ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libecb/ecb.h

Comparing libecb/ecb.h (file contents):
Revision 1.139 by root, Tue Oct 14 14:39:06 2014 UTC vs.
Revision 1.212 by root, Fri Mar 25 15:31:22 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 0x00010003	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
		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
107	#endif	140	#endif
108		141
109	#define ECB_CPP (__cplusplus+0)	142	#define ECB_CPP (__cplusplus+0)
110	#define ECB_CPP11 (__cplusplus >= 201103L)	143	#define ECB_CPP11 (__cplusplus >= 201103L)
		144	#define ECB_CPP14 (__cplusplus >= 201402L)
		145	#define ECB_CPP17 (__cplusplus >= 201703L)
111		146
112	#if ECB_CPP	147	#if ECB_CPP
113	#define ECB_C 0	148	#define ECB_C 0
114	#define ECB_STDC_VERSION 0	149	#define ECB_STDC_VERSION 0
115	#else	150	#else
…		…
117	#define ECB_STDC_VERSION __STDC_VERSION__	152	#define ECB_STDC_VERSION __STDC_VERSION__
118	#endif	153	#endif
119		154
120	#define ECB_C99 (ECB_STDC_VERSION >= 199901L)	155	#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
121	#define ECB_C11 (ECB_STDC_VERSION >= 201112L)	156	#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
		157	#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
122		158
123	#if ECB_CPP	159	#if ECB_CPP
124	#define ECB_EXTERN_C extern "C"	160	#define ECB_EXTERN_C extern "C"
125	#define ECB_EXTERN_C_BEG ECB_EXTERN_C {	161	#define ECB_EXTERN_C_BEG ECB_EXTERN_C {
126	#define ECB_EXTERN_C_END }	162	#define ECB_EXTERN_C_END }
…		…
141		177
142	#if ECB_NO_SMP	178	#if ECB_NO_SMP
143	#define ECB_MEMORY_FENCE do { } while (0)	179	#define ECB_MEMORY_FENCE do { } while (0)
144	#endif	180	#endif
145		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
146	#ifndef ECB_MEMORY_FENCE	191	#ifndef ECB_MEMORY_FENCE
147	#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")
148	#if __i386 || __i386__	194	#if __i386 || __i386__
149	#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")
150	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")	196	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
151	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")	197	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
152	#elif __amd64 || __amd64__ || __x86_64 || __x86_64__	198	#elif ECB_GCC_AMD64
153	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")	199	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
154	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")	200	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
155	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")	201	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
156	#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__	202	#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
157	#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 */
158	#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \	211	#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
159	|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__	212	|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
		213	|| defined __ARM_ARCH_6T2__
160	#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")
161	#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \	215	#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
162	|| defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__	216	|| defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
163	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")	217	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
164	#elif __aarch64__	218	#elif __aarch64__
165	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")	219	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
166	#elif (__sparc || __sparc__) && !__sparcv8	220	#elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
167	#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")
168	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")	222	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
169	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")	223	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
170	#elif defined __s390__ || defined __s390x__	224	#elif defined __s390__ || defined __s390x__
171	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")	225	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
…		…
194	#if ECB_GCC_VERSION(4,7)	248	#if ECB_GCC_VERSION(4,7)
195	/* see comment below (stdatomic.h) about the C11 memory model. */	249	/* see comment below (stdatomic.h) about the C11 memory model. */
196	#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)	250	#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
197	#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)	251	#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
198	#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)
199		255
200	/* The __has_feature syntax from clang is so misdesigned that we cannot use it	256	#elif ECB_CLANG_EXTENSION(c_atomic)
201	* without risking compile time errors with other compilers. We *could*
202	* define our own ecb_clang_has_feature, but I just can't be bothered to work
203	* around this shit time and again.
204	* #elif defined __clang && __has_feature (cxx_atomic)
205	* // see comment below (stdatomic.h) about the C11 memory model.	257	/* see comment below (stdatomic.h) about the C11 memory model. */
206	* #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)	258	#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
207	* #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)	259	#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
208	* #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)	260	#define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
209	*/	261	#undef ECB_MEMORY_FENCE_RELAXED
		262	#define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
210		263
211	#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__	264	#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
212	#define ECB_MEMORY_FENCE __sync_synchronize ()	265	#define ECB_MEMORY_FENCE __sync_synchronize ()
213	#elif _MSC_VER >= 1500 /* VC++ 2008 */	266	#elif _MSC_VER >= 1500 /* VC++ 2008 */
214	/* 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... */
…		…
224	#elif defined _WIN32	277	#elif defined _WIN32
225	#include <WinNT.h>	278	#include <WinNT.h>
226	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */	279	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
227	#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110	280	#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
228	#include <mbarrier.h>	281	#include <mbarrier.h>
229	#define ECB_MEMORY_FENCE __machine_rw_barrier ()	282	#define ECB_MEMORY_FENCE __machine_rw_barrier ()
230	#define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()	283	#define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
231	#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 ()
232	#elif __xlC__	286	#elif __xlC__
233	#define ECB_MEMORY_FENCE __sync ()	287	#define ECB_MEMORY_FENCE __sync ()
234	#endif	288	#endif
235	#endif	289	#endif
236		290
237	#ifndef ECB_MEMORY_FENCE	291	#ifndef ECB_MEMORY_FENCE
238	#if ECB_C11 && !defined __STDC_NO_ATOMICS__	292	#if ECB_C11 && !defined __STDC_NO_ATOMICS__
239	/* 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, */
240	/* not just C11 atomics and atomic accesses */	294	/* not just C11 atomics and atomic accesses */
241	#include <stdatomic.h>	295	#include <stdatomic.h>
242	/* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
243	/* any fence other than seq_cst, which isn't very efficient for us. */
244	/* Why that is, we don't know - either the C11 memory model is quite useless */
245	/* for most usages, or gcc and clang have a bug */
246	/* I *currently* lean towards the latter, and inefficiently implement */
247	/* all three of ecb's fences as a seq_cst fence */
248	/* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
249	/* for all __atomic_thread_fence's except seq_cst */
250	#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)
251	#endif	299	#endif
252	#endif	300	#endif
253		301
254	#ifndef ECB_MEMORY_FENCE	302	#ifndef ECB_MEMORY_FENCE
255	#if !ECB_AVOID_PTHREADS	303	#if !ECB_AVOID_PTHREADS
…		…
275		323
276	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE	324	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
277	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE	325	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
278	#endif	326	#endif
279		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
280	/*****************************************************************************/	332	/*****************************************************************************/
281		333
282	#if __cplusplus	334	#if ECB_CPP
283	#define ecb_inline static inline	335	#define ecb_inline static inline
284	#elif ECB_GCC_VERSION(2,5)	336	#elif ECB_GCC_VERSION(2,5)
285	#define ecb_inline static __inline__	337	#define ecb_inline static __inline__
286	#elif ECB_C99	338	#elif ECB_C99
287	#define ecb_inline static inline	339	#define ecb_inline static inline
…		…
301		353
302	#define ECB_CONCAT_(a, b) a ## b	354	#define ECB_CONCAT_(a, b) a ## b
303	#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)	355	#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
304	#define ECB_STRINGIFY_(a) # a	356	#define ECB_STRINGIFY_(a) # a
305	#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))
306		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? */
307	#define ecb_function_ ecb_inline	364	#define ecb_function_ ecb_inline
308		365
309	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)	366	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
310	#define ecb_attribute(attrlist) __attribute__(attrlist)	367	#define ecb_attribute(attrlist) __attribute__ (attrlist)
311	#else	368	#else
312	#define ecb_attribute(attrlist)	369	#define ecb_attribute(attrlist)
313	#endif	370	#endif
314		371
315	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)	372	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
…		…
333	#else	390	#else
334	#define ecb_prefetch(addr,rw,locality)	391	#define ecb_prefetch(addr,rw,locality)
335	#endif	392	#endif
336		393
337	/* no emulation for ecb_decltype */	394	/* no emulation for ecb_decltype */
338	#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; };
339	#define ecb_decltype(x) __decltype(x)	398	#define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
340	#elif ECB_GCC_VERSION(3,0)	399	#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
341	#define ecb_decltype(x) __typeof(x)	400	#define ecb_decltype(x) __typeof__ (x)
342	#endif	401	#endif
343		402
344	#if _MSC_VER >= 1300	403	#if _MSC_VER >= 1300
345	#define ecb_deprecated __declspec(deprecated)	404	#define ecb_deprecated __declspec (deprecated)
346	#else	405	#else
347	#define ecb_deprecated ecb_attribute ((__deprecated__))	406	#define ecb_deprecated ecb_attribute ((__deprecated__))
348	#endif	407	#endif
349		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
350	#define ecb_noinline ecb_attribute ((__noinline__))	420	#define ecb_noinline ecb_attribute ((__noinline__))
		421	#endif
		422
351	#define ecb_unused ecb_attribute ((__unused__))	423	#define ecb_unused ecb_attribute ((__unused__))
352	#define ecb_const ecb_attribute ((__const__))	424	#define ecb_const ecb_attribute ((__const__))
353	#define ecb_pure ecb_attribute ((__pure__))	425	#define ecb_pure ecb_attribute ((__pure__))
354		426
355	/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */	427	#if ECB_C11 || __IBMC_NORETURN
356	#if ECB_C11	428	/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
357	#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)
358	#else	435	#else
359	#define ecb_noreturn ecb_attribute ((__noreturn__))	436	#define ecb_noreturn ecb_attribute ((__noreturn__))
360	#endif	437	#endif
361		438
362	#if ECB_GCC_VERSION(4,3)	439	#if ECB_GCC_VERSION(4,3)
…		…
381	/* count trailing zero bits and count # of one bits */	458	/* count trailing zero bits and count # of one bits */
382	#if ECB_GCC_VERSION(3,4) \	459	#if ECB_GCC_VERSION(3,4) \
383	|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \	460	|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
384	&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \	461	&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
385	&& ECB_CLANG_BUILTIN(__builtin_popcount))	462	&& ECB_CLANG_BUILTIN(__builtin_popcount))
386	/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
387	#define ecb_ld32(x) (__builtin_clz (x) ^ 31)
388	#define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
389	#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))
390	#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)
391	#define ecb_popcount32(x) __builtin_popcount (x)	469	#define ecb_popcount32(x) __builtin_popcount (x)
392	/* no popcountll */	470	/* ecb_popcount64 is more difficult, see below */
393	#else	471	#else
394	ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;	472	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
395	ecb_function_ int	473	ecb_function_ ecb_const int
396	ecb_ctz32 (uint32_t x)	474	ecb_ctz32 (uint32_t x)
397	{	475	{
		476	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		477	unsigned long r;
		478	_BitScanForward (&r, x);
		479	return (int)r;
		480	#else
398	int r = 0;	481	int r;
399		482
400	x &= ~x + 1; /* this isolates the lowest bit */	483	x &= ~x + 1; /* this isolates the lowest bit */
401		484
402	#if ECB_branchless_on_i386	485	#if 1
		486	/* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */
		487	/* This happens to return 32 for x == 0, but the API does not support this */
		488
		489	/* -0 marks unused entries */
		490	static unsigned char table[64] =
		491	{
		492	32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14,
		493	10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15,
		494	31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26,
		495	30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0
		496	};
		497
		498	/* magic constant results in 33 unique values in the upper 6 bits */
		499	x *= 0x0450fbafU; /* == 17 * 65 * 65535 */
		500
		501	r = table [x >> 26];
		502	#elif 0 /* branchless on i386, typically */
		503	r = 0;
403	r += !!(x & 0xaaaaaaaa) << 0;	504	r += !!(x & 0xaaaaaaaa) << 0;
404	r += !!(x & 0xcccccccc) << 1;	505	r += !!(x & 0xcccccccc) << 1;
405	r += !!(x & 0xf0f0f0f0) << 2;	506	r += !!(x & 0xf0f0f0f0) << 2;
406	r += !!(x & 0xff00ff00) << 3;	507	r += !!(x & 0xff00ff00) << 3;
407	r += !!(x & 0xffff0000) << 4;	508	r += !!(x & 0xffff0000) << 4;
408	#else	509	#else /* branchless on modern compilers, typically */
		510	r = 0;
409	if (x & 0xaaaaaaaa) r += 1;	511	if (x & 0xaaaaaaaa) r += 1;
410	if (x & 0xcccccccc) r += 2;	512	if (x & 0xcccccccc) r += 2;
411	if (x & 0xf0f0f0f0) r += 4;	513	if (x & 0xf0f0f0f0) r += 4;
412	if (x & 0xff00ff00) r += 8;	514	if (x & 0xff00ff00) r += 8;
413	if (x & 0xffff0000) r += 16;	515	if (x & 0xffff0000) r += 16;
414	#endif	516	#endif
415		517
416	return r;	518	return r;
		519	#endif
417	}	520	}
418		521
419	ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;	522	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
420	ecb_function_ int	523	ecb_function_ ecb_const int
421	ecb_ctz64 (uint64_t x)	524	ecb_ctz64 (uint64_t x)
422	{	525	{
		526	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		527	unsigned long r;
		528	_BitScanForward64 (&r, x);
		529	return (int)r;
		530	#else
423	int shift = x & 0xffffffffU ? 0 : 32;	531	int shift = x & 0xffffffff ? 0 : 32;
424	return ecb_ctz32 (x >> shift) + shift;	532	return ecb_ctz32 (x >> shift) + shift;
		533	#endif
425	}	534	}
426		535
		536	ecb_function_ ecb_const int ecb_clz32 (uint32_t x);
		537	ecb_function_ ecb_const int
		538	ecb_clz32 (uint32_t x)
		539	{
		540	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		541	unsigned long r;
		542	_BitScanReverse (&r, x);
		543	return (int)r;
		544	#else
		545
		546	/* Robert Harley's algorithm from comp.arch 1996-12-07 */
		547	/* This happens to return 32 for x == 0, but the API does not support this */
		548
		549	/* -0 marks unused table elements */
		550	static unsigned char table[64] =
		551	{
		552	32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0,
		553	-0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0,
		554	17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18,
		555	5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0
		556	};
		557
		558	/* propagate leftmost 1 bit to the right */
		559	x |= x >> 1;
		560	x |= x >> 2;
		561	x |= x >> 4;
		562	x |= x >> 8;
		563	x |= x >> 16;
		564
		565	/* magic constant results in 33 unique values in the upper 6 bits */
		566	x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */
		567
		568	return table [x >> 26];
		569	#endif
		570	}
		571
		572	ecb_function_ ecb_const int ecb_clz64 (uint64_t x);
		573	ecb_function_ ecb_const int
		574	ecb_clz64 (uint64_t x)
		575	{
		576	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		577	unsigned long r;
		578	_BitScanReverse64 (&r, x);
		579	return (int)r;
		580	#else
		581	uint32_t l = x >> 32;
		582	int shift = l ? 0 : 32;
		583	return ecb_clz32 (l ? l : x) + shift;
		584	#endif
		585	}
		586
427	ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;	587	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
428	ecb_function_ int	588	ecb_function_ ecb_const int
429	ecb_popcount32 (uint32_t x)	589	ecb_popcount32 (uint32_t x)
430	{	590	{
431	x -= (x >> 1) & 0x55555555;	591	x -= (x >> 1) & 0x55555555;
432	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);	592	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
433	x = ((x >> 4) + x) & 0x0f0f0f0f;	593	x = ((x >> 4) + x) & 0x0f0f0f0f;
434	x *= 0x01010101;	594	x *= 0x01010101;
435		595
436	return x >> 24;	596	return x >> 24;
437	}	597	}
438		598
439	ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;	599	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
440	ecb_function_ int ecb_ld32 (uint32_t x)	600	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
441	{	601	{
		602	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		603	unsigned long r;
		604	_BitScanReverse (&r, x);
		605	return (int)r;
		606	#else
442	int r = 0;	607	int r = 0;
443		608
444	if (x >> 16) { x >>= 16; r += 16; }	609	if (x >> 16) { x >>= 16; r += 16; }
445	if (x >> 8) { x >>= 8; r += 8; }	610	if (x >> 8) { x >>= 8; r += 8; }
446	if (x >> 4) { x >>= 4; r += 4; }	611	if (x >> 4) { x >>= 4; r += 4; }
447	if (x >> 2) { x >>= 2; r += 2; }	612	if (x >> 2) { x >>= 2; r += 2; }
448	if (x >> 1) { r += 1; }	613	if (x >> 1) { r += 1; }
449		614
450	return r;	615	return r;
		616	#endif
451	}	617	}
452		618
453	ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;	619	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
454	ecb_function_ int ecb_ld64 (uint64_t x)	620	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
455	{	621	{
		622	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		623	unsigned long r;
		624	_BitScanReverse64 (&r, x);
		625	return (int)r;
		626	#else
456	int r = 0;	627	int r = 0;
457		628
458	if (x >> 32) { x >>= 32; r += 32; }	629	if (x >> 32) { x >>= 32; r += 32; }
459		630
460	return r + ecb_ld32 (x);	631	return r + ecb_ld32 (x);
461	}
462	#endif	632	#endif
		633	}
		634	#endif
463		635
464	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;	636	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
465	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }	637	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
466	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;	638	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
467	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }	639	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
468		640
469	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;	641	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
470	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)	642	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
471	{	643	{
472	return ( (x * 0x0802U & 0x22110U)	644	return ( (x * 0x0802U & 0x22110U)
473	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;	645	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
474	}	646	}
475		647
476	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;	648	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
477	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)	649	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
478	{	650	{
479	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);	651	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
480	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);	652	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
481	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);	653	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
482	x = ( x >> 8 ) | ( x << 8);	654	x = ( x >> 8 ) | ( x << 8);
483		655
484	return x;	656	return x;
485	}	657	}
486		658
487	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;	659	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
488	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)	660	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
489	{	661	{
490	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);	662	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
491	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);	663	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
492	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);	664	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
493	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);	665	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
494	x = ( x >> 16 ) | ( x << 16);	666	x = ( x >> 16 ) | ( x << 16);
495		667
496	return x;	668	return x;
497	}	669	}
498		670
499	/* popcount64 is only available on 64 bit cpus as gcc builtin */
500	/* so for this version we are lazy */
501	ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;	671	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
502	ecb_function_ int	672	ecb_function_ ecb_const int
503	ecb_popcount64 (uint64_t x)	673	ecb_popcount64 (uint64_t x)
504	{	674	{
		675	/* popcount64 is only available on 64 bit cpus as gcc builtin. */
		676	/* also, gcc/clang make this surprisingly difficult to use */
		677	#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
		678	return __builtin_popcountl (x);
		679	#else
505	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);	680	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
		681	#endif
506	}	682	}
507		683
508	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;	684	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
509	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;	685	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
510	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;	686	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
511	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;	687	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
512	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;	688	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
513	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;	689	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
514	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;	690	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
515	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;	691	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
516		692
517	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }	693	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
518	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }	694	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
519	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }	695	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
520	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }	696	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
521	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }	697	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
522	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }	698	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
523	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }	699	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
524	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }	700	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
		701
		702	#if ECB_CPP
		703
		704	inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
		705	inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
		706	inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
		707	inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
		708
		709	inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
		710	inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
		711	inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
		712	inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
		713
		714	inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
		715	inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
		716	inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
		717	inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
		718
		719	inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
		720	inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
		721	inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
		722	inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
		723
		724	inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
		725	inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
		726	inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
		727
		728	inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
		729	inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
		730	inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
		731	inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
		732
		733	inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
		734	inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
		735	inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
		736	inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
		737
		738	#endif
525		739
526	#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))	740	#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
		741	#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
		742	#define ecb_bswap16(x) __builtin_bswap16 (x)
		743	#else
527	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)	744	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
		745	#endif
528	#define ecb_bswap32(x) __builtin_bswap32 (x)	746	#define ecb_bswap32(x) __builtin_bswap32 (x)
529	#define ecb_bswap64(x) __builtin_bswap64 (x)	747	#define ecb_bswap64(x) __builtin_bswap64 (x)
		748	#elif _MSC_VER
		749	#include <stdlib.h>
		750	#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
		751	#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
		752	#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
530	#else	753	#else
531	ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;	754	ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
532	ecb_function_ uint16_t	755	ecb_function_ ecb_const uint16_t
533	ecb_bswap16 (uint16_t x)	756	ecb_bswap16 (uint16_t x)
534	{	757	{
535	return ecb_rotl16 (x, 8);	758	return ecb_rotl16 (x, 8);
536	}	759	}
537		760
538	ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;	761	ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
539	ecb_function_ uint32_t	762	ecb_function_ ecb_const uint32_t
540	ecb_bswap32 (uint32_t x)	763	ecb_bswap32 (uint32_t x)
541	{	764	{
542	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);	765	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
543	}	766	}
544		767
545	ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;	768	ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
546	ecb_function_ uint64_t	769	ecb_function_ ecb_const uint64_t
547	ecb_bswap64 (uint64_t x)	770	ecb_bswap64 (uint64_t x)
548	{	771	{
549	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);	772	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
550	}	773	}
551	#endif	774	#endif
552		775
553	#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)	776	#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
554	#define ecb_unreachable() __builtin_unreachable ()	777	#define ecb_unreachable() __builtin_unreachable ()
555	#else	778	#else
556	/* this seems to work fine, but gcc always emits a warning for it :/ */	779	/* this seems to work fine, but gcc always emits a warning for it :/ */
557	ecb_inline void ecb_unreachable (void) ecb_noreturn;	780	ecb_inline ecb_noreturn void ecb_unreachable (void);
558	ecb_inline void ecb_unreachable (void) { }	781	ecb_inline ecb_noreturn void ecb_unreachable (void) { }
559	#endif	782	#endif
560		783
561	/* try to tell the compiler that some condition is definitely true */	784	/* try to tell the compiler that some condition is definitely true */
562	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0	785	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
563		786
564	ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;	787	ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
565	ecb_inline unsigned char	788	ecb_inline ecb_const uint32_t
566	ecb_byteorder_helper (void)	789	ecb_byteorder_helper (void)
567	{	790	{
568	/* the union code still generates code under pressure in gcc, */	791	/* the union code still generates code under pressure in gcc, */
569	/* but less than using pointers, and always seems to */	792	/* but less than using pointers, and always seems to */
570	/* successfully return a constant. */	793	/* successfully return a constant. */
571	/* the reason why we have this horrible preprocessor mess */	794	/* the reason why we have this horrible preprocessor mess */
572	/* is to avoid it in all cases, at least on common architectures */	795	/* is to avoid it in all cases, at least on common architectures */
573	/* or when using a recent enough gcc version (>= 4.6) */	796	/* or when using a recent enough gcc version (>= 4.6) */
574	#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
575	return 0x44;
576	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__	797	#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
		798	|| ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
		799	#define ECB_LITTLE_ENDIAN 1
577	return 0x44;	800	return 0x44332211;
578	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__	801	#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
		802	|| ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
		803	#define ECB_BIG_ENDIAN 1
579	return 0x11;	804	return 0x11223344;
580	#else	805	#else
581	union	806	union
582	{	807	{
		808	uint8_t c[4];
583	uint32_t i;	809	uint32_t u;
584	uint8_t c;
585	} u = { 0x11223344 };	810	} u = { 0x11, 0x22, 0x33, 0x44 };
586	return u.c;	811	return u.u;
587	#endif	812	#endif
588	}	813	}
589		814
590	ecb_inline ecb_bool ecb_big_endian (void) ecb_const;	815	ecb_inline ecb_const ecb_bool ecb_big_endian (void);
591	ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }	816	ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
592	ecb_inline ecb_bool ecb_little_endian (void) ecb_const;	817	ecb_inline ecb_const ecb_bool ecb_little_endian (void);
593	ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }	818	ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
		819
		820	/*****************************************************************************/
		821	/* unaligned load/store */
		822
		823	ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		824	ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		825	ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		826
		827	ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		828	ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		829	ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		830
		831	ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		832	ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		833	ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		834
		835	ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
		836	ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
		837	ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
		838
		839	ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
		840	ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
		841	ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
		842
		843	ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		844	ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		845	ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		846
		847	ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		848	ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		849	ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		850
		851	ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
		852	ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
		853	ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
		854
		855	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)); }
		856	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)); }
		857	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)); }
		858
		859	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)); }
		860	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)); }
		861	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)); }
		862
		863	#if ECB_CPP
		864
		865	inline uint8_t ecb_bswap (uint8_t v) { return v; }
		866	inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
		867	inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
		868	inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
		869
		870	template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		871	template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		872	template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
		873	template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
		874	template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
		875	template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
		876	template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
		877	template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
		878
		879	template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		880	template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		881	template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
		882	template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
		883	template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
		884	template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
		885	template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
		886	template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
		887
		888	#endif
		889
		890	/*****************************************************************************/
		891	/* pointer/integer hashing */
		892
		893	/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
		894	ecb_function_ uint32_t ecb_mix32 (uint32_t v);
		895	ecb_function_ uint32_t ecb_mix32 (uint32_t v)
		896	{
		897	v ^= v >> 16; v *= 0x7feb352dU;
		898	v ^= v >> 15; v *= 0x846ca68bU;
		899	v ^= v >> 16;
		900	return v;
		901	}
		902
		903	ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
		904	ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
		905	{
		906	v ^= v >> 16 ; v *= 0x43021123U;
		907	v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
		908	v ^= v >> 16 ;
		909	return v;
		910	}
		911
		912	/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
		913	ecb_function_ uint64_t ecb_mix64 (uint64_t v);
		914	ecb_function_ uint64_t ecb_mix64 (uint64_t v)
		915	{
		916	v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
		917	v ^= v >> 27; v *= 0x94d049bb133111ebU;
		918	v ^= v >> 31;
		919	return v;
		920	}
		921
		922	ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
		923	ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
		924	{
		925	v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
		926	v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
		927	v ^= v >> 30 ^ v >> 60;
		928	return v;
		929	}
		930
		931	ecb_function_ uintptr_t ecb_ptrmix (void *p);
		932	ecb_function_ uintptr_t ecb_ptrmix (void *p)
		933	{
		934	#if ECB_PTRSIZE <= 4
		935	return ecb_mix32 ((uint32_t)p);
		936	#else
		937	return ecb_mix64 ((uint64_t)p);
		938	#endif
		939	}
		940
		941	ecb_function_ void *ecb_ptrunmix (uintptr_t v);
		942	ecb_function_ void *ecb_ptrunmix (uintptr_t v)
		943	{
		944	#if ECB_PTRSIZE <= 4
		945	return (void *)ecb_unmix32 (v);
		946	#else
		947	return (void *)ecb_unmix64 (v);
		948	#endif
		949	}
		950
		951	#if ECB_CPP
		952
		953	template<typename T>
		954	inline uintptr_t ecb_ptrmix (T *p)
		955	{
		956	return ecb_ptrmix (static_cast<void *>(p));
		957	}
		958
		959	template<typename T>
		960	inline T *ecb_ptrunmix (uintptr_t v)
		961	{
		962	return static_cast<T *>(ecb_ptrunmix (v));
		963	}
		964
		965	#endif
		966
		967	/*****************************************************************************/
		968	/* gray code */
		969
		970	ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); }
		971	ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); }
		972	ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); }
		973	ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); }
		974
		975	ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g);
		976	ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g)
		977	{
		978	g ^= g >> 1;
		979	g ^= g >> 2;
		980	g ^= g >> 4;
		981
		982	return g;
		983	}
		984
		985	ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g);
		986	ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g)
		987	{
		988	g ^= g >> 1;
		989	g ^= g >> 2;
		990	g ^= g >> 4;
		991	g ^= g >> 8;
		992
		993	return g;
		994	}
		995
		996	ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g);
		997	ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g)
		998	{
		999	g ^= g >> 1;
		1000	g ^= g >> 2;
		1001	g ^= g >> 4;
		1002	g ^= g >> 8;
		1003	g ^= g >> 16;
		1004
		1005	return g;
		1006	}
		1007
		1008	ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g);
		1009	ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g)
		1010	{
		1011	g ^= g >> 1;
		1012	g ^= g >> 2;
		1013	g ^= g >> 4;
		1014	g ^= g >> 8;
		1015	g ^= g >> 16;
		1016	g ^= g >> 32;
		1017
		1018	return g;
		1019	}
		1020
		1021	#if ECB_CPP
		1022
		1023	ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); }
		1024	ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); }
		1025	ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); }
		1026	ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); }
		1027
		1028	ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); }
		1029	ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); }
		1030	ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); }
		1031	ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); }
		1032
		1033	#endif
		1034
		1035	/*****************************************************************************/
		1036	/* 2d hilbert curves */
		1037
		1038	/* algorithm from the book Hacker's Delight, modified to not */
		1039	/* run into undefined behaviour for n==16 */
		1040	static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s);
		1041	static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
		1042	{
		1043	uint32_t comp, swap, cs, t, sr;
		1044
		1045	/* pad s on the left (unused) bits with 01 (no change groups) */
		1046	s |= 0x55555555U << n << n;
		1047	/* "s shift right" */
		1048	sr = (s >> 1) & 0x55555555U;
		1049	/* compute complement and swap info in two-bit groups */
		1050	cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
		1051
		1052	/* parallel prefix xor op to propagate both complement
		1053	* and swap info together from left to right (there is
		1054	* no step "cs ^= cs >> 1", so in effect it computes
		1055	* two independent parallel prefix operations on two
		1056	* interleaved sets of sixteen bits).
		1057	*/
		1058	cs ^= cs >> 2;
		1059	cs ^= cs >> 4;
		1060	cs ^= cs >> 8;
		1061	cs ^= cs >> 16;
		1062
		1063	/* separate swap and complement bits */
		1064	swap = cs & 0x55555555U;
		1065	comp = (cs >> 1) & 0x55555555U;
		1066
		1067	/* calculate coordinates in odd and even bit positions */
		1068	t = (s & swap) ^ comp;
		1069	s = s ^ sr ^ t ^ (t << 1);
		1070
		1071	/* unpad/clear out any junk on the left */
		1072	s = s & ((1 << n << n) - 1);
		1073
		1074	/* Now "unshuffle" to separate the x and y bits. */
		1075	t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
		1076	t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
		1077	t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
		1078	t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
		1079
		1080	/* now s contains two 16-bit coordinates */
		1081	return s;
		1082	}
		1083
		1084	/* 64 bit, a straightforward extension to the 32 bit case */
		1085	static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s);
		1086	static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
		1087	{
		1088	uint64_t comp, swap, cs, t, sr;
		1089
		1090	/* pad s on the left (unused) bits with 01 (no change groups) */
		1091	s |= 0x5555555555555555U << n << n;
		1092	/* "s shift right" */
		1093	sr = (s >> 1) & 0x5555555555555555U;
		1094	/* compute complement and swap info in two-bit groups */
		1095	cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
		1096
		1097	/* parallel prefix xor op to propagate both complement
		1098	* and swap info together from left to right (there is
		1099	* no step "cs ^= cs >> 1", so in effect it computes
		1100	* two independent parallel prefix operations on two
		1101	* interleaved sets of thirty-two bits).
		1102	*/
		1103	cs ^= cs >> 2;
		1104	cs ^= cs >> 4;
		1105	cs ^= cs >> 8;
		1106	cs ^= cs >> 16;
		1107	cs ^= cs >> 32;
		1108
		1109	/* separate swap and complement bits */
		1110	swap = cs & 0x5555555555555555U;
		1111	comp = (cs >> 1) & 0x5555555555555555U;
		1112
		1113	/* calculate coordinates in odd and even bit positions */
		1114	t = (s & swap) ^ comp;
		1115	s = s ^ sr ^ t ^ (t << 1);
		1116
		1117	/* unpad/clear out any junk on the left */
		1118	s = s & ((1 << n << n) - 1);
		1119
		1120	/* Now "unshuffle" to separate the x and y bits. */
		1121	t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
		1122	t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
		1123	t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
		1124	t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
		1125	t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
		1126
		1127	/* now s contains two 32-bit coordinates */
		1128	return s;
		1129	}
		1130
		1131	/* algorithm from the book Hacker's Delight, but a similar algorithm*/
		1132	/* is given in https://doi.org/10.1002/spe.4380160103 */
		1133	/* this has been slightly improved over the original version */
		1134	ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy);
		1135	ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
		1136	{
		1137	uint32_t row;
		1138	uint32_t state = 0;
		1139	uint32_t s = 0;
		1140
		1141	do
		1142	{
		1143	--n;
		1144
		1145	row = 4 * state
		1146	| (2 & (xy >> n >> 15))
		1147	| (1 & (xy >> n ));
		1148
		1149	/* these funky constants are lookup tables for two-bit values */
		1150	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1151	state = (0x8fe65831U >> 2 * row) & 3;
		1152	}
		1153	while (n > 0);
		1154
		1155	return s;
		1156	}
		1157
		1158	/* 64 bit, essentially the same as 32 bit */
		1159	ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy);
		1160	ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
		1161	{
		1162	uint32_t row;
		1163	uint32_t state = 0;
		1164	uint64_t s = 0;
		1165
		1166	do
		1167	{
		1168	--n;
		1169
		1170	row = 4 * state
		1171	| (2 & (xy >> n >> 31))
		1172	| (1 & (xy >> n ));
		1173
		1174	/* these funky constants are lookup tables for two-bit values */
		1175	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1176	state = (0x8fe65831U >> 2 * row) & 3;
		1177	}
		1178	while (n > 0);
		1179
		1180	return s;
		1181	}
		1182
		1183	/*****************************************************************************/
		1184	/* division */
594		1185
595	#if ECB_GCC_VERSION(3,0) || ECB_C99	1186	#if ECB_GCC_VERSION(3,0) || ECB_C99
		1187	/* C99 tightened the definition of %, so we can use a more efficient version */
596	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))	1188	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
597	#else	1189	#else
598	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))	1190	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
599	#endif	1191	#endif
600		1192
601	#if __cplusplus	1193	#if ECB_CPP
602	template<typename T>	1194	template<typename T>
603	static inline T ecb_div_rd (T val, T div)	1195	static inline T ecb_div_rd (T val, T div)
604	{	1196	{
605	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;	1197	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
606	}	1198	}
…		…
611	}	1203	}
612	#else	1204	#else
613	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))	1205	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
614	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))	1206	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
615	#endif	1207	#endif
		1208
		1209	/*****************************************************************************/
		1210	/* array length */
616		1211
617	#if ecb_cplusplus_does_not_suck	1212	#if ecb_cplusplus_does_not_suck
618	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */	1213	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
619	template<typename T, int N>	1214	template<typename T, int N>
620	static inline int ecb_array_length (const T (&arr)[N])	1215	static inline int ecb_array_length (const T (&arr)[N])
…		…
623	}	1218	}
624	#else	1219	#else
625	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))	1220	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
626	#endif	1221	#endif
627		1222
		1223	/*****************************************************************************/
		1224	/* IEEE 754-2008 half float conversions */
		1225
		1226	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
		1227	ecb_function_ ecb_const uint32_t
		1228	ecb_binary16_to_binary32 (uint32_t x)
		1229	{
		1230	unsigned int s = (x & 0x8000) << (31 - 15);
		1231	int e = (x >> 10) & 0x001f;
		1232	unsigned int m = x & 0x03ff;
		1233
		1234	if (ecb_expect_false (e == 31))
		1235	/* infinity or NaN */
		1236	e = 255 - (127 - 15);
		1237	else if (ecb_expect_false (!e))
		1238	{
		1239	if (ecb_expect_true (!m))
		1240	/* zero, handled by code below by forcing e to 0 */
		1241	e = 0 - (127 - 15);
		1242	else
		1243	{
		1244	/* subnormal, renormalise */
		1245	unsigned int s = 10 - ecb_ld32 (m);
		1246
		1247	m = (m << s) & 0x3ff; /* mask implicit bit */
		1248	e -= s - 1;
		1249	}
		1250	}
		1251
		1252	/* e and m now are normalised, or zero, (or inf or nan) */
		1253	e += 127 - 15;
		1254
		1255	return s | (e << 23) | (m << (23 - 10));
		1256	}
		1257
		1258	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
		1259	ecb_function_ ecb_const uint16_t
		1260	ecb_binary32_to_binary16 (uint32_t x)
		1261	{
		1262	unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
		1263	int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
		1264	unsigned int m = x & 0x007fffff;
		1265
		1266	x &= 0x7fffffff;
		1267
		1268	/* if it's within range of binary16 normals, use fast path */
		1269	if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
		1270	{
		1271	/* mantissa round-to-even */
		1272	m += 0x00000fff + ((m >> (23 - 10)) & 1);
		1273
		1274	/* handle overflow */
		1275	if (ecb_expect_false (m >= 0x00800000))
		1276	{
		1277	m >>= 1;
		1278	e += 1;
		1279	}
		1280
		1281	return s | (e << 10) | (m >> (23 - 10));
		1282	}
		1283
		1284	/* handle large numbers and infinity */
		1285	if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
		1286	return s | 0x7c00;
		1287
		1288	/* handle zero, subnormals and small numbers */
		1289	if (ecb_expect_true (x < 0x38800000))
		1290	{
		1291	/* zero */
		1292	if (ecb_expect_true (!x))
		1293	return s;
		1294
		1295	/* handle subnormals */
		1296
		1297	/* too small, will be zero */
		1298	if (e < (14 - 24)) /* might not be sharp, but is good enough */
		1299	return s;
		1300
		1301	m |= 0x00800000; /* make implicit bit explicit */
		1302
		1303	/* very tricky - we need to round to the nearest e (+10) bit value */
		1304	{
		1305	unsigned int bits = 14 - e;
		1306	unsigned int half = (1 << (bits - 1)) - 1;
		1307	unsigned int even = (m >> bits) & 1;
		1308
		1309	/* if this overflows, we will end up with a normalised number */
		1310	m = (m + half + even) >> bits;
		1311	}
		1312
		1313	return s | m;
		1314	}
		1315
		1316	/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
		1317	m >>= 13;
		1318
		1319	return s | 0x7c00 | m | !m;
		1320	}
		1321
		1322	/*******************************************************************************/
		1323	/* fast integer to ascii */
		1324
		1325	/*
		1326	* This code is pretty complicated because it is general. The idea behind it,
		1327	* however, is pretty simple: first, the number is multiplied with a scaling
		1328	* factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
		1329	* number with the first digit in the upper bits.
		1330	* Then this digit is converted to text and masked out. The resulting number
		1331	* is then multiplied by 10, by multiplying the fixed point representation
		1332	* by 5 and shifting the (binary) decimal point one to the right, so a 4.28
		1333	* format becomes 5.27, 6.26 and so on.
		1334	* The rest involves only advancing the pointer if we already generated a
		1335	* non-zero digit, so leading zeroes are overwritten.
		1336	*/
		1337
		1338	/* simply return a mask with "bits" bits set */
		1339	#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
		1340
		1341	/* oputput a single digit. maskvalue is 10**digitidx */
		1342	#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
		1343	if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
		1344	{ \
		1345	char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
		1346	*ptr = digit + '0'; /* output it */ \
		1347	nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
		1348	ptr += nz; /* output digit only if non-zero digit seen */ \
		1349	x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
		1350	}
		1351
		1352	/* convert integer to fixed point format and multiply out digits, highest first */
		1353	/* requires magic constants: max. digits and number of bits after the decimal point */
		1354	#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
		1355	ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
		1356	{ \
		1357	char nz = lz; /* non-zero digit seen? */ \
		1358	/* convert to x.bits fixed-point */ \
		1359	type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
		1360	/* output up to 10 digits */ \
		1361	ecb_i2a_digit (type,bits,digitmask, 1, 0); \
		1362	ecb_i2a_digit (type,bits,digitmask, 10, 1); \
		1363	ecb_i2a_digit (type,bits,digitmask, 100, 2); \
		1364	ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
		1365	ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
		1366	ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
		1367	ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
		1368	ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
		1369	ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
		1370	ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
		1371	return ptr; \
		1372	}
		1373
		1374	/* predefined versions of the above, for various digits */
		1375	/* ecb_i2a_xN = almost N digits, limit defined by macro */
		1376	/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
		1377	/* ecb_i2a_0N = exactly N digits, including leading zeroes */
		1378
		1379	/* non-leading-zero versions, limited range */
		1380	#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
		1381	#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
		1382	ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
		1383	ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
		1384
		1385	/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
		1386	ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
		1387	ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
		1388	ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
		1389	ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
		1390	ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
		1391	ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
		1392	ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
		1393	ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
		1394
		1395	/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
		1396	ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
		1397	ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
		1398	ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
		1399	ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
		1400	ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
		1401	ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
		1402	ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
		1403	ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
		1404
		1405	#define ECB_I2A_I32_DIGITS 11
		1406	#define ECB_I2A_U32_DIGITS 10
		1407	#define ECB_I2A_I64_DIGITS 20
		1408	#define ECB_I2A_U64_DIGITS 21
		1409	#define ECB_I2A_MAX_DIGITS 21
		1410
		1411	ecb_inline char *
		1412	ecb_i2a_u32 (char *ptr, uint32_t u)
		1413	{
		1414	#if ECB_64BIT_NATIVE
		1415	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1416	ptr = ecb_i2a_x10 (ptr, u);
		1417	else /* x10 almost, but not fully, covers 32 bit */
		1418	{
		1419	uint32_t u1 = u % 1000000000;
		1420	uint32_t u2 = u / 1000000000;
		1421
		1422	*ptr++ = u2 + '0';
		1423	ptr = ecb_i2a_09 (ptr, u1);
		1424	}
		1425	#else
		1426	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1427	ecb_i2a_x5 (ptr, u);
		1428	else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
		1429	{
		1430	uint32_t u1 = u % 10000;
		1431	uint32_t u2 = u / 10000;
		1432
		1433	ptr = ecb_i2a_x5 (ptr, u2);
		1434	ptr = ecb_i2a_04 (ptr, u1);
		1435	}
		1436	else
		1437	{
		1438	uint32_t u1 = u % 10000;
		1439	uint32_t ua = u / 10000;
		1440	uint32_t u2 = ua % 10000;
		1441	uint32_t u3 = ua / 10000;
		1442
		1443	ptr = ecb_i2a_2 (ptr, u3);
		1444	ptr = ecb_i2a_04 (ptr, u2);
		1445	ptr = ecb_i2a_04 (ptr, u1);
		1446	}
		1447	#endif
		1448
		1449	return ptr;
		1450	}
		1451
		1452	ecb_inline char *
		1453	ecb_i2a_i32 (char *ptr, int32_t v)
		1454	{
		1455	*ptr = '-'; ptr += v < 0;
		1456	uint32_t u = v < 0 ? -(uint32_t)v : v;
		1457
		1458	#if ECB_64BIT_NATIVE
		1459	ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
		1460	#else
		1461	ptr = ecb_i2a_u32 (ptr, u);
		1462	#endif
		1463
		1464	return ptr;
		1465	}
		1466
		1467	ecb_inline char *
		1468	ecb_i2a_u64 (char *ptr, uint64_t u)
		1469	{
		1470	#if ECB_64BIT_NATIVE
		1471	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1472	ptr = ecb_i2a_x10 (ptr, u);
		1473	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1474	{
		1475	uint64_t u1 = u % 1000000000;
		1476	uint64_t u2 = u / 1000000000;
		1477
		1478	ptr = ecb_i2a_x10 (ptr, u2);
		1479	ptr = ecb_i2a_09 (ptr, u1);
		1480	}
		1481	else
		1482	{
		1483	uint64_t u1 = u % 1000000000;
		1484	uint64_t ua = u / 1000000000;
		1485	uint64_t u2 = ua % 1000000000;
		1486	uint64_t u3 = ua / 1000000000;
		1487
		1488	ptr = ecb_i2a_2 (ptr, u3);
		1489	ptr = ecb_i2a_09 (ptr, u2);
		1490	ptr = ecb_i2a_09 (ptr, u1);
		1491	}
		1492	#else
		1493	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1494	ptr = ecb_i2a_x5 (ptr, u);
		1495	else
		1496	{
		1497	uint64_t u1 = u % 10000;
		1498	uint64_t u2 = u / 10000;
		1499
		1500	ptr = ecb_i2a_u64 (ptr, u2);
		1501	ptr = ecb_i2a_04 (ptr, u1);
		1502	}
		1503	#endif
		1504
		1505	return ptr;
		1506	}
		1507
		1508	ecb_inline char *
		1509	ecb_i2a_i64 (char *ptr, int64_t v)
		1510	{
		1511	*ptr = '-'; ptr += v < 0;
		1512	uint64_t u = v < 0 ? -(uint64_t)v : v;
		1513
		1514	#if ECB_64BIT_NATIVE
		1515	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1516	ptr = ecb_i2a_x10 (ptr, u);
		1517	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1518	{
		1519	uint64_t u1 = u % 1000000000;
		1520	uint64_t u2 = u / 1000000000;
		1521
		1522	ptr = ecb_i2a_x10 (ptr, u2);
		1523	ptr = ecb_i2a_09 (ptr, u1);
		1524	}
		1525	else
		1526	{
		1527	uint64_t u1 = u % 1000000000;
		1528	uint64_t ua = u / 1000000000;
		1529	uint64_t u2 = ua % 1000000000;
		1530	uint64_t u3 = ua / 1000000000;
		1531
		1532	/* 2**31 is 19 digits, so the top is exactly one digit */
		1533	*ptr++ = u3 + '0';
		1534	ptr = ecb_i2a_09 (ptr, u2);
		1535	ptr = ecb_i2a_09 (ptr, u1);
		1536	}
		1537	#else
		1538	ptr = ecb_i2a_u64 (ptr, u);
		1539	#endif
		1540
		1541	return ptr;
		1542	}
		1543
628	/*******************************************************************************/	1544	/*******************************************************************************/
629	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */	1545	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
630		1546
631	/* basically, everything uses "ieee pure-endian" floating point numbers */	1547	/* basically, everything uses "ieee pure-endian" floating point numbers */
632	/* the only noteworthy exception is ancient armle, which uses order 43218765 */	1548	/* the only noteworthy exception is ancient armle, which uses order 43218765 */
633	#if 0 \	1549	#if 0 \
634	|| __i386 || __i386__ \	1550	|| __i386 || __i386__ \
635	|| __amd64 || __amd64__ || __x86_64 || __x86_64__ \	1551	|| ECB_GCC_AMD64 \
636	|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \	1552	|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
637	|| defined __s390__ || defined __s390x__ \	1553	|| defined __s390__ || defined __s390x__ \
638	|| defined __mips__ \	1554	|| defined __mips__ \
639	|| defined __alpha__ \	1555	|| defined __alpha__ \
640	|| defined __hppa__ \	1556	|| defined __hppa__ \
641	|| defined __ia64__ \	1557	|| defined __ia64__ \
642	|| defined __m68k__ \	1558	|| defined __m68k__ \
643	|| defined __m88k__ \	1559	|| defined __m88k__ \
644	|| defined __sh__ \	1560	|| defined __sh__ \
645	|| defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \	1561	|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
646	|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \	1562	|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
647	|| defined __aarch64__	1563	|| defined __aarch64__
648	#define ECB_STDFP 1	1564	#define ECB_STDFP 1
649	#include <string.h> /* for memcpy */
650	#else	1565	#else
651	#define ECB_STDFP 0	1566	#define ECB_STDFP 0
652	#endif	1567	#endif
653		1568
654	#ifndef ECB_NO_LIBM	1569	#ifndef ECB_NO_LIBM
…		…
666	#define ECB_NAN NAN	1581	#define ECB_NAN NAN
667	#else	1582	#else
668	#define ECB_NAN ECB_INFINITY	1583	#define ECB_NAN ECB_INFINITY
669	#endif	1584	#endif
670		1585
671	/* converts an ieee half/binary16 to a float */	1586	#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
672	ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;	1587	#define ecb_ldexpf(x,e) ldexpf ((x), (e))
673	ecb_function_ float	1588	#define ecb_frexpf(x,e) frexpf ((x), (e))
674	ecb_binary16_to_float (uint16_t x)	1589	#else
675	{	1590	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
676	int e = (x >> 10) & 0x1f;	1591	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
677	int m = x & 0x3ff;	1592	#endif
678	float r;
679
680	if (!e ) r = ldexpf (m , -24);
681	else if (e != 31) r = ldexpf (m + 0x400, e - 25);
682	else if (m ) r = ECB_NAN;
683	else r = ECB_INFINITY;
684
685	return x & 0x8000 ? -r : r;
686	}
687		1593
688	/* convert a float to ieee single/binary32 */	1594	/* convert a float to ieee single/binary32 */
689	ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;	1595	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
690	ecb_function_ uint32_t	1596	ecb_function_ ecb_const uint32_t
691	ecb_float_to_binary32 (float x)	1597	ecb_float_to_binary32 (float x)
692	{	1598	{
693	uint32_t r;	1599	uint32_t r;
694		1600
695	#if ECB_STDFP	1601	#if ECB_STDFP
…		…
702	if (x == 0e0f ) return 0x00000000U;	1608	if (x == 0e0f ) return 0x00000000U;
703	if (x > +3.40282346638528860e+38f) return 0x7f800000U;	1609	if (x > +3.40282346638528860e+38f) return 0x7f800000U;
704	if (x < -3.40282346638528860e+38f) return 0xff800000U;	1610	if (x < -3.40282346638528860e+38f) return 0xff800000U;
705	if (x != x ) return 0x7fbfffffU;	1611	if (x != x ) return 0x7fbfffffU;
706		1612
707	m = frexpf (x, &e) * 0x1000000U;	1613	m = ecb_frexpf (x, &e) * 0x1000000U;
708		1614
709	r = m & 0x80000000U;	1615	r = m & 0x80000000U;
710		1616
711	if (r)	1617	if (r)
712	m = -m;	1618	m = -m;
…		…
724		1630
725	return r;	1631	return r;
726	}	1632	}
727		1633
728	/* converts an ieee single/binary32 to a float */	1634	/* converts an ieee single/binary32 to a float */
729	ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;	1635	ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
730	ecb_function_ float	1636	ecb_function_ ecb_const float
731	ecb_binary32_to_float (uint32_t x)	1637	ecb_binary32_to_float (uint32_t x)
732	{	1638	{
733	float r;	1639	float r;
734		1640
735	#if ECB_STDFP	1641	#if ECB_STDFP
…		…
745	x |= 0x800000U;	1651	x |= 0x800000U;
746	else	1652	else
747	e = 1;	1653	e = 1;
748		1654
749	/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */	1655	/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
750	r = ldexpf (x * (0.5f / 0x800000U), e - 126);	1656	r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
751		1657
752	r = neg ? -r : r;	1658	r = neg ? -r : r;
753	#endif	1659	#endif
754		1660
755	return r;	1661	return r;
756	}	1662	}
757		1663
758	/* convert a double to ieee double/binary64 */	1664	/* convert a double to ieee double/binary64 */
759	ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;	1665	ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
760	ecb_function_ uint64_t	1666	ecb_function_ ecb_const uint64_t
761	ecb_double_to_binary64 (double x)	1667	ecb_double_to_binary64 (double x)
762	{	1668	{
763	uint64_t r;	1669	uint64_t r;
764		1670
765	#if ECB_STDFP	1671	#if ECB_STDFP
…		…
794		1700
795	return r;	1701	return r;
796	}	1702	}
797		1703
798	/* converts an ieee double/binary64 to a double */	1704	/* converts an ieee double/binary64 to a double */
799	ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;	1705	ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
800	ecb_function_ double	1706	ecb_function_ ecb_const double
801	ecb_binary64_to_double (uint64_t x)	1707	ecb_binary64_to_double (uint64_t x)
802	{	1708	{
803	double r;	1709	double r;
804		1710
805	#if ECB_STDFP	1711	#if ECB_STDFP
…		…
823	#endif	1729	#endif
824		1730
825	return r;	1731	return r;
826	}	1732	}
827		1733
828	#endif	1734	/* convert a float to ieee half/binary16 */
		1735	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
		1736	ecb_function_ ecb_const uint16_t
		1737	ecb_float_to_binary16 (float x)
		1738	{
		1739	return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
		1740	}
829		1741
830	#endif	1742	/* convert an ieee half/binary16 to float */
		1743	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
		1744	ecb_function_ ecb_const float
		1745	ecb_binary16_to_float (uint16_t x)
		1746	{
		1747	return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
		1748	}
831		1749
		1750	#endif
		1751
		1752	#endif
		1753

Diff Legend

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