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

Comparing libecb/ecb.h (file contents):
Revision 1.165 by sf-exg, Fri May 1 12:02:18 2015 UTC vs.
Revision 1.192 by root, Mon Jun 21 23:59:58 2021 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 0x00010009
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__) && !__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")
182	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore \| #LoadLoad" : : : "memory")	222	#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore \| #LoadLoad" : : : "memory")
183	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore \| #StoreStore")	223	#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore \| #StoreStore")
184	#elif defined __s390__ \|\| defined __s390x__	224	#elif defined __s390__ \|\| defined __s390x__
185	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")	225	#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "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;
429		477
430	x &= ~x + 1; /* this isolates the lowest bit */	478	x &= ~x + 1; /* this isolates the lowest bit */
431		479
432	#if ECB_branchless_on_i386	480	#if ECB_branchless_on_i386
…		…
442	if (x & 0xff00ff00) r += 8;	490	if (x & 0xff00ff00) r += 8;
443	if (x & 0xffff0000) r += 16;	491	if (x & 0xffff0000) r += 16;
444	#endif	492	#endif
445		493
446	return r;	494	return r;
		495	#endif
447	}	496	}
448		497
449	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);	498	ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
450	ecb_function_ ecb_const int	499	ecb_function_ ecb_const int
451	ecb_ctz64 (uint64_t x)	500	ecb_ctz64 (uint64_t x)
452	{	501	{
		502	#if 1400 <= _MSC_VER && (_M_X64 \|\| _M_IA64 \|\| _M_ARM)
		503	unsigned long r;
		504	_BitScanForward64 (&r, x);
		505	return (int)r;
		506	#else
453	int shift = x & 0xffffffffU ? 0 : 32;	507	int shift = x & 0xffffffff ? 0 : 32;
454	return ecb_ctz32 (x >> shift) + shift;	508	return ecb_ctz32 (x >> shift) + shift;
		509	#endif
455	}	510	}
456		511
457	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);	512	ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
458	ecb_function_ ecb_const int	513	ecb_function_ ecb_const int
459	ecb_popcount32 (uint32_t x)	514	ecb_popcount32 (uint32_t x)
…		…
467	}	522	}
468		523
469	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);	524	ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
470	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)	525	ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
471	{	526	{
		527	#if 1400 <= _MSC_VER && (_M_IX86 \|\| _M_X64 \|\| _M_IA64 \|\| _M_ARM)
		528	unsigned long r;
		529	_BitScanReverse (&r, x);
		530	return (int)r;
		531	#else
472	int r = 0;	532	int r = 0;
473		533
474	if (x >> 16) { x >>= 16; r += 16; }	534	if (x >> 16) { x >>= 16; r += 16; }
475	if (x >> 8) { x >>= 8; r += 8; }	535	if (x >> 8) { x >>= 8; r += 8; }
476	if (x >> 4) { x >>= 4; r += 4; }	536	if (x >> 4) { x >>= 4; r += 4; }
477	if (x >> 2) { x >>= 2; r += 2; }	537	if (x >> 2) { x >>= 2; r += 2; }
478	if (x >> 1) { r += 1; }	538	if (x >> 1) { r += 1; }
479		539
480	return r;	540	return r;
		541	#endif
481	}	542	}
482		543
483	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);	544	ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
484	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)	545	ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
485	{	546	{
		547	#if 1400 <= _MSC_VER && (_M_X64 \|\| _M_IA64 \|\| _M_ARM)
		548	unsigned long r;
		549	_BitScanReverse64 (&r, x);
		550	return (int)r;
		551	#else
486	int r = 0;	552	int r = 0;
487		553
488	if (x >> 32) { x >>= 32; r += 32; }	554	if (x >> 32) { x >>= 32; r += 32; }
489		555
490	return r + ecb_ld32 (x);	556	return r + ecb_ld32 (x);
		557	#endif
491	}	558	}
492	#endif	559	#endif
493		560
494	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);	561	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)); }	562	ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
…		…
551	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) \| (x << count); }	618	ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) \| (x << count); }
552	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) \| (x >> count); }	619	ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) \| (x >> count); }
553	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) \| (x << count); }	620	ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) \| (x << count); }
554	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) \| (x >> count); }	621	ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) \| (x >> count); }
555		622
		623	#if ECB_CPP
		624
		625	inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
		626	inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
		627	inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
		628	inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
		629
		630	inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
		631	inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
		632	inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
		633	inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
		634
		635	inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
		636	inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
		637	inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
		638	inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
		639
		640	inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
		641	inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
		642	inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
		643	inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
		644
		645	inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
		646	inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
		647	inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
		648
		649	inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
		650	inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
		651	inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
		652	inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
		653
		654	inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
		655	inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
		656	inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
		657	inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
		658
		659	#endif
		660
556	#if ECB_GCC_VERSION(4,3) \|\| (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))	661	#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)	662	#if ECB_GCC_VERSION(4,8) \|\| ECB_CLANG_BUILTIN(__builtin_bswap16)
558	#define ecb_bswap16(x) __builtin_bswap16 (x)	663	#define ecb_bswap16(x) __builtin_bswap16 (x)
559	#else	664	#else
560	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)	665	#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
…		…
598	#endif	703	#endif
599		704
600	/* try to tell the compiler that some condition is definitely true */	705	/* try to tell the compiler that some condition is definitely true */
601	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0	706	#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
602		707
603	ecb_inline ecb_const unsigned char ecb_byteorder_helper (void);	708	ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
604	ecb_inline ecb_const unsigned char	709	ecb_inline ecb_const uint32_t
605	ecb_byteorder_helper (void)	710	ecb_byteorder_helper (void)
606	{	711	{
607	/* the union code still generates code under pressure in gcc, */	712	/* the union code still generates code under pressure in gcc, */
608	/* but less than using pointers, and always seems to */	713	/* but less than using pointers, and always seems to */
609	/* successfully return a constant. */	714	/* successfully return a constant. */
610	/* the reason why we have this horrible preprocessor mess */	715	/* the reason why we have this horrible preprocessor mess */
611	/* is to avoid it in all cases, at least on common architectures */	716	/* is to avoid it in all cases, at least on common architectures */
612	/* or when using a recent enough gcc version (>= 4.6) */	717	/* 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__	718	#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
		719	\|\| ((__i386 \|\| __i386__ \|\| _M_IX86 \|\| ECB_GCC_AMD64 \|\| ECB_MSVC_AMD64) && !__VOS__)
		720	#define ECB_LITTLE_ENDIAN 1
616	return 0x44;	721	return 0x44332211;
617	#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__	722	#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
		723	\|\| ((__AARCH64EB__ \|\| __MIPSEB__ \|\| __ARMEB__) && !__VOS__)
		724	#define ECB_BIG_ENDIAN 1
618	return 0x11;	725	return 0x11223344;
619	#else	726	#else
620	union	727	union
621	{	728	{
		729	uint8_t c[4];
622	uint32_t i;	730	uint32_t u;
623	uint8_t c;
624	} u = { 0x11223344 };	731	} u = { 0x11, 0x22, 0x33, 0x44 };
625	return u.c;	732	return u.u;
626	#endif	733	#endif
627	}	734	}
628		735
629	ecb_inline ecb_const ecb_bool ecb_big_endian (void);	736	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; }	737	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);	738	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; }	739	ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
		740
		741	/*****************************************************************************/
		742	/* unaligned load/store */
		743
		744	ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		745	ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		746	ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		747
		748	ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		749	ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		750	ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		751
		752	ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		753	ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		754	ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
		755
		756	ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
		757	ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
		758	ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
		759
		760	ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
		761	ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
		762	ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
		763
		764	ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
		765	ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
		766	ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
		767
		768	ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
		769	ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
		770	ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
		771
		772	ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
		773	ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
		774	ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
		775
		776	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)); }
		777	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)); }
		778	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)); }
		779
		780	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)); }
		781	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)); }
		782	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)); }
		783
		784	#if ECB_CPP
		785
		786	inline uint8_t ecb_bswap (uint8_t v) { return v; }
		787	inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
		788	inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
		789	inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
		790
		791	template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		792	template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		793	template<typename T> inline T ecb_peek (const void ptr) { return (const T *)ptr; }
		794	template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
		795	template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
		796	template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
		797	template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
		798	template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
		799
		800	template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
		801	template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
		802	template<typename T> inline void ecb_poke (void ptr, T v) { (T *)ptr = v; }
		803	template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
		804	template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
		805	template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
		806	template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
		807	template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
		808
		809	#endif
		810
		811	/*****************************************************************************/
		812	/* division */
633		813
634	#if ECB_GCC_VERSION(3,0) \|\| ECB_C99	814	#if ECB_GCC_VERSION(3,0) \|\| ECB_C99
		815	/* 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))	816	#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
636	#else	817	#else
637	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))	818	#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
638	#endif	819	#endif
639		820
…		…
650	}	831	}
651	#else	832	#else
652	#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))	833	#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))	834	#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
654	#endif	835	#endif
		836
		837	/*****************************************************************************/
		838	/* array length */
655		839
656	#if ecb_cplusplus_does_not_suck	840	#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) */	841	/* 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>	842	template<typename T, int N>
659	static inline int ecb_array_length (const T (&arr)[N])	843	static inline int ecb_array_length (const T (&arr)[N])
…		…
661	return N;	845	return N;
662	}	846	}
663	#else	847	#else
664	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))	848	#define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
665	#endif	849	#endif
		850
		851	/*****************************************************************************/
		852	/* IEEE 754-2008 half float conversions */
		853
		854	ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
		855	ecb_function_ ecb_const uint32_t
		856	ecb_binary16_to_binary32 (uint32_t x)
		857	{
		858	unsigned int s = (x & 0x8000) << (31 - 15);
		859	int e = (x >> 10) & 0x001f;
		860	unsigned int m = x & 0x03ff;
		861
		862	if (ecb_expect_false (e == 31))
		863	/* infinity or NaN */
		864	e = 255 - (127 - 15);
		865	else if (ecb_expect_false (!e))
		866	{
		867	if (ecb_expect_true (!m))
		868	/* zero, handled by code below by forcing e to 0 */
		869	e = 0 - (127 - 15);
		870	else
		871	{
		872	/* subnormal, renormalise */
		873	unsigned int s = 10 - ecb_ld32 (m);
		874
		875	m = (m << s) & 0x3ff; /* mask implicit bit */
		876	e -= s - 1;
		877	}
		878	}
		879
		880	/* e and m now are normalised, or zero, (or inf or nan) */
		881	e += 127 - 15;
		882
		883	return s \| (e << 23) \| (m << (23 - 10));
		884	}
		885
		886	ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
		887	ecb_function_ ecb_const uint16_t
		888	ecb_binary32_to_binary16 (uint32_t x)
		889	{
		890	unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
		891	int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
		892	unsigned int m = x & 0x007fffff;
		893
		894	x &= 0x7fffffff;
		895
		896	/* if it's within range of binary16 normals, use fast path */
		897	if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
		898	{
		899	/* mantissa round-to-even */
		900	m += 0x00000fff + ((m >> (23 - 10)) & 1);
		901
		902	/* handle overflow */
		903	if (ecb_expect_false (m >= 0x00800000))
		904	{
		905	m >>= 1;
		906	e += 1;
		907	}
		908
		909	return s \| (e << 10) \| (m >> (23 - 10));
		910	}
		911
		912	/* handle large numbers and infinity */
		913	if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
		914	return s \| 0x7c00;
		915
		916	/* handle zero, subnormals and small numbers */
		917	if (ecb_expect_true (x < 0x38800000))
		918	{
		919	/* zero */
		920	if (ecb_expect_true (!x))
		921	return s;
		922
		923	/* handle subnormals */
		924
		925	/* too small, will be zero */
		926	if (e < (14 - 24)) /* might not be sharp, but is good enough */
		927	return s;
		928
		929	m \|= 0x00800000; /* make implicit bit explicit */
		930
		931	/* very tricky - we need to round to the nearest e (+10) bit value */
		932	{
		933	unsigned int bits = 14 - e;
		934	unsigned int half = (1 << (bits - 1)) - 1;
		935	unsigned int even = (m >> bits) & 1;
		936
		937	/* if this overflows, we will end up with a normalised number */
		938	m = (m + half + even) >> bits;
		939	}
		940
		941	return s \| m;
		942	}
		943
		944	/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
		945	m >>= 13;
		946
		947	return s \| 0x7c00 \| m \| !m;
		948	}
		949
		950	/*******************************************************************************/
		951	/* fast integer to ascii */
		952
		953	// simply return a mask with "bits" bits set
		954	#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
		955
		956	// oputput a single digit. maskvalue is 10**digitidx
		957	#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
		958	if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
		959	{ \
		960	char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
		961	ptr = digit + '0'; / output it */ \
		962	nz = (digitmask == maskvalue) \|\| nz \|\| digit; /* first term == always output last digit */ \
		963	ptr += nz; /* output digit only if non-zero digit seen */ \
		964	x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* 10, but shift decimal point right / \
		965	}
		966
		967	// convert integer to fixed point format and multiply out digits, highest first
		968	// requires magic constants: max. digits and number of bits after the decimal point
		969	#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
		970	ecb_inline char ecb_i2a_ ## suffix (char ptr, uint32_t u) \
		971	{ \
		972	char nz = lz; /* non-zero digit seen? */ \
		973	/* convert to x.bits fixed-point */ \
		974	type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
		975	/* output up to 10 digits */ \
		976	ecb_i2a_digit (type,bits,digitmask, 1, 0); \
		977	ecb_i2a_digit (type,bits,digitmask, 10, 1); \
		978	ecb_i2a_digit (type,bits,digitmask, 100, 2); \
		979	ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
		980	ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
		981	ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
		982	ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
		983	ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
		984	ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
		985	ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
		986	return ptr; \
		987	}
		988
		989	// predefined versions of the above, for various digits
		990	// ecb_i2a_xN = almost N digits, limit defined by macro
		991	// ecb_i2a_N = up to N digits, leading zeroes suppressed
		992	// ecb_i2a_0N = exactly N digits, including leading zeroes
		993
		994	// non-leading-zero versions, limited range
		995	#define ECB_I2A_MAX_X5 59074 // limit for ecb_i2a_x5
		996	#define ECB_I2A_MAX_X10 2932500665 // limit for ecb_i2a_x10
		997	ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
		998	ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
		999
		1000	// non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit
		1001	ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
		1002	ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
		1003	ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
		1004	ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
		1005	ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
		1006	ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
		1007	ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
		1008	ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
		1009
		1010	// leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit
		1011	ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
		1012	ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
		1013	ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
		1014	ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
		1015	ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
		1016	ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
		1017	ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
		1018	ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
		1019
		1020	#define ECB_I2A_I32_DIGITS 11
		1021	#define ECB_I2A_U32_DIGITS 10
		1022	#define ECB_I2A_I64_DIGITS 20
		1023	#define ECB_I2A_U32_DIGITS 21
		1024	#define ECB_I2A_DIGITS 21
		1025
		1026	ecb_inline char *
		1027	ecb_i2a_u32 (char *ptr, uint32_t u)
		1028	{
		1029	#if ECB_64BIT_NATIVE
		1030	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1031	ptr = ecb_i2a_x10 (ptr, u);
		1032	else // x10 almost, but not fully, covers 32 bit
		1033	{
		1034	uint32_t u1 = u % 1000000000;
		1035	uint32_t u2 = u / 1000000000;
		1036
		1037	*ptr++ = u2 + '0';
		1038	ptr = ecb_i2a_09 (ptr, u1);
		1039	}
		1040	#else
		1041	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1042	ecb_i2a_x5 (ptr, u);
		1043	else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
		1044	{
		1045	uint32_t u1 = u % 10000;
		1046	uint32_t u2 = u / 10000;
		1047
		1048	ptr = ecb_i2a_x5 (ptr, u2);
		1049	ptr = ecb_i2a_04 (ptr, u1);
		1050	}
		1051	else
		1052	{
		1053	uint32_t u1 = u % 10000;
		1054	uint32_t ua = u / 10000;
		1055	uint32_t u2 = ua % 10000;
		1056	uint32_t u3 = ua / 10000;
		1057
		1058	ptr = ecb_i2a_2 (ptr, u3);
		1059	ptr = ecb_i2a_04 (ptr, u2);
		1060	ptr = ecb_i2a_04 (ptr, u1);
		1061	}
		1062	#endif
		1063
		1064	return ptr;
		1065	}
		1066
		1067	ecb_inline char *
		1068	ecb_i2a_i32 (char *ptr, int32_t v)
		1069	{
		1070	*ptr = '-'; ptr += v < 0;
		1071	uint32_t u = v < 0 ? -(uint32_t)v : v;
		1072
		1073	#if ECB_64BIT_NATIVE
		1074	ptr = ecb_i2a_x10 (ptr, u); // x10 fully covers 31 bit
		1075	#else
		1076	ptr = ecb_i2a_u32 (ptr, u);
		1077	#endif
		1078
		1079	return ptr;
		1080	}
		1081
		1082	ecb_inline char *
		1083	ecb_i2a_u64 (char *ptr, uint64_t u)
		1084	{
		1085	#if ECB_64BIT_NATIVE
		1086	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1087	ptr = ecb_i2a_x10 (ptr, u);
		1088	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1089	{
		1090	uint64_t u1 = u % 1000000000;
		1091	uint64_t u2 = u / 1000000000;
		1092
		1093	ptr = ecb_i2a_x10 (ptr, u2);
		1094	ptr = ecb_i2a_09 (ptr, u1);
		1095	}
		1096	else
		1097	{
		1098	uint64_t u1 = u % 1000000000;
		1099	uint64_t ua = u / 1000000000;
		1100	uint64_t u2 = ua % 1000000000;
		1101	uint64_t u3 = ua / 1000000000;
		1102
		1103	ptr = ecb_i2a_2 (ptr, u3);
		1104	ptr = ecb_i2a_09 (ptr, u2);
		1105	ptr = ecb_i2a_09 (ptr, u1);
		1106	}
		1107	#else
		1108	if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
		1109	ptr = ecb_i2a_x5 (ptr, u);
		1110	else
		1111	{
		1112	uint64_t u1 = u % 10000;
		1113	uint64_t u2 = u / 10000;
		1114
		1115	ptr = ecb_i2a_u64 (ptr, u2);
		1116	ptr = ecb_i2a_04 (ptr, u1);
		1117	}
		1118	#endif
		1119
		1120	return ptr;
		1121	}
		1122
		1123	ecb_inline char *
		1124	ecb_i2a_i64 (char *ptr, int64_t v)
		1125	{
		1126	*ptr = '-'; ptr += v < 0;
		1127	uint64_t u = v < 0 ? -(uint64_t)v : v;
		1128
		1129	#if ECB_64BIT_NATIVE
		1130	if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
		1131	ptr = ecb_i2a_x10 (ptr, u);
		1132	else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
		1133	{
		1134	uint64_t u1 = u % 1000000000;
		1135	uint64_t u2 = u / 1000000000;
		1136
		1137	ptr = ecb_i2a_x10 (ptr, u2);
		1138	ptr = ecb_i2a_09 (ptr, u1);
		1139	}
		1140	else
		1141	{
		1142	uint64_t u1 = u % 1000000000;
		1143	uint64_t ua = u / 1000000000;
		1144	uint64_t u2 = ua % 1000000000;
		1145	uint64_t u3 = ua / 1000000000;
		1146
		1147	// 2**31 is 19 digits, so the top is exactly one digit
		1148	*ptr++ = u3 + '0';
		1149	ptr = ecb_i2a_09 (ptr, u2);
		1150	ptr = ecb_i2a_09 (ptr, u1);
		1151	}
		1152	#else
		1153	ptr = ecb_i2a_u64 (ptr, u);
		1154	#endif
		1155
		1156	return ptr;
		1157	}
666		1158
667	/*******************************************************************************/	1159	/*******************************************************************************/
668	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */	1160	/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
669		1161
670	/* basically, everything uses "ieee pure-endian" floating point numbers */	1162	/* basically, everything uses "ieee pure-endian" floating point numbers */
…		…
683	\|\| defined __sh__ \	1175	\|\| defined __sh__ \
684	\|\| defined _M_IX86 \|\| defined ECB_MSVC_AMD64 \|\| defined _M_IA64 \	1176	\|\| 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__)) \	1177	\|\| (defined __arm__ && (defined __ARM_EABI__ \|\| defined __EABI__ \|\| defined __VFP_FP__ \|\| defined _WIN32_WCE \|\| defined __ANDROID__)) \
686	\|\| defined __aarch64__	1178	\|\| defined __aarch64__
687	#define ECB_STDFP 1	1179	#define ECB_STDFP 1
688	#include <string.h> /* for memcpy */
689	#else	1180	#else
690	#define ECB_STDFP 0	1181	#define ECB_STDFP 0
691	#endif	1182	#endif
692		1183
693	#ifndef ECB_NO_LIBM	1184	#ifndef ECB_NO_LIBM
…		…
713	#else	1204	#else
714	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))	1205	#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
715	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))	1206	#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
716	#endif	1207	#endif
717		1208
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 */	1209	/* convert a float to ieee single/binary32 */
736	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);	1210	ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
737	ecb_function_ ecb_const uint32_t	1211	ecb_function_ ecb_const uint32_t
738	ecb_float_to_binary32 (float x)	1212	ecb_float_to_binary32 (float x)
739	{	1213	{
…		…
870	#endif	1344	#endif
871		1345
872	return r;	1346	return r;
873	}	1347	}
874		1348
875	#endif	1349	/* convert a float to ieee half/binary16 */
		1350	ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
		1351	ecb_function_ ecb_const uint16_t
		1352	ecb_float_to_binary16 (float x)
		1353	{
		1354	return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
		1355	}
876		1356
877	#endif	1357	/* convert an ieee half/binary16 to float */
		1358	ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
		1359	ecb_function_ ecb_const float
		1360	ecb_binary16_to_float (uint16_t x)
		1361	{
		1362	return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
		1363	}
878		1364
		1365	#endif
		1366
		1367	#endif
		1368

Diff Legend

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

Comparing libecb/ecb.h (file contents): Revision 1.165 by sf-exg, Fri May 1 12:02:18 2015 UTC vs. Revision 1.192 by root, Mon Jun 21 23:59:58 2021 UTC

Diff Legend

Comparing libecb/ecb.h (file contents):
Revision 1.165 by sf-exg, Fri May 1 12:02:18 2015 UTC vs.
Revision 1.192 by root, Mon Jun 21 23:59:58 2021 UTC