[ViewVC] Diff of: cvs/libecb/ecb.h

Comparing libecb/ecb.h (file contents):
Revision 1.166 by root, Sun Aug 9 00:10:21 2015 UTC vs.
Revision 1.205 by root, Fri Mar 25 14:21:14 2022 UTC

…		…
1	/*	1	/*
2	* libecb - http://software.schmorp.de/pkg/libecb	2	* libecb - http://software.schmorp.de/pkg/libecb
3	*	3	*
4	* Copyright (©) 2009-2015 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
79	#define ECB_GCC_AMD64 (__amd64 \|\| __amd64__ \|\| __x86_64 \|\| __x86_64__)	89	#define ECB_GCC_AMD64 (__amd64 \|\| __amd64__ \|\| __x86_64 \|\| __x86_64__)
80	#define ECB_MSVC_AMD64 (_M_AMD64 \|\| _M_X64)	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
81		99
82	/* work around x32 idiocy by defining proper macros */	100	/* work around x32 idiocy by defining proper macros */
83	#if ECB_GCC_AMD64 \|\| ECB_MSVC_AMD64	101	#if ECB_GCC_AMD64 \|\| ECB_MSVC_AMD64
84	#if _ILP32	102	#if _ILP32
85	#define ECB_AMD64_X32 1	103	#define ECB_AMD64_X32 1
86	#else	104	#else
87	#define ECB_AMD64 1	105	#define ECB_AMD64 1
88	#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
89	#endif	113	#endif
90		114
91	/* many compilers define _GNUC_ to some versions but then only implement	115	/* many compilers define _GNUC_ to some versions but then only implement
92	* what their idiot authors think are the "more important" extensions,	116	* what their idiot authors think are the "more important" extensions,
93	* causing enormous grief in return for some better fake benchmark numbers.	117	* causing enormous grief in return for some better fake benchmark numbers.
…		…
115	#define ECB_CLANG_EXTENSION(x) 0	139	#define ECB_CLANG_EXTENSION(x) 0
116	#endif	140	#endif
117		141
118	#define ECB_CPP (__cplusplus+0)	142	#define ECB_CPP (__cplusplus+0)
119	#define ECB_CPP11 (__cplusplus >= 201103L)	143	#define ECB_CPP11 (__cplusplus >= 201103L)
		144	#define ECB_CPP14 (__cplusplus >= 201402L)
		145	#define ECB_CPP17 (__cplusplus >= 201703L)
120		146
121	#if ECB_CPP	147	#if ECB_CPP
122	#define ECB_C 0	148	#define ECB_C 0
123	#define ECB_STDC_VERSION 0	149	#define ECB_STDC_VERSION 0
124	#else	150	#else
…		…
126	#define ECB_STDC_VERSION __STDC_VERSION__	152	#define ECB_STDC_VERSION __STDC_VERSION__
127	#endif	153	#endif
128		154
129	#define ECB_C99 (ECB_STDC_VERSION >= 199901L)	155	#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
130	#define ECB_C11 (ECB_STDC_VERSION >= 201112L)	156	#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
		157	#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
131		158
132	#if ECB_CPP	159	#if ECB_CPP
133	#define ECB_EXTERN_C extern "C"	160	#define ECB_EXTERN_C extern "C"
134	#define ECB_EXTERN_C_BEG ECB_EXTERN_C {	161	#define ECB_EXTERN_C_BEG ECB_EXTERN_C {
135	#define ECB_EXTERN_C_END }	162	#define ECB_EXTERN_C_END }
…		…
155	/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */	182	/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
156	#if __xlC__ && ECB_CPP	183	#if __xlC__ && ECB_CPP
157	#include <builtins.h>	184	#include <builtins.h>
158	#endif	185	#endif
159		186
		187	#if 1400 <= _MSC_VER
		188	#include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
		189	#endif
		190
160	#ifndef ECB_MEMORY_FENCE	191	#ifndef ECB_MEMORY_FENCE
161	#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")
162	#if __i386 \|\| __i386__	194	#if __i386 \|\| __i386__
163	#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")
164	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")	196	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
165	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")	197	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
166	#elif ECB_GCC_AMD64	198	#elif ECB_GCC_AMD64
167	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")	199	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
168	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")	200	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
169	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")	201	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
170	#elif __powerpc__ \|\| __ppc__ \|\| __powerpc64__ \|\| __ppc64__	202	#elif __powerpc__ \|\| __ppc__ \|\| __powerpc64__ \|\| __ppc64__
171	#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 */
172	#elif defined __ARM_ARCH_6__ \|\| defined __ARM_ARCH_6J__ \	211	#elif defined __ARM_ARCH_6__ \|\| defined __ARM_ARCH_6J__ \
173	\|\| defined __ARM_ARCH_6K__ \|\| defined __ARM_ARCH_6ZK__	212	\|\| defined __ARM_ARCH_6K__ \|\| defined __ARM_ARCH_6ZK__ \
		213	\|\| defined __ARM_ARCH_6T2__
174	#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")
175	#elif defined __ARM_ARCH_7__ \|\| defined __ARM_ARCH_7A__ \	215	#elif defined __ARM_ARCH_7__ \|\| defined __ARM_ARCH_7A__ \
176	\|\| defined __ARM_ARCH_7M__ \|\| defined __ARM_ARCH_7R__	216	\|\| defined __ARM_ARCH_7R__ \|\| defined __ARM_ARCH_7M__
177	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")	217	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
178	#elif __aarch64__	218	#elif __aarch64__
179	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")	219	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
180	#elif (__sparc \|\| __sparc__) && !(__sparc_v8__ \|\| defined __sparcv8)	220	#elif (__sparc \|\| __sparc__) && !(__sparc_v8__ \|\| defined __sparcv8)
181	#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")
…		…
208	#if ECB_GCC_VERSION(4,7)	248	#if ECB_GCC_VERSION(4,7)
209	/* see comment below (stdatomic.h) about the C11 memory model. */	249	/* see comment below (stdatomic.h) about the C11 memory model. */
210	#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)	250	#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
211	#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)	251	#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
212	#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)
213		255
214	#elif ECB_CLANG_EXTENSION(c_atomic)	256	#elif ECB_CLANG_EXTENSION(c_atomic)
215	/* see comment below (stdatomic.h) about the C11 memory model. */	257	/* see comment below (stdatomic.h) about the C11 memory model. */
216	#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)	258	#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
217	#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)	259	#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
218	#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)
219		263
220	#elif ECB_GCC_VERSION(4,4) \|\| defined __INTEL_COMPILER \|\| defined __clang__	264	#elif ECB_GCC_VERSION(4,4) \|\| defined __INTEL_COMPILER \|\| defined __clang__
221	#define ECB_MEMORY_FENCE __sync_synchronize ()	265	#define ECB_MEMORY_FENCE __sync_synchronize ()
222	#elif _MSC_VER >= 1500 /* VC++ 2008 */	266	#elif _MSC_VER >= 1500 /* VC++ 2008 */
223	/* 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... */
…		…
233	#elif defined _WIN32	277	#elif defined _WIN32
234	#include <WinNT.h>	278	#include <WinNT.h>
235	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */	279	#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
236	#elif __SUNPRO_C >= 0x5110 \|\| __SUNPRO_CC >= 0x5110	280	#elif __SUNPRO_C >= 0x5110 \|\| __SUNPRO_CC >= 0x5110
237	#include <mbarrier.h>	281	#include <mbarrier.h>
238	#define ECB_MEMORY_FENCE __machine_rw_barrier ()	282	#define ECB_MEMORY_FENCE __machine_rw_barrier ()
239	#define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()	283	#define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
240	#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 ()
241	#elif __xlC__	286	#elif __xlC__
242	#define ECB_MEMORY_FENCE __sync ()	287	#define ECB_MEMORY_FENCE __sync ()
243	#endif	288	#endif
244	#endif	289	#endif
245		290
246	#ifndef ECB_MEMORY_FENCE	291	#ifndef ECB_MEMORY_FENCE
247	#if ECB_C11 && !defined __STDC_NO_ATOMICS__	292	#if ECB_C11 && !defined __STDC_NO_ATOMICS__
248	/* 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, */
249	/* not just C11 atomics and atomic accesses */	294	/* not just C11 atomics and atomic accesses */
250	#include <stdatomic.h>	295	#include <stdatomic.h>
251	/* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
252	/* any fence other than seq_cst, which isn't very efficient for us. */
253	/* Why that is, we don't know - either the C11 memory model is quite useless */
254	/* for most usages, or gcc and clang have a bug */
255	/* I currently lean towards the latter, and inefficiently implement */
256	/* all three of ecb's fences as a seq_cst fence */
257	/* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
258	/* for all __atomic_thread_fence's except seq_cst */
259	#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)
260	#endif	299	#endif
261	#endif	300	#endif
262		301
263	#ifndef ECB_MEMORY_FENCE	302	#ifndef ECB_MEMORY_FENCE
264	#if !ECB_AVOID_PTHREADS	303	#if !ECB_AVOID_PTHREADS
…		…
284		323
285	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE	324	#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
286	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE	325	#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
287	#endif	326	#endif
288		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
289	/*****************************************************************************/	332	/*****************************************************************************/
290		333
291	#if ECB_CPP	334	#if ECB_CPP
292	#define ecb_inline static inline	335	#define ecb_inline static inline
293	#elif ECB_GCC_VERSION(2,5)	336	#elif ECB_GCC_VERSION(2,5)
…		…
423	#else	466	#else
424	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);	467	ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
425	ecb_function_ ecb_const int	468	ecb_function_ ecb_const int
426	ecb_ctz32 (uint32_t x)	469	ecb_ctz32 (uint32_t x)
427	{	470	{
		471	#if 1400 <= _MSC_VER && (_M_IX86 \|\| _M_X64 \|\| _M_IA64 \|\| _M_ARM)
		472	unsigned long r;
		473	_BitScanForward (&r, x);
		474	return (int)r;
		475	#else
428	int r = 0;	476	int r = 0;
		477
		478	/* todo: use david seal's algorithm */
429		479
430	x &= ~x + 1; /* this isolates the lowest bit */	480	x &= ~x + 1; /* this isolates the lowest bit */
431		481
432	#if ECB_branchless_on_i386	482	#if ECB_branchless_on_i386
433	r += !!(x & 0xaaaaaaaa) << 0;	483	r += !!(x & 0xaaaaaaaa) << 0;
…		…
442	if (x & 0xff00ff00) r += 8;	492	if (x & 0xff00ff00) r += 8;
443	if (x & 0xffff0000) r += 16;	493	if (x & 0xffff0000) r += 16;
444	#endif	494	#endif
445		495
446	return r;	496	return r;
		497	#endif
447	}	498	}
448		499
449	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);	500	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
450	ecb_function_ ecb_const int	501	ecb_function_ ecb_const int
451	ecb_ctz64 (uint64_t x)	502	ecb_ctz64 (uint64_t x)
452	{	503	{
		504	#if 1400 <= _MSC_VER && (_M_X64 \|\| _M_IA64 \|\| _M_ARM)
		505	unsigned long r;
		506	_BitScanForward64 (&r, x);
		507	return (int)r;
		508	#else
453	int shift = x & 0xffffffffU ? 0 : 32;	509	int shift = x & 0xffffffff ? 0 : 32;
454	return ecb_ctz32 (x >> shift) + shift;	510	return ecb_ctz32 (x >> shift) + shift;
		511	#endif
455	}	512	}
456		513
457	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);	514	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
458	ecb_function_ ecb_const int	515	ecb_function_ ecb_const int
459	ecb_popcount32 (uint32_t x)	516	ecb_popcount32 (uint32_t x)
…		…
467	}	524	}
468		525
469	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);	526	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
470	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)	527	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
471	{	528	{
		529	#if 1400 <= _MSC_VER && (_M_IX86 \|\| _M_X64 \|\| _M_IA64 \|\| _M_ARM)
		530	unsigned long r;
		531	_BitScanReverse (&r, x);
		532	return (int)r;
		533	#else
472	int r = 0;	534	int r = 0;
473		535
474	if (x >> 16) { x >>= 16; r += 16; }	536	if (x >> 16) { x >>= 16; r += 16; }
475	if (x >> 8) { x >>= 8; r += 8; }	537	if (x >> 8) { x >>= 8; r += 8; }
476	if (x >> 4) { x >>= 4; r += 4; }	538	if (x >> 4) { x >>= 4; r += 4; }
477	if (x >> 2) { x >>= 2; r += 2; }	539	if (x >> 2) { x >>= 2; r += 2; }
478	if (x >> 1) { r += 1; }	540	if (x >> 1) { r += 1; }
479		541
480	return r;	542	return r;
		543	#endif
481	}	544	}
482		545
483	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);	546	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
484	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)	547	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
485	{	548	{
		549	#if 1400 <= _MSC_VER && (_M_X64 \|\| _M_IA64 \|\| _M_ARM)
		550	unsigned long r;
		551	_BitScanReverse64 (&r, x);
		552	return (int)r;
		553	#else
486	int r = 0;	554	int r = 0;
487		555
488	if (x >> 32) { x >>= 32; r += 32; }	556	if (x >> 32) { x >>= 32; r += 32; }
489		557
490	return r + ecb_ld32 (x);	558	return r + ecb_ld32 (x);
		559	#endif
491	}	560	}
492	#endif	561	#endif
493		562
494	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);	563	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
495	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }	564	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
…		…
524	x = ( x >> 16 ) \| ( x << 16);	593	x = ( x >> 16 ) \| ( x << 16);
525		594
526	return x;	595	return x;
527	}	596	}
528		597
529	/* popcount64 is only available on 64 bit cpus as gcc builtin */
530	/* so for this version we are lazy */
531	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);	598	ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
532	ecb_function_ ecb_const int	599	ecb_function_ ecb_const int
533	ecb_popcount64 (uint64_t x)	600	ecb_popcount64 (uint64_t x)
534	{	601	{
		602	/* popcount64 is only available on 64 bit cpus as gcc builtin. */
		603	/* also, gcc/clang make this surprisingly difficult to use */
		604	#if __LP64__ && (ECB_GCC_VERSION(3,4) \|\| ECB_CLANG_BUILTIN (__builtin_popcountl))
		605	return __builtin_popcountl (x);
		606	#else
535	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);	607	return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
		608	#endif
536	}	609	}
537		610
538	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);	611	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
539	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);	612	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
540	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);	613	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
…		…
542	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);	615	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
543	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);	616	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
544	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);	617	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
545	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);	618	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
546		619
547	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) \| (x << count); }	620	ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) \| (x << (count & 7)); }
548	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) \| (x >> count); }	621	ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) \| (x >> (count & 7)); }
549	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) \| (x << count); }	622	ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) \| (x << (count & 15)); }
550	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) \| (x >> count); }	623	ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) \| (x >> (count & 15)); }
551	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) \| (x << count); }	624	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) \| (x << (count & 31)); }
552	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) \| (x >> count); }	625	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) \| (x >> (count & 31)); }
553	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) \| (x << count); }	626	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) \| (x << (count & 63)); }
554	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) \| (x >> count); }	627	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) \| (x >> (count & 63)); }
		628
		629	#if ECB_CPP
		630
		631	inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
		632	inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
		633	inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
		634	inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
		635
		636	inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
		637	inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
		638	inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
		639	inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
		640
		641	inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
		642	inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
		643	inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
		644	inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
		645
		646	inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
		647	inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
		648	inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
		649	inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
		650
		651	inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
		652	inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
		653	inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
		654
		655	inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
		656	inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
		657	inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
		658	inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
		659
		660	inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
		661	inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
		662	inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
		663	inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
		664
		665	#endif
555		666
556	#if ECB_GCC_VERSION(4,3) \|\| (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))	667	#if ECB_GCC_VERSION(4,3) \|\| (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
557	#if ECB_GCC_VERSION(4,8) \|\| ECB_CLANG_BUILTIN(__builtin_bswap16)	668	#if ECB_GCC_VERSION(4,8) \|\| ECB_CLANG_BUILTIN(__builtin_bswap16)
558	#define ecb_bswap16(x) __builtin_bswap16 (x)	669	#define ecb_bswap16(x) __builtin_bswap16 (x)
559	#else	670	#else
…		…
598	#endif	709	#endif
599		710
600	/* try to tell the compiler that some condition is definitely true */	711	/* try to tell the compiler that some condition is definitely true */
601	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0	712	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
602		713
603	ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);	714	ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
604	ecb_inline ecb_const unsigned char	715	ecb_inline ecb_const uint32_t
605	ecb_byteorder_helper (void)	716	ecb_byteorder_helper (void)
606	{	717	{
607	/* the union code still generates code under pressure in gcc, */	718	/* the union code still generates code under pressure in gcc, */
608	/* but less than using pointers, and always seems to */	719	/* but less than using pointers, and always seems to */
609	/* successfully return a constant. */	720	/* successfully return a constant. */
610	/* the reason why we have this horrible preprocessor mess */	721	/* the reason why we have this horrible preprocessor mess */
611	/* is to avoid it in all cases, at least on common architectures */	722	/* is to avoid it in all cases, at least on common architectures */
612	/* or when using a recent enough gcc version (>= 4.6) */	723	/* or when using a recent enough gcc version (>= 4.6) */
613	#if ((__i386 \|\| __i386__) && !__VOS__) \|\| _M_IX86 \|\| ECB_GCC_AMD64 \|\| ECB_MSVC_AMD64
614	return 0x44;
615	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__	724	#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
		725	\|\| ((__i386 \|\| __i386__ \|\| _M_IX86 \|\| ECB_GCC_AMD64 \|\| ECB_MSVC_AMD64) && !__VOS__)
		726	#define ECB_LITTLE_ENDIAN 1
616	return 0x44;	727	return 0x44332211;
617	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__	728	#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
		729	\|\| ((__AARCH64EB__ \|\| __MIPSEB__ \|\| __ARMEB__) && !__VOS__)
		730	#define ECB_BIG_ENDIAN 1
618	return 0x11;	731	return 0x11223344;
619	#else	732	#else
620	union	733	union
621	{	734	{
		735	uint8_t c[4];
622	uint32_t i;	736	uint32_t u;
623	uint8_t c;
624	} u = { 0x11223344 };	737	} u = { 0x11, 0x22, 0x33, 0x44 };
625	return u.c;	738	return u.u;
626	#endif	739	#endif
627	}	740	}
628		741
629	ecb_inline ecb_const ecb_bool ecb_big_endian (void);	742	ecb_inline ecb_const ecb_bool ecb_big_endian (void);
630	ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }	743	ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
631	ecb_inline ecb_const ecb_bool ecb_little_endian (void);	744	ecb_inline ecb_const ecb_bool ecb_little_endian (void);
632	ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }	745	ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
		746
		747	/*****************************************************************************/
		748	/* unaligned load/store */
		749
		750	ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		751	ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		752	ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		753
		754	ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		755	ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		756	ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		757
		758	ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		759	ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		760	ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		761
		762	ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
		763	ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
		764	ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
		765
		766	ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
		767	ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
		768	ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
		769
		770	ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		771	ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		772	ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		773
		774	ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		775	ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		776	ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		777
		778	ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
		779	ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
		780	ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
		781
		782	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)); }
		783	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)); }
		784	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)); }
		785
		786	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)); }
		787	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)); }
		788	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)); }
		789
		790	#if ECB_CPP
		791
		792	inline uint8_t ecb_bswap (uint8_t v) { return v; }
		793	inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
		794	inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
		795	inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
		796
		797	template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		798	template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		799	template<typename T> inline T ecb_peek (const void ptr) { return (const T *)ptr; }
		800	template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
		801	template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
		802	template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
		803	template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
		804	template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
		805
		806	template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		807	template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		808	template<typename T> inline void ecb_poke (void ptr, T v) { (T *)ptr = v; }
		809	template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
		810	template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
		811	template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
		812	template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
		813	template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
		814
		815	#endif
		816
		817	/*****************************************************************************/
		818	/* pointer/integer hashing */
		819
		820	/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
		821	ecb_function_ uint32_t ecb_mix32 (uint32_t v);
		822	ecb_function_ uint32_t ecb_mix32 (uint32_t v)
		823	{
		824	v ^= v >> 16; v *= 0x7feb352dU;
		825	v ^= v >> 15; v *= 0x846ca68bU;
		826	v ^= v >> 16;
		827	return v;
		828	}
		829
		830	ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
		831	ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
		832	{
		833	v ^= v >> 16 ; v *= 0x43021123U;
		834	v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
		835	v ^= v >> 16 ;
		836	return v;
		837	}
		838
		839	/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
		840	ecb_function_ uint64_t ecb_mix64 (uint64_t v);
		841	ecb_function_ uint64_t ecb_mix64 (uint64_t v)
		842	{
		843	v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
		844	v ^= v >> 27; v *= 0x94d049bb133111ebU;
		845	v ^= v >> 31;
		846	return v;
		847	}
		848
		849	ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
		850	ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
		851	{
		852	v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
		853	v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
		854	v ^= v >> 30 ^ v >> 60;
		855	return v;
		856	}
		857
		858	ecb_function_ uintptr_t ecb_ptrmix (void *p);
		859	ecb_function_ uintptr_t ecb_ptrmix (void *p)
		860	{
		861	#if ECB_PTRSIZE <= 4
		862	return ecb_mix32 ((uint32_t)p);
		863	#else
		864	return ecb_mix64 ((uint64_t)p);
		865	#endif
		866	}
		867
		868	ecb_function_ void *ecb_ptrunmix (uintptr_t v);
		869	ecb_function_ void *ecb_ptrunmix (uintptr_t v)
		870	{
		871	#if ECB_PTRSIZE <= 4
		872	return (void *)ecb_unmix32 (v);
		873	#else
		874	return (void *)ecb_unmix64 (v);
		875	#endif
		876	}
		877
		878	#if ECB_CPP
		879
		880	template<typename T>
		881	inline uintptr_t ecb_ptrmix (T *p)
		882	{
		883	return ecb_ptrmix (static_cast<void *>(p));
		884	}
		885
		886	template<typename T>
		887	inline T *ecb_ptrunmix (uintptr_t v)
		888	{
		889	return static_cast<T *>(ecb_ptrunmix (v));
		890	}
		891
		892	#endif
		893
		894	/*****************************************************************************/
		895	/* gray code */
		896
		897	ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
		898	ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
		899	ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
		900	ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
		901
		902	ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
		903	{
		904	g ^= g >> 1;
		905	g ^= g >> 2;
		906	g ^= g >> 4;
		907
		908	return g;
		909	}
		910
		911	ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
		912	{
		913	g ^= g >> 1;
		914	g ^= g >> 2;
		915	g ^= g >> 4;
		916	g ^= g >> 8;
		917
		918	return g;
		919	}
		920
		921	ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
		922	{
		923	g ^= g >> 1;
		924	g ^= g >> 2;
		925	g ^= g >> 4;
		926	g ^= g >> 8;
		927	g ^= g >> 16;
		928
		929	return g;
		930	}
		931
		932	ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
		933	{
		934	g ^= g >> 1;
		935	g ^= g >> 2;
		936	g ^= g >> 4;
		937	g ^= g >> 8;
		938	g ^= g >> 16;
		939	g ^= g >> 32;
		940
		941	return g;
		942	}
		943
		944	#if ECB_CPP
		945
		946	ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
		947	ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
		948	ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
		949	ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
		950
		951	ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
		952	ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
		953	ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
		954	ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
		955
		956	#endif
		957
		958	/*****************************************************************************/
		959	/* 2d hilbert curves */
		960
		961	/* algorithm from the book Hacker's Delight, modified to not */
		962	/* run into undefined behaviour for n==16 */
		963	static uint32_t
		964	ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
		965	{
		966	uint32_t comp, swap, cs, t, sr;
		967
		968	/* pad s on the left (unused) bits with 01 (no change groups) */
		969	s \|= 0x55555555U << n << n;
		970	/* "s shift right" */
		971	sr = (s >> 1) & 0x55555555U;
		972	/* compute complement and swap info in two-bit groups */
		973	cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
		974
		975	/* parallel prefix xor op to propagate both complement
		976	* and swap info together from left to right (there is
		977	* no step "cs ^= cs >> 1", so in effect it computes
		978	* two independent parallel prefix operations on two
		979	* interleaved sets of sixteen bits).
		980	*/
		981	cs ^= cs >> 2;
		982	cs ^= cs >> 4;
		983	cs ^= cs >> 8;
		984	cs ^= cs >> 16;
		985
		986	/* separate swap and complement bits */
		987	swap = cs & 0x55555555U;
		988	comp = (cs >> 1) & 0x55555555U;
		989
		990	/* calculate coordinates in odd and even bit positions */
		991	t = (s & swap) ^ comp;
		992	s = s ^ sr ^ t ^ (t << 1);
		993
		994	/* unpad/clear out any junk on the left */
		995	s = s & ((1 << n << n) - 1);
		996
		997	/* Now "unshuffle" to separate the x and y bits. */
		998	t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
		999	t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
		1000	t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
		1001	t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
		1002
		1003	/* now s contains two 16-bit coordinates */
		1004	return s;
		1005	}
		1006
		1007	/* 64 bit, a straightforward extension to the 32 bit case */
		1008	static uint64_t
		1009	ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
		1010	{
		1011	uint64_t comp, swap, cs, t, sr;
		1012
		1013	/* pad s on the left (unused) bits with 01 (no change groups) */
		1014	s \|= 0x5555555555555555U << n << n;
		1015	/* "s shift right" */
		1016	sr = (s >> 1) & 0x5555555555555555U;
		1017	/* compute complement and swap info in two-bit groups */
		1018	cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
		1019
		1020	/* parallel prefix xor op to propagate both complement
		1021	* and swap info together from left to right (there is
		1022	* no step "cs ^= cs >> 1", so in effect it computes
		1023	* two independent parallel prefix operations on two
		1024	* interleaved sets of thirty-two bits).
		1025	*/
		1026	cs ^= cs >> 2;
		1027	cs ^= cs >> 4;
		1028	cs ^= cs >> 8;
		1029	cs ^= cs >> 16;
		1030	cs ^= cs >> 32;
		1031
		1032	/* separate swap and complement bits */
		1033	swap = cs & 0x5555555555555555U;
		1034	comp = (cs >> 1) & 0x5555555555555555U;
		1035
		1036	/* calculate coordinates in odd and even bit positions */
		1037	t = (s & swap) ^ comp;
		1038	s = s ^ sr ^ t ^ (t << 1);
		1039
		1040	/* unpad/clear out any junk on the left */
		1041	s = s & ((1 << n << n) - 1);
		1042
		1043	/* Now "unshuffle" to separate the x and y bits. */
		1044	t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
		1045	t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
		1046	t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
		1047	t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
		1048	t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
		1049
		1050	/* now s contains two 32-bit coordinates */
		1051	return s;
		1052	}
		1053
		1054	/* algorithm from the book Hacker's Delight, but a similar algorithm*/
		1055	/* is given in https://doi.org/10.1002/spe.4380160103 */
		1056	/* this has been slightly improved over the original version */
		1057	ecb_function_ uint32_t
		1058	ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
		1059	{
		1060	uint32_t row;
		1061	uint32_t state = 0;
		1062	uint32_t s = 0;
		1063
		1064	do
		1065	{
		1066	--n;
		1067
		1068	row = 4 * state
		1069	\| (2 & (xy >> n >> 15))
		1070	\| (1 & (xy >> n ));
		1071
		1072	/* these funky constants are lookup tables for two-bit values */
		1073	s = (s << 2) \| (0x361e9cb4U >> 2 * row) & 3;
		1074	state = (0x8fe65831U >> 2 * row) & 3;
		1075	}
		1076	while (n > 0);
		1077
		1078	return s;
		1079	}
		1080
		1081	/* 64 bit, essentially the same as 32 bit */
		1082	ecb_function_ uint64_t
		1083	ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
		1084	{
		1085	uint32_t row;
		1086	uint32_t state = 0;
		1087	uint64_t s = 0;
		1088
		1089	do
		1090	{
		1091	--n;
		1092
		1093	row = 4 * state
		1094	\| (2 & (xy >> n >> 31))
		1095	\| (1 & (xy >> n ));
		1096
		1097	/* these funky constants are lookup tables for two-bit values */
		1098	s = (s << 2) \| (0x361e9cb4U >> 2 * row) & 3;
		1099	state = (0x8fe65831U >> 2 * row) & 3;
		1100	}
		1101	while (n > 0);
		1102
		1103	return s;
		1104	}
		1105
		1106	/*****************************************************************************/
		1107	/* division */
633		1108
634	#if ECB_GCC_VERSION(3,0) \|\| ECB_C99	1109	#if ECB_GCC_VERSION(3,0) \|\| ECB_C99
		1110	/* C99 tightened the definition of %, so we can use a more efficient version */
635	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))	1111	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
636	#else	1112	#else
637	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))	1113	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
638	#endif	1114	#endif
639		1115
…		…
651	#else	1127	#else
652	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))	1128	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
653	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))	1129	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
654	#endif	1130	#endif
655		1131
		1132	/*****************************************************************************/
		1133	/* array length */
		1134
656	#if ecb_cplusplus_does_not_suck	1135	#if ecb_cplusplus_does_not_suck
657	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */	1136	/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
658	template<typename T, int N>	1137	template<typename T, int N>
659	static inline int ecb_array_length (const T (&arr)[N])	1138	static inline int ecb_array_length (const T (&arr)[N])
660	{	1139	{
661	return N;	1140	return N;
662	}	1141	}
663	#else	1142	#else
664	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))	1143	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
665	#endif	1144	#endif
		1145
		1146	/*****************************************************************************/
		1147	/* IEEE 754-2008 half float conversions */
		1148
		1149	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
		1150	ecb_function_ ecb_const uint32_t
		1151	ecb_binary16_to_binary32 (uint32_t x)
		1152	{
		1153	unsigned int s = (x & 0x8000) << (31 - 15);
		1154	int e = (x >> 10) & 0x001f;
		1155	unsigned int m = x & 0x03ff;
		1156
		1157	if (ecb_expect_false (e == 31))
		1158	/* infinity or NaN */
		1159	e = 255 - (127 - 15);
		1160	else if (ecb_expect_false (!e))
		1161	{
		1162	if (ecb_expect_true (!m))
		1163	/* zero, handled by code below by forcing e to 0 */
		1164	e = 0 - (127 - 15);
		1165	else
		1166	{
		1167	/* subnormal, renormalise */
		1168	unsigned int s = 10 - ecb_ld32 (m);
		1169
		1170	m = (m << s) & 0x3ff; /* mask implicit bit */
		1171	e -= s - 1;
		1172	}
		1173	}
		1174
		1175	/* e and m now are normalised, or zero, (or inf or nan) */
		1176	e += 127 - 15;
		1177
		1178	return s \| (e << 23) \| (m << (23 - 10));
		1179	}
		1180
		1181	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
		1182	ecb_function_ ecb_const uint16_t
		1183	ecb_binary32_to_binary16 (uint32_t x)
		1184	{
		1185	unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
		1186	int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
		1187	unsigned int m = x & 0x007fffff;
		1188
		1189	x &= 0x7fffffff;
		1190
		1191	/* if it's within range of binary16 normals, use fast path */
		1192	if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
		1193	{
		1194	/* mantissa round-to-even */
		1195	m += 0x00000fff + ((m >> (23 - 10)) & 1);
		1196
		1197	/* handle overflow */
		1198	if (ecb_expect_false (m >= 0x00800000))
		1199	{
		1200	m >>= 1;
		1201	e += 1;
		1202	}
		1203
		1204	return s \| (e << 10) \| (m >> (23 - 10));
		1205	}
		1206
		1207	/* handle large numbers and infinity */
		1208	if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
		1209	return s \| 0x7c00;
		1210
		1211	/* handle zero, subnormals and small numbers */
		1212	if (ecb_expect_true (x < 0x38800000))
		1213	{
		1214	/* zero */
		1215	if (ecb_expect_true (!x))
		1216	return s;
		1217
		1218	/* handle subnormals */
		1219
		1220	/* too small, will be zero */
		1221	if (e < (14 - 24)) /* might not be sharp, but is good enough */
		1222	return s;
		1223
		1224	m \|= 0x00800000; /* make implicit bit explicit */
		1225
		1226	/* very tricky - we need to round to the nearest e (+10) bit value */
		1227	{
		1228	unsigned int bits = 14 - e;
		1229	unsigned int half = (1 << (bits - 1)) - 1;
		1230	unsigned int even = (m >> bits) & 1;
		1231
		1232	/* if this overflows, we will end up with a normalised number */
		1233	m = (m + half + even) >> bits;
		1234	}
		1235
		1236	return s \| m;
		1237	}
		1238
		1239	/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
		1240	m >>= 13;
		1241
		1242	return s \| 0x7c00 \| m \| !m;
		1243	}
		1244
		1245	/*******************************************************************************/
		1246	/* fast integer to ascii */
		1247
		1248	/*
		1249	* This code is pretty complicated because it is general. The idea behind it,
		1250	* however, is pretty simple: first, the number is multiplied with a scaling
		1251	* factor (2bits / 10(digits-1)) to convert the integer into a fixed-point
		1252	* number with the first digit in the upper bits.
		1253	* Then this digit is converted to text and masked out. The resulting number
		1254	* is then multiplied by 10, by multiplying the fixed point representation
		1255	* by 5 and shifting the (binary) decimal point one to the right, so a 4.28
		1256	* format becomes 5.27, 6.26 and so on.
		1257	* The rest involves only advancing the pointer if we already generated a
		1258	* non-zero digit, so leading zeroes are overwritten.
		1259	*/
		1260
		1261	/* simply return a mask with "bits" bits set */
		1262	#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
		1263
		1264	/* oputput a single digit. maskvalue is 10*digitidx /
		1265	#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
		1266	if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
		1267	{ \
		1268	char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
		1269	ptr = digit + '0'; / output it */ \
		1270	nz = (digitmask == maskvalue) \|\| nz \|\| digit; /* first term == always output last digit */ \
		1271	ptr += nz; /* output digit only if non-zero digit seen */ \
		1272	x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* 10, but shift decimal point right / \
		1273	}
		1274
		1275	/* convert integer to fixed point format and multiply out digits, highest first */
		1276	/* requires magic constants: max. digits and number of bits after the decimal point */
		1277	#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
		1278	ecb_inline char ecb_i2a_ ## suffix (char ptr, uint32_t u) \
		1279	{ \
		1280	char nz = lz; /* non-zero digit seen? */ \
		1281	/* convert to x.bits fixed-point */ \
		1282	type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
		1283	/* output up to 10 digits */ \
		1284	ecb_i2a_digit (type,bits,digitmask, 1, 0); \
		1285	ecb_i2a_digit (type,bits,digitmask, 10, 1); \
		1286	ecb_i2a_digit (type,bits,digitmask, 100, 2); \
		1287	ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
		1288	ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
		1289	ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
		1290	ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
		1291	ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
		1292	ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
		1293	ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
		1294	return ptr; \
		1295	}
		1296
		1297	/* predefined versions of the above, for various digits */
		1298	/* ecb_i2a_xN = almost N digits, limit defined by macro */
		1299	/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
		1300	/* ecb_i2a_0N = exactly N digits, including leading zeroes */
		1301
		1302	/* non-leading-zero versions, limited range */
		1303	#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
		1304	#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
		1305	ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
		1306	ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
		1307
		1308	/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
		1309	ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
		1310	ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
		1311	ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
		1312	ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
		1313	ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
		1314	ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
		1315	ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
		1316	ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
		1317
		1318	/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
		1319	ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
		1320	ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
		1321	ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
		1322	ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
		1323	ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
		1324	ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
		1325	ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
		1326	ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
		1327
		1328	#define ECB_I2A_I32_DIGITS 11
		1329	#define ECB_I2A_U32_DIGITS 10
		1330	#define ECB_I2A_I64_DIGITS 20
		1331	#define ECB_I2A_U64_DIGITS 21
		1332	#define ECB_I2A_MAX_DIGITS 21
		1333
		1334	ecb_inline char *
		1335	ecb_i2a_u32 (char *ptr, uint32_t u)
		1336	{
		1337	#if ECB_64BIT_NATIVE
		1338	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1339	ptr = ecb_i2a_x10 (ptr, u);
		1340	else /* x10 almost, but not fully, covers 32 bit */
		1341	{
		1342	uint32_t u1 = u % 1000000000;
		1343	uint32_t u2 = u / 1000000000;
		1344
		1345	*ptr++ = u2 + '0';
		1346	ptr = ecb_i2a_09 (ptr, u1);
		1347	}
		1348	#else
		1349	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1350	ecb_i2a_x5 (ptr, u);
		1351	else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
		1352	{
		1353	uint32_t u1 = u % 10000;
		1354	uint32_t u2 = u / 10000;
		1355
		1356	ptr = ecb_i2a_x5 (ptr, u2);
		1357	ptr = ecb_i2a_04 (ptr, u1);
		1358	}
		1359	else
		1360	{
		1361	uint32_t u1 = u % 10000;
		1362	uint32_t ua = u / 10000;
		1363	uint32_t u2 = ua % 10000;
		1364	uint32_t u3 = ua / 10000;
		1365
		1366	ptr = ecb_i2a_2 (ptr, u3);
		1367	ptr = ecb_i2a_04 (ptr, u2);
		1368	ptr = ecb_i2a_04 (ptr, u1);
		1369	}
		1370	#endif
		1371
		1372	return ptr;
		1373	}
		1374
		1375	ecb_inline char *
		1376	ecb_i2a_i32 (char *ptr, int32_t v)
		1377	{
		1378	*ptr = '-'; ptr += v < 0;
		1379	uint32_t u = v < 0 ? -(uint32_t)v : v;
		1380
		1381	#if ECB_64BIT_NATIVE
		1382	ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
		1383	#else
		1384	ptr = ecb_i2a_u32 (ptr, u);
		1385	#endif
		1386
		1387	return ptr;
		1388	}
		1389
		1390	ecb_inline char *
		1391	ecb_i2a_u64 (char *ptr, uint64_t u)
		1392	{
		1393	#if ECB_64BIT_NATIVE
		1394	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1395	ptr = ecb_i2a_x10 (ptr, u);
		1396	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1397	{
		1398	uint64_t u1 = u % 1000000000;
		1399	uint64_t u2 = u / 1000000000;
		1400
		1401	ptr = ecb_i2a_x10 (ptr, u2);
		1402	ptr = ecb_i2a_09 (ptr, u1);
		1403	}
		1404	else
		1405	{
		1406	uint64_t u1 = u % 1000000000;
		1407	uint64_t ua = u / 1000000000;
		1408	uint64_t u2 = ua % 1000000000;
		1409	uint64_t u3 = ua / 1000000000;
		1410
		1411	ptr = ecb_i2a_2 (ptr, u3);
		1412	ptr = ecb_i2a_09 (ptr, u2);
		1413	ptr = ecb_i2a_09 (ptr, u1);
		1414	}
		1415	#else
		1416	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1417	ptr = ecb_i2a_x5 (ptr, u);
		1418	else
		1419	{
		1420	uint64_t u1 = u % 10000;
		1421	uint64_t u2 = u / 10000;
		1422
		1423	ptr = ecb_i2a_u64 (ptr, u2);
		1424	ptr = ecb_i2a_04 (ptr, u1);
		1425	}
		1426	#endif
		1427
		1428	return ptr;
		1429	}
		1430
		1431	ecb_inline char *
		1432	ecb_i2a_i64 (char *ptr, int64_t v)
		1433	{
		1434	*ptr = '-'; ptr += v < 0;
		1435	uint64_t u = v < 0 ? -(uint64_t)v : v;
		1436
		1437	#if ECB_64BIT_NATIVE
		1438	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1439	ptr = ecb_i2a_x10 (ptr, u);
		1440	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1441	{
		1442	uint64_t u1 = u % 1000000000;
		1443	uint64_t u2 = u / 1000000000;
		1444
		1445	ptr = ecb_i2a_x10 (ptr, u2);
		1446	ptr = ecb_i2a_09 (ptr, u1);
		1447	}
		1448	else
		1449	{
		1450	uint64_t u1 = u % 1000000000;
		1451	uint64_t ua = u / 1000000000;
		1452	uint64_t u2 = ua % 1000000000;
		1453	uint64_t u3 = ua / 1000000000;
		1454
		1455	/* 2*31 is 19 digits, so the top is exactly one digit /
		1456	*ptr++ = u3 + '0';
		1457	ptr = ecb_i2a_09 (ptr, u2);
		1458	ptr = ecb_i2a_09 (ptr, u1);
		1459	}
		1460	#else
		1461	ptr = ecb_i2a_u64 (ptr, u);
		1462	#endif
		1463
		1464	return ptr;
		1465	}
666		1466
667	/*******************************************************************************/	1467	/*******************************************************************************/
668	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */	1468	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
669		1469
670	/* basically, everything uses "ieee pure-endian" floating point numbers */	1470	/* basically, everything uses "ieee pure-endian" floating point numbers */
…		…
683	\|\| defined __sh__ \	1483	\|\| defined __sh__ \
684	\|\| defined _M_IX86 \|\| defined ECB_MSVC_AMD64 \|\| defined _M_IA64 \	1484	\|\| defined _M_IX86 \|\| defined ECB_MSVC_AMD64 \|\| defined _M_IA64 \
685	\|\| (defined __arm__ && (defined __ARM_EABI__ \|\| defined __EABI__ \|\| defined __VFP_FP__ \|\| defined _WIN32_WCE \|\| defined __ANDROID__)) \	1485	\|\| (defined __arm__ && (defined __ARM_EABI__ \|\| defined __EABI__ \|\| defined __VFP_FP__ \|\| defined _WIN32_WCE \|\| defined __ANDROID__)) \
686	\|\| defined __aarch64__	1486	\|\| defined __aarch64__
687	#define ECB_STDFP 1	1487	#define ECB_STDFP 1
688	#include <string.h> /* for memcpy */
689	#else	1488	#else
690	#define ECB_STDFP 0	1489	#define ECB_STDFP 0
691	#endif	1490	#endif
692		1491
693	#ifndef ECB_NO_LIBM	1492	#ifndef ECB_NO_LIBM
…		…
713	#else	1512	#else
714	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))	1513	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
715	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))	1514	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
716	#endif	1515	#endif
717		1516
718	/* converts an ieee half/binary16 to a float */
719	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
720	ecb_function_ ecb_const float
721	ecb_binary16_to_float (uint16_t x)
722	{
723	int e = (x >> 10) & 0x1f;
724	int m = x & 0x3ff;
725	float r;
726
727	if (!e ) r = ecb_ldexpf (m , -24);
728	else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25);
729	else if (m ) r = ECB_NAN;
730	else r = ECB_INFINITY;
731
732	return x & 0x8000 ? -r : r;
733	}
734
735	/* convert a float to ieee single/binary32 */	1517	/* convert a float to ieee single/binary32 */
736	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);	1518	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
737	ecb_function_ ecb_const uint32_t	1519	ecb_function_ ecb_const uint32_t
738	ecb_float_to_binary32 (float x)	1520	ecb_float_to_binary32 (float x)
739	{	1521	{
…		…
870	#endif	1652	#endif
871		1653
872	return r;	1654	return r;
873	}	1655	}
874		1656
875	#endif	1657	/* convert a float to ieee half/binary16 */
		1658	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
		1659	ecb_function_ ecb_const uint16_t
		1660	ecb_float_to_binary16 (float x)
		1661	{
		1662	return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
		1663	}
876		1664
877	#endif	1665	/* convert an ieee half/binary16 to float */
		1666	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
		1667	ecb_function_ ecb_const float
		1668	ecb_binary16_to_float (uint16_t x)
		1669	{
		1670	return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
		1671	}
878		1672
		1673	#endif
		1674
		1675	#endif
		1676

Diff Legend

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

Comparing libecb/ecb.h (file contents): Revision 1.166 by root, Sun Aug 9 00:10:21 2015 UTC vs. Revision 1.205 by root, Fri Mar 25 14:21:14 2022 UTC

Diff Legend

Comparing libecb/ecb.h (file contents):
Revision 1.166 by root, Sun Aug 9 00:10:21 2015 UTC vs.
Revision 1.205 by root, Fri Mar 25 14:21:14 2022 UTC