766	#if ECB_STDFP	1645	#if ECB_STDFP
767	memcpy (&r, &x, 8);	1646	memcpy (&r, &x, 8);
…		…
795		1674
796	return r;	1675	return r;
797	}	1676	}
798		1677
799	/* converts an ieee double/binary64 to a double */	1678	/* converts an ieee double/binary64 to a double */
800	ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;	1679	ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
801	ecb_function_ double	1680	ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x)
802	ecb_binary64_to_double (uint64_t x)
803	{	1681	{
804	double r;	1682	double r;
805		1683
806	#if ECB_STDFP	1684	#if ECB_STDFP
807	memcpy (&r, &x, 8);	1685	memcpy (&r, &x, 8);
…		…
824	#endif	1702	#endif
825		1703
826	return r;	1704	return r;
827	}	1705	}
828		1706
829	#endif	1707	/* convert a float to ieee half/binary16 */
		1708	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
		1709	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x)
		1710	{
		1711	return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
		1712	}
830		1713
831	#endif	1714	/* convert an ieee half/binary16 to float */
		1715	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
		1716	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x)
		1717	{
		1718	return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
		1719	}
832		1720
		1721	#endif
		1722
		1723	#endif
		1724

Diff Legend

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