ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libecb/ecb.h

Comparing libecb/ecb.h (file contents):
Revision 1.148 by root, Fri Oct 24 04:50:13 2014 UTC vs.
Revision 1.210 by root, Fri Mar 25 15:28:08 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.
…		…
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	/* TODO: popssibly static would be best for these at the moment? */
308	#define ecb_function_ ecb_inline	362	#define ecb_function_ ecb_inline
309		363
310	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)	364	#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
311	#define ecb_attribute(attrlist) __attribute__ (attrlist)	365	#define ecb_attribute(attrlist) __attribute__ (attrlist)
312	#else	366	#else
…		…
343	#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)	397	#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
344	#define ecb_decltype(x) __typeof__ (x)	398	#define ecb_decltype(x) __typeof__ (x)
345	#endif	399	#endif
346		400
347	#if _MSC_VER >= 1300	401	#if _MSC_VER >= 1300
348	#define ecb_deprecated __declspec(deprecated)	402	#define ecb_deprecated __declspec (deprecated)
349	#else	403	#else
350	#define ecb_deprecated ecb_attribute ((__deprecated__))	404	#define ecb_deprecated ecb_attribute ((__deprecated__))
351	#endif	405	#endif
352		406
		407	#if _MSC_VER >= 1500
		408	#define ecb_deprecated_message(msg) __declspec (deprecated (msg))
		409	#elif ECB_GCC_VERSION(4,5)
		410	#define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
		411	#else
		412	#define ecb_deprecated_message(msg) ecb_deprecated
		413	#endif
		414
		415	#if _MSC_VER >= 1400
		416	#define ecb_noinline __declspec (noinline)
		417	#else
353	#define ecb_noinline ecb_attribute ((__noinline__))	418	#define ecb_noinline ecb_attribute ((__noinline__))
		419	#endif
		420
354	#define ecb_unused ecb_attribute ((__unused__))	421	#define ecb_unused ecb_attribute ((__unused__))
355	#define ecb_const ecb_attribute ((__const__))	422	#define ecb_const ecb_attribute ((__const__))
356	#define ecb_pure ecb_attribute ((__pure__))	423	#define ecb_pure ecb_attribute ((__pure__))
357		424
358	/* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
359	#if ECB_C11 || __IBMC_NORETURN	425	#if ECB_C11 || __IBMC_NORETURN
360	/* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */	426	/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
361	#define ecb_noreturn _Noreturn	427	#define ecb_noreturn _Noreturn
		428	#elif ECB_CPP11
		429	#define ecb_noreturn [[noreturn]]
		430	#elif _MSC_VER >= 1200
		431	/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
		432	#define ecb_noreturn __declspec (noreturn)
362	#else	433	#else
363	#define ecb_noreturn ecb_attribute ((__noreturn__))	434	#define ecb_noreturn ecb_attribute ((__noreturn__))
364	#endif	435	#endif
365		436
366	#if ECB_GCC_VERSION(4,3)	437	#if ECB_GCC_VERSION(4,3)
…		…
385	/* count trailing zero bits and count # of one bits */	456	/* count trailing zero bits and count # of one bits */
386	#if ECB_GCC_VERSION(3,4) \	457	#if ECB_GCC_VERSION(3,4) \
387	|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \	458	|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
388	&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \	459	&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
389	&& ECB_CLANG_BUILTIN(__builtin_popcount))	460	&& ECB_CLANG_BUILTIN(__builtin_popcount))
390	/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
391	#define ecb_ld32(x) (__builtin_clz (x) ^ 31)
392	#define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
393	#define ecb_ctz32(x) __builtin_ctz (x)	461	#define ecb_ctz32(x) __builtin_ctz (x)
		462	#define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x))
394	#define ecb_ctz64(x) __builtin_ctzll (x)	463	#define ecb_clz32(x) __builtin_clz (x)
		464	#define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x))
		465	#define ecb_ld32(x) (ecb_clz32 (x) ^ 31)
		466	#define ecb_ld64(x) (ecb_clz64 (x) ^ 63)
395	#define ecb_popcount32(x) __builtin_popcount (x)	467	#define ecb_popcount32(x) __builtin_popcount (x)
396	/* no popcountll */	468	/* ecb_popcount64 is more difficult, see below */
397	#else	469	#else
398	ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;	470	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
399	ecb_function_ int	471	ecb_function_ ecb_const int
400	ecb_ctz32 (uint32_t x)	472	ecb_ctz32 (uint32_t x)
401	{	473	{
		474	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		475	unsigned long r;
		476	_BitScanForward (&r, x);
		477	return (int)r;
		478	#else
402	int r = 0;	479	int r;
403		480
404	x &= ~x + 1; /* this isolates the lowest bit */	481	x &= ~x + 1; /* this isolates the lowest bit */
405		482
406	#if ECB_branchless_on_i386	483	#if 1
		484	/* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */
		485	/* This happens to return 32 for x == 0, but the API does not support this */
		486
		487	/* -0 marks unused entries */
		488	static unsigned char table[64] =
		489	{
		490	32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14,
		491	10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15,
		492	31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26,
		493	30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0
		494	};
		495
		496	/* magic constant results in 33 unique values in the upper 6 bits */
		497	x *= 0x0450fbafU; /* == 17 * 65 * 65535 */
		498
		499	r = table [x >> 26];
		500	#elif 0 /* branchless on i386, typically */
		501	r = 0;
407	r += !!(x & 0xaaaaaaaa) << 0;	502	r += !!(x & 0xaaaaaaaa) << 0;
408	r += !!(x & 0xcccccccc) << 1;	503	r += !!(x & 0xcccccccc) << 1;
409	r += !!(x & 0xf0f0f0f0) << 2;	504	r += !!(x & 0xf0f0f0f0) << 2;
410	r += !!(x & 0xff00ff00) << 3;	505	r += !!(x & 0xff00ff00) << 3;
411	r += !!(x & 0xffff0000) << 4;	506	r += !!(x & 0xffff0000) << 4;
412	#else	507	#else /* branchless on modern compilers, typically */
		508	r = 0;
413	if (x & 0xaaaaaaaa) r += 1;	509	if (x & 0xaaaaaaaa) r += 1;
414	if (x & 0xcccccccc) r += 2;	510	if (x & 0xcccccccc) r += 2;
415	if (x & 0xf0f0f0f0) r += 4;	511	if (x & 0xf0f0f0f0) r += 4;
416	if (x & 0xff00ff00) r += 8;	512	if (x & 0xff00ff00) r += 8;
417	if (x & 0xffff0000) r += 16;	513	if (x & 0xffff0000) r += 16;
418	#endif	514	#endif
419		515
420	return r;	516	return r;
		517	#endif
421	}	518	}
422		519
423	ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;	520	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
424	ecb_function_ int	521	ecb_function_ ecb_const int
425	ecb_ctz64 (uint64_t x)	522	ecb_ctz64 (uint64_t x)
426	{	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
427	int shift = x & 0xffffffffU ? 0 : 32;	529	int shift = x & 0xffffffff ? 0 : 32;
428	return ecb_ctz32 (x >> shift) + shift;	530	return ecb_ctz32 (x >> shift) + shift;
		531	#endif
429	}	532	}
430		533
		534	ecb_function_ ecb_const int ecb_clz32 (uint32_t x);
		535	ecb_function_ ecb_const int
		536	ecb_clz32 (uint32_t x)
		537	{
		538	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		539	unsigned long r;
		540	_BitScanReverse (&r, x);
		541	return (int)r;
		542	#else
		543
		544	/* Robert Harley's algorithm from comp.arch 1996-12-07 */
		545	/* This happens to return 32 for x == 0, but the API does not support this */
		546
		547	/* -0 marks unused table elements */
		548	static unsigned char table[64] =
		549	{
		550	32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0,
		551	-0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0,
		552	17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18,
		553	5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0
		554	};
		555
		556	/* propagate leftmost 1 bit to the right */
		557	x |= x >> 1;
		558	x |= x >> 2;
		559	x |= x >> 4;
		560	x |= x >> 8;
		561	x |= x >> 16;
		562
		563	/* magic constant results in 33 unique values in the upper 6 bits */
		564	x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */
		565
		566	return table [x >> 26];
		567	#endif
		568	}
		569
		570	ecb_function_ ecb_const int ecb_clz64 (uint64_t x);
		571	ecb_function_ ecb_const int
		572	ecb_clz64 (uint64_t x)
		573	{
		574	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		575	unsigned long r;
		576	_BitScanReverse64 (&r, x);
		577	return (int)r;
		578	#else
		579	uint32_t l = x >> 32;
		580	int shift = l ? 0 : 32;
		581	return ecb_clz32 (l ? l : x) + shift;
		582	#endif
		583	}
		584
431	ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;	585	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
432	ecb_function_ int	586	ecb_function_ ecb_const int
433	ecb_popcount32 (uint32_t x)	587	ecb_popcount32 (uint32_t x)
434	{	588	{
435	x -= (x >> 1) & 0x55555555;	589	x -= (x >> 1) & 0x55555555;
436	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);	590	x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
437	x = ((x >> 4) + x) & 0x0f0f0f0f;	591	x = ((x >> 4) + x) & 0x0f0f0f0f;
438	x *= 0x01010101;	592	x *= 0x01010101;
439		593
440	return x >> 24;	594	return x >> 24;
441	}	595	}
442		596
443	ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;	597	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
444	ecb_function_ int ecb_ld32 (uint32_t x)	598	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
445	{	599	{
		600	#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
		601	unsigned long r;
		602	_BitScanReverse (&r, x);
		603	return (int)r;
		604	#else
446	int r = 0;	605	int r = 0;
447		606
448	if (x >> 16) { x >>= 16; r += 16; }	607	if (x >> 16) { x >>= 16; r += 16; }
449	if (x >> 8) { x >>= 8; r += 8; }	608	if (x >> 8) { x >>= 8; r += 8; }
450	if (x >> 4) { x >>= 4; r += 4; }	609	if (x >> 4) { x >>= 4; r += 4; }
451	if (x >> 2) { x >>= 2; r += 2; }	610	if (x >> 2) { x >>= 2; r += 2; }
452	if (x >> 1) { r += 1; }	611	if (x >> 1) { r += 1; }
453		612
454	return r;	613	return r;
		614	#endif
455	}	615	}
456		616
457	ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;	617	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
458	ecb_function_ int ecb_ld64 (uint64_t x)	618	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
459	{	619	{
		620	#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
		621	unsigned long r;
		622	_BitScanReverse64 (&r, x);
		623	return (int)r;
		624	#else
460	int r = 0;	625	int r = 0;
461		626
462	if (x >> 32) { x >>= 32; r += 32; }	627	if (x >> 32) { x >>= 32; r += 32; }
463		628
464	return r + ecb_ld32 (x);	629	return r + ecb_ld32 (x);
465	}
466	#endif	630	#endif
		631	}
		632	#endif
467		633
468	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;	634	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
469	ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }	635	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
470	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;	636	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
471	ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }	637	ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
472		638
473	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;	639	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
474	ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)	640	ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
475	{	641	{
476	return ( (x * 0x0802U & 0x22110U)	642	return ( (x * 0x0802U & 0x22110U)
477	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;	643	| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
478	}	644	}
479		645
480	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;	646	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
481	ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)	647	ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
482	{	648	{
483	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);	649	x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
484	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);	650	x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
485	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);	651	x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
486	x = ( x >> 8 ) | ( x << 8);	652	x = ( x >> 8 ) | ( x << 8);
487		653
488	return x;	654	return x;
489	}	655	}
490		656
491	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;	657	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
492	ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)	658	ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
493	{	659	{
494	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);	660	x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
495	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);	661	x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
496	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);	662	x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
497	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);	663	x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
498	x = ( x >> 16 ) | ( x << 16);	664	x = ( x >> 16 ) | ( x << 16);
499		665
500	return x;	666	return x;
501	}	667	}
502		668
503	/* popcount64 is only available on 64 bit cpus as gcc builtin */
504	/* so for this version we are lazy */
505	ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;	669	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
506	ecb_function_ int	670	ecb_function_ ecb_const int
507	ecb_popcount64 (uint64_t x)	671	ecb_popcount64 (uint64_t x)
508	{	672	{
		673	/* popcount64 is only available on 64 bit cpus as gcc builtin. */
		674	/* also, gcc/clang make this surprisingly difficult to use */
		675	#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
		676	return __builtin_popcountl (x);
		677	#else
509	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);	678	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
		679	#endif
510	}	680	}
511		681
512	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;	682	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
513	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;	683	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
514	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;	684	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
515	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;	685	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
516	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;	686	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
517	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;	687	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
518	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;	688	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
519	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;	689	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
520		690
521	ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }	691	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
522	ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }	692	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
523	ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }	693	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
524	ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }	694	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
525	ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }	695	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
526	ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }	696	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
527	ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }	697	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
528	ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }	698	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
		699
		700	#if ECB_CPP
		701
		702	inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
		703	inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
		704	inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
		705	inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
		706
		707	inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
		708	inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
		709	inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
		710	inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
		711
		712	inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
		713	inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
		714	inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
		715	inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
		716
		717	inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
		718	inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
		719	inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
		720	inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
		721
		722	inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
		723	inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
		724	inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
		725
		726	inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
		727	inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
		728	inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
		729	inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
		730
		731	inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
		732	inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
		733	inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
		734	inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
		735
		736	#endif
529		737
530	#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))	738	#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
		739	#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
		740	#define ecb_bswap16(x) __builtin_bswap16 (x)
		741	#else
531	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)	742	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
		743	#endif
532	#define ecb_bswap32(x) __builtin_bswap32 (x)	744	#define ecb_bswap32(x) __builtin_bswap32 (x)
533	#define ecb_bswap64(x) __builtin_bswap64 (x)	745	#define ecb_bswap64(x) __builtin_bswap64 (x)
		746	#elif _MSC_VER
		747	#include <stdlib.h>
		748	#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
		749	#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
		750	#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
534	#else	751	#else
535	ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;	752	ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
536	ecb_function_ uint16_t	753	ecb_function_ ecb_const uint16_t
537	ecb_bswap16 (uint16_t x)	754	ecb_bswap16 (uint16_t x)
538	{	755	{
539	return ecb_rotl16 (x, 8);	756	return ecb_rotl16 (x, 8);
540	}	757	}
541		758
542	ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;	759	ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
543	ecb_function_ uint32_t	760	ecb_function_ ecb_const uint32_t
544	ecb_bswap32 (uint32_t x)	761	ecb_bswap32 (uint32_t x)
545	{	762	{
546	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);	763	return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
547	}	764	}
548		765
549	ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;	766	ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
550	ecb_function_ uint64_t	767	ecb_function_ ecb_const uint64_t
551	ecb_bswap64 (uint64_t x)	768	ecb_bswap64 (uint64_t x)
552	{	769	{
553	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);	770	return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
554	}	771	}
555	#endif	772	#endif
556		773
557	#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)	774	#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
558	#define ecb_unreachable() __builtin_unreachable ()	775	#define ecb_unreachable() __builtin_unreachable ()
559	#else	776	#else
560	/* this seems to work fine, but gcc always emits a warning for it :/ */	777	/* this seems to work fine, but gcc always emits a warning for it :/ */
561	ecb_inline void ecb_unreachable (void) ecb_noreturn;	778	ecb_inline ecb_noreturn void ecb_unreachable (void);
562	ecb_inline void ecb_unreachable (void) { }	779	ecb_inline ecb_noreturn void ecb_unreachable (void) { }
563	#endif	780	#endif
564		781
565	/* try to tell the compiler that some condition is definitely true */	782	/* try to tell the compiler that some condition is definitely true */
566	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0	783	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
567		784
568	ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;	785	ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
569	ecb_inline unsigned char	786	ecb_inline ecb_const uint32_t
570	ecb_byteorder_helper (void)	787	ecb_byteorder_helper (void)
571	{	788	{
572	/* the union code still generates code under pressure in gcc, */	789	/* the union code still generates code under pressure in gcc, */
573	/* but less than using pointers, and always seems to */	790	/* but less than using pointers, and always seems to */
574	/* successfully return a constant. */	791	/* successfully return a constant. */
575	/* the reason why we have this horrible preprocessor mess */	792	/* the reason why we have this horrible preprocessor mess */
576	/* is to avoid it in all cases, at least on common architectures */	793	/* is to avoid it in all cases, at least on common architectures */
577	/* or when using a recent enough gcc version (>= 4.6) */	794	/* or when using a recent enough gcc version (>= 4.6) */
578	#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
579	return 0x44;
580	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__	795	#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
		796	|| ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
		797	#define ECB_LITTLE_ENDIAN 1
581	return 0x44;	798	return 0x44332211;
582	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__	799	#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
		800	|| ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
		801	#define ECB_BIG_ENDIAN 1
583	return 0x11;	802	return 0x11223344;
584	#else	803	#else
585	union	804	union
586	{	805	{
		806	uint8_t c[4];
587	uint32_t i;	807	uint32_t u;
588	uint8_t c;
589	} u = { 0x11223344 };	808	} u = { 0x11, 0x22, 0x33, 0x44 };
590	return u.c;	809	return u.u;
591	#endif	810	#endif
592	}	811	}
593		812
594	ecb_inline ecb_bool ecb_big_endian (void) ecb_const;	813	ecb_inline ecb_const ecb_bool ecb_big_endian (void);
595	ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }	814	ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
596	ecb_inline ecb_bool ecb_little_endian (void) ecb_const;	815	ecb_inline ecb_const ecb_bool ecb_little_endian (void);
597	ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }	816	ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
		817
		818	/*****************************************************************************/
		819	/* unaligned load/store */
		820
		821	ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		822	ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		823	ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		824
		825	ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		826	ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		827	ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		828
		829	ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		830	ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		831	ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		832
		833	ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
		834	ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
		835	ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
		836
		837	ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
		838	ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
		839	ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
		840
		841	ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		842	ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		843	ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		844
		845	ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		846	ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		847	ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		848
		849	ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
		850	ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
		851	ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
		852
		853	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)); }
		854	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)); }
		855	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)); }
		856
		857	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)); }
		858	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)); }
		859	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)); }
		860
		861	#if ECB_CPP
		862
		863	inline uint8_t ecb_bswap (uint8_t v) { return v; }
		864	inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
		865	inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
		866	inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
		867
		868	template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		869	template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		870	template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
		871	template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
		872	template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
		873	template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
		874	template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
		875	template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
		876
		877	template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		878	template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		879	template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
		880	template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
		881	template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
		882	template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
		883	template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
		884	template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
		885
		886	#endif
		887
		888	/*****************************************************************************/
		889	/* pointer/integer hashing */
		890
		891	/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
		892	ecb_function_ uint32_t ecb_mix32 (uint32_t v);
		893	ecb_function_ uint32_t ecb_mix32 (uint32_t v)
		894	{
		895	v ^= v >> 16; v *= 0x7feb352dU;
		896	v ^= v >> 15; v *= 0x846ca68bU;
		897	v ^= v >> 16;
		898	return v;
		899	}
		900
		901	ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
		902	ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
		903	{
		904	v ^= v >> 16 ; v *= 0x43021123U;
		905	v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
		906	v ^= v >> 16 ;
		907	return v;
		908	}
		909
		910	/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
		911	ecb_function_ uint64_t ecb_mix64 (uint64_t v);
		912	ecb_function_ uint64_t ecb_mix64 (uint64_t v)
		913	{
		914	v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
		915	v ^= v >> 27; v *= 0x94d049bb133111ebU;
		916	v ^= v >> 31;
		917	return v;
		918	}
		919
		920	ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
		921	ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
		922	{
		923	v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
		924	v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
		925	v ^= v >> 30 ^ v >> 60;
		926	return v;
		927	}
		928
		929	ecb_function_ uintptr_t ecb_ptrmix (void *p);
		930	ecb_function_ uintptr_t ecb_ptrmix (void *p)
		931	{
		932	#if ECB_PTRSIZE <= 4
		933	return ecb_mix32 ((uint32_t)p);
		934	#else
		935	return ecb_mix64 ((uint64_t)p);
		936	#endif
		937	}
		938
		939	ecb_function_ void *ecb_ptrunmix (uintptr_t v);
		940	ecb_function_ void *ecb_ptrunmix (uintptr_t v)
		941	{
		942	#if ECB_PTRSIZE <= 4
		943	return (void *)ecb_unmix32 (v);
		944	#else
		945	return (void *)ecb_unmix64 (v);
		946	#endif
		947	}
		948
		949	#if ECB_CPP
		950
		951	template<typename T>
		952	inline uintptr_t ecb_ptrmix (T *p)
		953	{
		954	return ecb_ptrmix (static_cast<void *>(p));
		955	}
		956
		957	template<typename T>
		958	inline T *ecb_ptrunmix (uintptr_t v)
		959	{
		960	return static_cast<T *>(ecb_ptrunmix (v));
		961	}
		962
		963	#endif
		964
		965	/*****************************************************************************/
		966	/* gray code */
		967
		968	ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); }
		969	ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); }
		970	ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); }
		971	ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); }
		972
		973	ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g);
		974	ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g)
		975	{
		976	g ^= g >> 1;
		977	g ^= g >> 2;
		978	g ^= g >> 4;
		979
		980	return g;
		981	}
		982
		983	ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g);
		984	ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g)
		985	{
		986	g ^= g >> 1;
		987	g ^= g >> 2;
		988	g ^= g >> 4;
		989	g ^= g >> 8;
		990
		991	return g;
		992	}
		993
		994	ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g);
		995	ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g)
		996	{
		997	g ^= g >> 1;
		998	g ^= g >> 2;
		999	g ^= g >> 4;
		1000	g ^= g >> 8;
		1001	g ^= g >> 16;
		1002
		1003	return g;
		1004	}
		1005
		1006	ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g);
		1007	ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g)
		1008	{
		1009	g ^= g >> 1;
		1010	g ^= g >> 2;
		1011	g ^= g >> 4;
		1012	g ^= g >> 8;
		1013	g ^= g >> 16;
		1014	g ^= g >> 32;
		1015
		1016	return g;
		1017	}
		1018
		1019	#if ECB_CPP
		1020
		1021	ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); }
		1022	ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); }
		1023	ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); }
		1024	ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); }
		1025
		1026	ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); }
		1027	ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); }
		1028	ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); }
		1029	ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); }
		1030
		1031	#endif
		1032
		1033	/*****************************************************************************/
		1034	/* 2d hilbert curves */
		1035
		1036	/* algorithm from the book Hacker's Delight, modified to not */
		1037	/* run into undefined behaviour for n==16 */
		1038	static uint32_t
		1039	ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
		1040	{
		1041	uint32_t comp, swap, cs, t, sr;
		1042
		1043	/* pad s on the left (unused) bits with 01 (no change groups) */
		1044	s |= 0x55555555U << n << n;
		1045	/* "s shift right" */
		1046	sr = (s >> 1) & 0x55555555U;
		1047	/* compute complement and swap info in two-bit groups */
		1048	cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
		1049
		1050	/* parallel prefix xor op to propagate both complement
		1051	* and swap info together from left to right (there is
		1052	* no step "cs ^= cs >> 1", so in effect it computes
		1053	* two independent parallel prefix operations on two
		1054	* interleaved sets of sixteen bits).
		1055	*/
		1056	cs ^= cs >> 2;
		1057	cs ^= cs >> 4;
		1058	cs ^= cs >> 8;
		1059	cs ^= cs >> 16;
		1060
		1061	/* separate swap and complement bits */
		1062	swap = cs & 0x55555555U;
		1063	comp = (cs >> 1) & 0x55555555U;
		1064
		1065	/* calculate coordinates in odd and even bit positions */
		1066	t = (s & swap) ^ comp;
		1067	s = s ^ sr ^ t ^ (t << 1);
		1068
		1069	/* unpad/clear out any junk on the left */
		1070	s = s & ((1 << n << n) - 1);
		1071
		1072	/* Now "unshuffle" to separate the x and y bits. */
		1073	t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
		1074	t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
		1075	t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
		1076	t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
		1077
		1078	/* now s contains two 16-bit coordinates */
		1079	return s;
		1080	}
		1081
		1082	/* 64 bit, a straightforward extension to the 32 bit case */
		1083	static uint64_t
		1084	ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
		1085	{
		1086	uint64_t comp, swap, cs, t, sr;
		1087
		1088	/* pad s on the left (unused) bits with 01 (no change groups) */
		1089	s |= 0x5555555555555555U << n << n;
		1090	/* "s shift right" */
		1091	sr = (s >> 1) & 0x5555555555555555U;
		1092	/* compute complement and swap info in two-bit groups */
		1093	cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
		1094
		1095	/* parallel prefix xor op to propagate both complement
		1096	* and swap info together from left to right (there is
		1097	* no step "cs ^= cs >> 1", so in effect it computes
		1098	* two independent parallel prefix operations on two
		1099	* interleaved sets of thirty-two bits).
		1100	*/
		1101	cs ^= cs >> 2;
		1102	cs ^= cs >> 4;
		1103	cs ^= cs >> 8;
		1104	cs ^= cs >> 16;
		1105	cs ^= cs >> 32;
		1106
		1107	/* separate swap and complement bits */
		1108	swap = cs & 0x5555555555555555U;
		1109	comp = (cs >> 1) & 0x5555555555555555U;
		1110
		1111	/* calculate coordinates in odd and even bit positions */
		1112	t = (s & swap) ^ comp;
		1113	s = s ^ sr ^ t ^ (t << 1);
		1114
		1115	/* unpad/clear out any junk on the left */
		1116	s = s & ((1 << n << n) - 1);
		1117
		1118	/* Now "unshuffle" to separate the x and y bits. */
		1119	t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
		1120	t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
		1121	t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
		1122	t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
		1123	t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
		1124
		1125	/* now s contains two 32-bit coordinates */
		1126	return s;
		1127	}
		1128
		1129	/* algorithm from the book Hacker's Delight, but a similar algorithm*/
		1130	/* is given in https://doi.org/10.1002/spe.4380160103 */
		1131	/* this has been slightly improved over the original version */
		1132	ecb_function_ uint32_t
		1133	ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
		1134	{
		1135	uint32_t row;
		1136	uint32_t state = 0;
		1137	uint32_t s = 0;
		1138
		1139	do
		1140	{
		1141	--n;
		1142
		1143	row = 4 * state
		1144	| (2 & (xy >> n >> 15))
		1145	| (1 & (xy >> n ));
		1146
		1147	/* these funky constants are lookup tables for two-bit values */
		1148	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1149	state = (0x8fe65831U >> 2 * row) & 3;
		1150	}
		1151	while (n > 0);
		1152
		1153	return s;
		1154	}
		1155
		1156	/* 64 bit, essentially the same as 32 bit */
		1157	ecb_function_ uint64_t
		1158	ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
		1159	{
		1160	uint32_t row;
		1161	uint32_t state = 0;
		1162	uint64_t s = 0;
		1163
		1164	do
		1165	{
		1166	--n;
		1167
		1168	row = 4 * state
		1169	| (2 & (xy >> n >> 31))
		1170	| (1 & (xy >> n ));
		1171
		1172	/* these funky constants are lookup tables for two-bit values */
		1173	s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
		1174	state = (0x8fe65831U >> 2 * row) & 3;
		1175	}
		1176	while (n > 0);
		1177
		1178	return s;
		1179	}
		1180
		1181	/*****************************************************************************/
		1182	/* division */
598		1183
599	#if ECB_GCC_VERSION(3,0) || ECB_C99	1184	#if ECB_GCC_VERSION(3,0) || ECB_C99
		1185	/* C99 tightened the definition of %, so we can use a more efficient version */
600	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))	1186	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
601	#else	1187	#else
602	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))	1188	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
603	#endif	1189	#endif
604		1190
605	#if __cplusplus	1191	#if ECB_CPP
606	template<typename T>	1192	template<typename T>
607	static inline T ecb_div_rd (T val, T div)	1193	static inline T ecb_div_rd (T val, T div)
608	{	1194	{
609	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;	1195	return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
610	}	1196	}
…		…
615	}	1201	}
616	#else	1202	#else
617	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))	1203	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
618	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))	1204	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
619	#endif	1205	#endif
		1206
		1207	/*****************************************************************************/
		1208	/* array length */
620		1209
621	#if ecb_cplusplus_does_not_suck	1210	#if ecb_cplusplus_does_not_suck
622	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */	1211	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
623	template<typename T, int N>	1212	template<typename T, int N>
624	static inline int ecb_array_length (const T (&arr)[N])	1213	static inline int ecb_array_length (const T (&arr)[N])
…		…
627	}	1216	}
628	#else	1217	#else
629	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))	1218	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
630	#endif	1219	#endif
631		1220
		1221	/*****************************************************************************/
		1222	/* IEEE 754-2008 half float conversions */
		1223
		1224	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
		1225	ecb_function_ ecb_const uint32_t
		1226	ecb_binary16_to_binary32 (uint32_t x)
		1227	{
		1228	unsigned int s = (x & 0x8000) << (31 - 15);
		1229	int e = (x >> 10) & 0x001f;
		1230	unsigned int m = x & 0x03ff;
		1231
		1232	if (ecb_expect_false (e == 31))
		1233	/* infinity or NaN */
		1234	e = 255 - (127 - 15);
		1235	else if (ecb_expect_false (!e))
		1236	{
		1237	if (ecb_expect_true (!m))
		1238	/* zero, handled by code below by forcing e to 0 */
		1239	e = 0 - (127 - 15);
		1240	else
		1241	{
		1242	/* subnormal, renormalise */
		1243	unsigned int s = 10 - ecb_ld32 (m);
		1244
		1245	m = (m << s) & 0x3ff; /* mask implicit bit */
		1246	e -= s - 1;
		1247	}
		1248	}
		1249
		1250	/* e and m now are normalised, or zero, (or inf or nan) */
		1251	e += 127 - 15;
		1252
		1253	return s | (e << 23) | (m << (23 - 10));
		1254	}
		1255
		1256	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
		1257	ecb_function_ ecb_const uint16_t
		1258	ecb_binary32_to_binary16 (uint32_t x)
		1259	{
		1260	unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
		1261	int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
		1262	unsigned int m = x & 0x007fffff;
		1263
		1264	x &= 0x7fffffff;
		1265
		1266	/* if it's within range of binary16 normals, use fast path */
		1267	if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
		1268	{
		1269	/* mantissa round-to-even */
		1270	m += 0x00000fff + ((m >> (23 - 10)) & 1);
		1271
		1272	/* handle overflow */
		1273	if (ecb_expect_false (m >= 0x00800000))
		1274	{
		1275	m >>= 1;
		1276	e += 1;
		1277	}
		1278
		1279	return s | (e << 10) | (m >> (23 - 10));
		1280	}
		1281
		1282	/* handle large numbers and infinity */
		1283	if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
		1284	return s | 0x7c00;
		1285
		1286	/* handle zero, subnormals and small numbers */
		1287	if (ecb_expect_true (x < 0x38800000))
		1288	{
		1289	/* zero */
		1290	if (ecb_expect_true (!x))
		1291	return s;
		1292
		1293	/* handle subnormals */
		1294
		1295	/* too small, will be zero */
		1296	if (e < (14 - 24)) /* might not be sharp, but is good enough */
		1297	return s;
		1298
		1299	m |= 0x00800000; /* make implicit bit explicit */
		1300
		1301	/* very tricky - we need to round to the nearest e (+10) bit value */
		1302	{
		1303	unsigned int bits = 14 - e;
		1304	unsigned int half = (1 << (bits - 1)) - 1;
		1305	unsigned int even = (m >> bits) & 1;
		1306
		1307	/* if this overflows, we will end up with a normalised number */
		1308	m = (m + half + even) >> bits;
		1309	}
		1310
		1311	return s | m;
		1312	}
		1313
		1314	/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
		1315	m >>= 13;
		1316
		1317	return s | 0x7c00 | m | !m;
		1318	}
		1319
		1320	/*******************************************************************************/
		1321	/* fast integer to ascii */
		1322
		1323	/*
		1324	* This code is pretty complicated because it is general. The idea behind it,
		1325	* however, is pretty simple: first, the number is multiplied with a scaling
		1326	* factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
		1327	* number with the first digit in the upper bits.
		1328	* Then this digit is converted to text and masked out. The resulting number
		1329	* is then multiplied by 10, by multiplying the fixed point representation
		1330	* by 5 and shifting the (binary) decimal point one to the right, so a 4.28
		1331	* format becomes 5.27, 6.26 and so on.
		1332	* The rest involves only advancing the pointer if we already generated a
		1333	* non-zero digit, so leading zeroes are overwritten.
		1334	*/
		1335
		1336	/* simply return a mask with "bits" bits set */
		1337	#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
		1338
		1339	/* oputput a single digit. maskvalue is 10**digitidx */
		1340	#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
		1341	if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
		1342	{ \
		1343	char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
		1344	*ptr = digit + '0'; /* output it */ \
		1345	nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
		1346	ptr += nz; /* output digit only if non-zero digit seen */ \
		1347	x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
		1348	}
		1349
		1350	/* convert integer to fixed point format and multiply out digits, highest first */
		1351	/* requires magic constants: max. digits and number of bits after the decimal point */
		1352	#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
		1353	ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
		1354	{ \
		1355	char nz = lz; /* non-zero digit seen? */ \
		1356	/* convert to x.bits fixed-point */ \
		1357	type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
		1358	/* output up to 10 digits */ \
		1359	ecb_i2a_digit (type,bits,digitmask, 1, 0); \
		1360	ecb_i2a_digit (type,bits,digitmask, 10, 1); \
		1361	ecb_i2a_digit (type,bits,digitmask, 100, 2); \
		1362	ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
		1363	ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
		1364	ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
		1365	ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
		1366	ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
		1367	ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
		1368	ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
		1369	return ptr; \
		1370	}
		1371
		1372	/* predefined versions of the above, for various digits */
		1373	/* ecb_i2a_xN = almost N digits, limit defined by macro */
		1374	/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
		1375	/* ecb_i2a_0N = exactly N digits, including leading zeroes */
		1376
		1377	/* non-leading-zero versions, limited range */
		1378	#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
		1379	#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
		1380	ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
		1381	ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
		1382
		1383	/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
		1384	ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
		1385	ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
		1386	ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
		1387	ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
		1388	ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
		1389	ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
		1390	ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
		1391	ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
		1392
		1393	/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
		1394	ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
		1395	ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
		1396	ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
		1397	ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
		1398	ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
		1399	ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
		1400	ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
		1401	ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
		1402
		1403	#define ECB_I2A_I32_DIGITS 11
		1404	#define ECB_I2A_U32_DIGITS 10
		1405	#define ECB_I2A_I64_DIGITS 20
		1406	#define ECB_I2A_U64_DIGITS 21
		1407	#define ECB_I2A_MAX_DIGITS 21
		1408
		1409	ecb_inline char *
		1410	ecb_i2a_u32 (char *ptr, uint32_t u)
		1411	{
		1412	#if ECB_64BIT_NATIVE
		1413	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1414	ptr = ecb_i2a_x10 (ptr, u);
		1415	else /* x10 almost, but not fully, covers 32 bit */
		1416	{
		1417	uint32_t u1 = u % 1000000000;
		1418	uint32_t u2 = u / 1000000000;
		1419
		1420	*ptr++ = u2 + '0';
		1421	ptr = ecb_i2a_09 (ptr, u1);
		1422	}
		1423	#else
		1424	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1425	ecb_i2a_x5 (ptr, u);
		1426	else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
		1427	{
		1428	uint32_t u1 = u % 10000;
		1429	uint32_t u2 = u / 10000;
		1430
		1431	ptr = ecb_i2a_x5 (ptr, u2);
		1432	ptr = ecb_i2a_04 (ptr, u1);
		1433	}
		1434	else
		1435	{
		1436	uint32_t u1 = u % 10000;
		1437	uint32_t ua = u / 10000;
		1438	uint32_t u2 = ua % 10000;
		1439	uint32_t u3 = ua / 10000;
		1440
		1441	ptr = ecb_i2a_2 (ptr, u3);
		1442	ptr = ecb_i2a_04 (ptr, u2);
		1443	ptr = ecb_i2a_04 (ptr, u1);
		1444	}
		1445	#endif
		1446
		1447	return ptr;
		1448	}
		1449
		1450	ecb_inline char *
		1451	ecb_i2a_i32 (char *ptr, int32_t v)
		1452	{
		1453	*ptr = '-'; ptr += v < 0;
		1454	uint32_t u = v < 0 ? -(uint32_t)v : v;
		1455
		1456	#if ECB_64BIT_NATIVE
		1457	ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
		1458	#else
		1459	ptr = ecb_i2a_u32 (ptr, u);
		1460	#endif
		1461
		1462	return ptr;
		1463	}
		1464
		1465	ecb_inline char *
		1466	ecb_i2a_u64 (char *ptr, uint64_t u)
		1467	{
		1468	#if ECB_64BIT_NATIVE
		1469	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1470	ptr = ecb_i2a_x10 (ptr, u);
		1471	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1472	{
		1473	uint64_t u1 = u % 1000000000;
		1474	uint64_t u2 = u / 1000000000;
		1475
		1476	ptr = ecb_i2a_x10 (ptr, u2);
		1477	ptr = ecb_i2a_09 (ptr, u1);
		1478	}
		1479	else
		1480	{
		1481	uint64_t u1 = u % 1000000000;
		1482	uint64_t ua = u / 1000000000;
		1483	uint64_t u2 = ua % 1000000000;
		1484	uint64_t u3 = ua / 1000000000;
		1485
		1486	ptr = ecb_i2a_2 (ptr, u3);
		1487	ptr = ecb_i2a_09 (ptr, u2);
		1488	ptr = ecb_i2a_09 (ptr, u1);
		1489	}
		1490	#else
		1491	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1492	ptr = ecb_i2a_x5 (ptr, u);
		1493	else
		1494	{
		1495	uint64_t u1 = u % 10000;
		1496	uint64_t u2 = u / 10000;
		1497
		1498	ptr = ecb_i2a_u64 (ptr, u2);
		1499	ptr = ecb_i2a_04 (ptr, u1);
		1500	}
		1501	#endif
		1502
		1503	return ptr;
		1504	}
		1505
		1506	ecb_inline char *
		1507	ecb_i2a_i64 (char *ptr, int64_t v)
		1508	{
		1509	*ptr = '-'; ptr += v < 0;
		1510	uint64_t u = v < 0 ? -(uint64_t)v : v;
		1511
		1512	#if ECB_64BIT_NATIVE
		1513	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1514	ptr = ecb_i2a_x10 (ptr, u);
		1515	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1516	{
		1517	uint64_t u1 = u % 1000000000;
		1518	uint64_t u2 = u / 1000000000;
		1519
		1520	ptr = ecb_i2a_x10 (ptr, u2);
		1521	ptr = ecb_i2a_09 (ptr, u1);
		1522	}
		1523	else
		1524	{
		1525	uint64_t u1 = u % 1000000000;
		1526	uint64_t ua = u / 1000000000;
		1527	uint64_t u2 = ua % 1000000000;
		1528	uint64_t u3 = ua / 1000000000;
		1529
		1530	/* 2**31 is 19 digits, so the top is exactly one digit */
		1531	*ptr++ = u3 + '0';
		1532	ptr = ecb_i2a_09 (ptr, u2);
		1533	ptr = ecb_i2a_09 (ptr, u1);
		1534	}
		1535	#else
		1536	ptr = ecb_i2a_u64 (ptr, u);
		1537	#endif
		1538
		1539	return ptr;
		1540	}
		1541
632	/*******************************************************************************/	1542	/*******************************************************************************/
633	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */	1543	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
634		1544
635	/* basically, everything uses "ieee pure-endian" floating point numbers */	1545	/* basically, everything uses "ieee pure-endian" floating point numbers */
636	/* the only noteworthy exception is ancient armle, which uses order 43218765 */	1546	/* the only noteworthy exception is ancient armle, which uses order 43218765 */
637	#if 0 \	1547	#if 0 \
638	|| __i386 || __i386__ \	1548	|| __i386 || __i386__ \
639	|| __amd64 || __amd64__ || __x86_64 || __x86_64__ \	1549	|| ECB_GCC_AMD64 \
640	|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \	1550	|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
641	|| defined __s390__ || defined __s390x__ \	1551	|| defined __s390__ || defined __s390x__ \
642	|| defined __mips__ \	1552	|| defined __mips__ \
643	|| defined __alpha__ \	1553	|| defined __alpha__ \
644	|| defined __hppa__ \	1554	|| defined __hppa__ \
645	|| defined __ia64__ \	1555	|| defined __ia64__ \
646	|| defined __m68k__ \	1556	|| defined __m68k__ \
647	|| defined __m88k__ \	1557	|| defined __m88k__ \
648	|| defined __sh__ \	1558	|| defined __sh__ \
649	|| defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \	1559	|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
650	|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \	1560	|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
651	|| defined __aarch64__	1561	|| defined __aarch64__
652	#define ECB_STDFP 1	1562	#define ECB_STDFP 1
653	#include <string.h> /* for memcpy */
654	#else	1563	#else
655	#define ECB_STDFP 0	1564	#define ECB_STDFP 0
656	#endif	1565	#endif
657		1566
658	#ifndef ECB_NO_LIBM	1567	#ifndef ECB_NO_LIBM
…		…
672	#define ECB_NAN ECB_INFINITY	1581	#define ECB_NAN ECB_INFINITY
673	#endif	1582	#endif
674		1583
675	#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L	1584	#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
676	#define ecb_ldexpf(x,e) ldexpf ((x), (e))	1585	#define ecb_ldexpf(x,e) ldexpf ((x), (e))
		1586	#define ecb_frexpf(x,e) frexpf ((x), (e))
677	#else	1587	#else
678	#define ecb_ldexpf(x,e) (float) ldexp ((double)(x), (e))	1588	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
		1589	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
679	#endif	1590	#endif
680
681	/* converts an ieee half/binary16 to a float */
682	ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
683	ecb_function_ float
684	ecb_binary16_to_float (uint16_t x)
685	{
686	int e = (x >> 10) & 0x1f;
687	int m = x & 0x3ff;
688	float r;
689
690	if (!e ) r = ecb_ldexpf (m , -24);
691	else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
692	else if (m ) r = ECB_NAN;
693	else r = ECB_INFINITY;
694
695	return x & 0x8000 ? -r : r;
696	}
697		1591
698	/* convert a float to ieee single/binary32 */	1592	/* convert a float to ieee single/binary32 */
699	ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;	1593	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
700	ecb_function_ uint32_t	1594	ecb_function_ ecb_const uint32_t
701	ecb_float_to_binary32 (float x)	1595	ecb_float_to_binary32 (float x)
702	{	1596	{
703	uint32_t r;	1597	uint32_t r;
704		1598
705	#if ECB_STDFP	1599	#if ECB_STDFP
…		…
712	if (x == 0e0f ) return 0x00000000U;	1606	if (x == 0e0f ) return 0x00000000U;
713	if (x > +3.40282346638528860e+38f) return 0x7f800000U;	1607	if (x > +3.40282346638528860e+38f) return 0x7f800000U;
714	if (x < -3.40282346638528860e+38f) return 0xff800000U;	1608	if (x < -3.40282346638528860e+38f) return 0xff800000U;
715	if (x != x ) return 0x7fbfffffU;	1609	if (x != x ) return 0x7fbfffffU;
716		1610
717	m = frexpf (x, &e) * 0x1000000U;	1611	m = ecb_frexpf (x, &e) * 0x1000000U;
718		1612
719	r = m & 0x80000000U;	1613	r = m & 0x80000000U;
720		1614
721	if (r)	1615	if (r)
722	m = -m;	1616	m = -m;
…		…
734		1628
735	return r;	1629	return r;
736	}	1630	}
737		1631
738	/* converts an ieee single/binary32 to a float */	1632	/* converts an ieee single/binary32 to a float */
739	ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;	1633	ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
740	ecb_function_ float	1634	ecb_function_ ecb_const float
741	ecb_binary32_to_float (uint32_t x)	1635	ecb_binary32_to_float (uint32_t x)
742	{	1636	{
743	float r;	1637	float r;
744		1638
745	#if ECB_STDFP	1639	#if ECB_STDFP
…		…
764		1658
765	return r;	1659	return r;
766	}	1660	}
767		1661
768	/* convert a double to ieee double/binary64 */	1662	/* convert a double to ieee double/binary64 */
769	ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;	1663	ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
770	ecb_function_ uint64_t	1664	ecb_function_ ecb_const uint64_t
771	ecb_double_to_binary64 (double x)	1665	ecb_double_to_binary64 (double x)
772	{	1666	{
773	uint64_t r;	1667	uint64_t r;
774		1668
775	#if ECB_STDFP	1669	#if ECB_STDFP
…		…
804		1698
805	return r;	1699	return r;
806	}	1700	}
807		1701
808	/* converts an ieee double/binary64 to a double */	1702	/* converts an ieee double/binary64 to a double */
809	ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;	1703	ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
810	ecb_function_ double	1704	ecb_function_ ecb_const double
811	ecb_binary64_to_double (uint64_t x)	1705	ecb_binary64_to_double (uint64_t x)
812	{	1706	{
813	double r;	1707	double r;
814		1708
815	#if ECB_STDFP	1709	#if ECB_STDFP
…		…
833	#endif	1727	#endif
834		1728
835	return r;	1729	return r;
836	}	1730	}
837		1731
838	#endif	1732	/* convert a float to ieee half/binary16 */
		1733	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
		1734	ecb_function_ ecb_const uint16_t
		1735	ecb_float_to_binary16 (float x)
		1736	{
		1737	return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
		1738	}
839		1739
840	#endif	1740	/* convert an ieee half/binary16 to float */
		1741	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
		1742	ecb_function_ ecb_const float
		1743	ecb_binary16_to_float (uint16_t x)
		1744	{
		1745	return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
		1746	}
841		1747
		1748	#endif
		1749
		1750	#endif
		1751

Diff Legend

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