ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/ecb.h
(Generate patch)

Comparing libeio/ecb.h (file contents):
Revision 1.18 by root, Sat Sep 7 23:18:23 2013 UTC vs.
Revision 1.32 by root, Tue Jul 27 07:58:39 2021 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2012 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:
… …
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- 25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED 26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27 * OF THE POSSIBILITY OF SUCH DAMAGE. 27 * OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * Alternatively, the contents of this file may be used under the terms of
30 * the GNU General Public License ("GPL") version 2 or any later version,
31 * in which case the provisions of the GPL are applicable instead of
32 * the above. If you wish to allow the use of your version of this file
33 * only under the terms of the GPL and not to allow others to use your
34 * version of this file under the BSD license, indicate your decision
35 * by deleting the provisions above and replace them with the notice
36 * and other provisions required by the GPL. If you do not delete the
37 * provisions above, a recipient may use your version of this file under
38 * either the BSD or the GPL.
28 */ 39 */
29 40
30#ifndef ECB_H 41#ifndef ECB_H
31#define ECB_H 42#define ECB_H
32 43
33/* 16 bits major, 16 bits minor */ 44/* 16 bits major, 16 bits minor */
34#define ECB_VERSION 0x00010003 45#define ECB_VERSION 0x00010009
35 46
36#ifdef _WIN32 47#include <string.h> /* for memcpy */
48
49#if defined (_WIN32) && !defined (__MINGW32__)
37 typedef signed char int8_t; 50 typedef signed char int8_t;
38 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;
39 typedef signed short int16_t; 54 typedef signed short int16_t;
40 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;
41 typedef signed int int32_t; 58 typedef signed int int32_t;
42 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;
43 #if __GNUC__ 62 #if __GNUC__
44 typedef signed long long int64_t; 63 typedef signed long long int64_t;
45 typedef unsigned long long uint64_t; 64 typedef unsigned long long uint64_t;
46 #else /* _MSC_VER || __BORLANDC__ */ 65 #else /* _MSC_VER || __BORLANDC__ */
47 typedef signed __int64 int64_t; 66 typedef signed __int64 int64_t;
48 typedef unsigned __int64 uint64_t; 67 typedef unsigned __int64 uint64_t;
49 #endif 68 #endif
69 typedef int64_t int_fast64_t;
70 typedef uint64_t uint_fast64_t;
50 #ifdef _WIN64 71 #ifdef _WIN64
51 #define ECB_PTRSIZE 8 72 #define ECB_PTRSIZE 8
52 typedef uint64_t uintptr_t; 73 typedef uint64_t uintptr_t;
53 typedef int64_t intptr_t; 74 typedef int64_t intptr_t;
54 #else 75 #else
… …
56 typedef uint32_t uintptr_t; 77 typedef uint32_t uintptr_t;
57 typedef int32_t intptr_t; 78 typedef int32_t intptr_t;
58 #endif 79 #endif
59#else 80#else
60 #include <inttypes.h> 81 #include <inttypes.h>
61 #if UINTMAX_MAX > 0xffffffffU 82 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
62 #define ECB_PTRSIZE 8 83 #define ECB_PTRSIZE 8
63 #else 84 #else
64 #define ECB_PTRSIZE 4 85 #define ECB_PTRSIZE 4
65 #endif 86 #endif
66#endif 87#endif
67 88
89#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
90#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
91
92#ifndef ECB_OPTIMIZE_SIZE
93 #if __OPTIMIZE_SIZE__
94 #define ECB_OPTIMIZE_SIZE 1
95 #else
96 #define ECB_OPTIMIZE_SIZE 0
97 #endif
98#endif
99
68/* work around x32 idiocy by defining proper macros */ 100/* work around x32 idiocy by defining proper macros */
69#if __x86_64 || _M_AMD64 101#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
70 #if __ILP32 102 #if _ILP32
71 #define ECB_AMD64_X32 1 103 #define ECB_AMD64_X32 1
72 #else 104 #else
73 #define ECB_AMD64 1 105 #define ECB_AMD64 1
74 #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
75#endif 113#endif
76 114
77/* many compilers define _GNUC_ to some versions but then only implement 115/* many compilers define _GNUC_ to some versions but then only implement
78 * what their idiot authors think are the "more important" extensions, 116 * what their idiot authors think are the "more important" extensions,
79 * causing enormous grief in return for some better fake benchmark numbers. 117 * causing enormous grief in return for some better fake benchmark numbers.
80 * or so. 118 * or so.
81 * we try to detect these and simply assume they are not gcc - if they have 119 * we try to detect these and simply assume they are not gcc - if they have
82 * an issue with that they should have done it right in the first place. 120 * an issue with that they should have done it right in the first place.
83 */ 121 */
84#ifndef ECB_GCC_VERSION
85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ 122#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
86 #define ECB_GCC_VERSION(major,minor) 0 123 #define ECB_GCC_VERSION(major,minor) 0
87 #else 124#else
88 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 125 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
89 #endif 126#endif
90#endif
91 127
92#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ 128#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
93#define ECB_C99 (__STDC_VERSION__ >= 199901L) 129
94#define ECB_C11 (__STDC_VERSION__ >= 201112L) 130#if __clang__ && defined __has_builtin
131 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
132#else
133 #define ECB_CLANG_BUILTIN(x) 0
134#endif
135
136#if __clang__ && defined __has_extension
137 #define ECB_CLANG_EXTENSION(x) __has_extension (x)
138#else
139 #define ECB_CLANG_EXTENSION(x) 0
140#endif
141
95#define ECB_CPP (__cplusplus+0) 142#define ECB_CPP (__cplusplus+0)
96#define ECB_CPP11 (__cplusplus >= 201103L) 143#define ECB_CPP11 (__cplusplus >= 201103L)
144#define ECB_CPP14 (__cplusplus >= 201402L)
145#define ECB_CPP17 (__cplusplus >= 201703L)
146
147#if ECB_CPP
148 #define ECB_C 0
149 #define ECB_STDC_VERSION 0
150#else
151 #define ECB_C 1
152 #define ECB_STDC_VERSION __STDC_VERSION__
153#endif
154
155#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
156#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
157#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
97 158
98#if ECB_CPP 159#if ECB_CPP
99 #define ECB_EXTERN_C extern "C" 160 #define ECB_EXTERN_C extern "C"
100 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 161 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
101 #define ECB_EXTERN_C_END } 162 #define ECB_EXTERN_C_END }
… …
116 177
117#if ECB_NO_SMP 178#if ECB_NO_SMP
118 #define ECB_MEMORY_FENCE do { } while (0) 179 #define ECB_MEMORY_FENCE do { } while (0)
119#endif 180#endif
120 181
182/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
183#if __xlC__ && ECB_CPP
184 #include <builtins.h>
185#endif
186
187#if 1400 <= _MSC_VER
188 #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189#endif
190
121#ifndef ECB_MEMORY_FENCE 191#ifndef ECB_MEMORY_FENCE
122 #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")
123 #if __i386 || __i386__ 194 #if __i386 || __i386__
124 #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")
125 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 196 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
126 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 197 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
127 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 198 #elif ECB_GCC_AMD64
128 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 199 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
129 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 200 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
130 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 201 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
131 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 202 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
132 #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 */
133 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 211 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
134 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 212 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213 || defined __ARM_ARCH_6T2__
135 #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")
136 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 215 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
137 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 216 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
138 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
139 #elif __sparc || __sparc__ 218 #elif __aarch64__
219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
220 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
140 #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")
141 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 222 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 223 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
143 #elif defined __s390__ || defined __s390x__ 224 #elif defined __s390__ || defined __s390x__
144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 225 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
145 #elif defined __mips__ 226 #elif defined __mips__
146 /* GNU/Linux emulates sync on mips1 architectures, so we force it's use */ 227 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
147 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ 228 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
148 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") 229 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
149 #elif defined __alpha__ 230 #elif defined __alpha__
150 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") 231 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
151 #elif defined __hppa__ 232 #elif defined __hppa__
152 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") 233 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
153 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 234 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
154 #elif defined __ia64__ 235 #elif defined __ia64__
155 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") 236 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
237 #elif defined __m68k__
238 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
239 #elif defined __m88k__
240 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
241 #elif defined __sh__
242 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
156 #endif 243 #endif
157 #endif 244 #endif
158#endif 245#endif
159 246
160#ifndef ECB_MEMORY_FENCE 247#ifndef ECB_MEMORY_FENCE
161 #if ECB_GCC_VERSION(4,7) 248 #if ECB_GCC_VERSION(4,7)
162 /* see comment below (stdatomic.h) about the C11 memory model. */ 249 /* see comment below (stdatomic.h) about the C11 memory model. */
163 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 250 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
251 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
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)
164 255
165 /* The __has_feature syntax from clang is so misdesigned that we cannot use it 256 #elif ECB_CLANG_EXTENSION(c_atomic)
166 * without risking compile time errors with other compilers. We *could*
167 * define our own ecb_clang_has_feature, but I just can't be bothered to work
168 * around this shit time and again.
169 * #elif defined __clang && __has_feature (cxx_atomic)
170 * // see comment below (stdatomic.h) about the C11 memory model. 257 /* see comment below (stdatomic.h) about the C11 memory model. */
171 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 258 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
172 */ 259 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
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)
173 263
174 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 264 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
175 #define ECB_MEMORY_FENCE __sync_synchronize () 265 #define ECB_MEMORY_FENCE __sync_synchronize ()
266 #elif _MSC_VER >= 1500 /* VC++ 2008 */
267 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
268 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
269 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
270 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
271 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
176 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 272 #elif _MSC_VER >= 1400 /* VC++ 2005 */
177 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 273 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
178 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 274 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
179 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 275 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
180 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 276 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
181 #elif defined _WIN32 277 #elif defined _WIN32
182 #include <WinNT.h> 278 #include <WinNT.h>
183 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 279 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
184 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 280 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
185 #include <mbarrier.h> 281 #include <mbarrier.h>
186 #define ECB_MEMORY_FENCE __machine_rw_barrier () 282 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
187 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 283 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
188 #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 ()
189 #elif __xlC__ 286 #elif __xlC__
190 #define ECB_MEMORY_FENCE __sync () 287 #define ECB_MEMORY_FENCE __sync ()
191 #endif 288 #endif
192#endif 289#endif
193 290
194#ifndef ECB_MEMORY_FENCE 291#ifndef ECB_MEMORY_FENCE
195 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 292 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
196 /* 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, */
197 /* not just C11 atomics and atomic accesses */ 294 /* not just C11 atomics and atomic accesses */
198 #include <stdatomic.h> 295 #include <stdatomic.h>
199 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
200 /* any fence other than seq_cst, which isn't very efficient for us. */
201 /* Why that is, we don't know - either the C11 memory model is quite useless */
202 /* for most usages, or gcc and clang have a bug */
203 /* I *currently* lean towards the latter, and inefficiently implement */
204 /* all three of ecb's fences as a seq_cst fence */
205 #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)
206 #endif 299 #endif
207#endif 300#endif
208 301
209#ifndef ECB_MEMORY_FENCE 302#ifndef ECB_MEMORY_FENCE
210 #if !ECB_AVOID_PTHREADS 303 #if !ECB_AVOID_PTHREADS
… …
230 323
231#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 324#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
232 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 325 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
233#endif 326#endif
234 327
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
235/*****************************************************************************/ 332/*****************************************************************************/
236 333
237#if __cplusplus 334#if ECB_CPP
238 #define ecb_inline static inline 335 #define ecb_inline static inline
239#elif ECB_GCC_VERSION(2,5) 336#elif ECB_GCC_VERSION(2,5)
240 #define ecb_inline static __inline__ 337 #define ecb_inline static __inline__
241#elif ECB_C99 338#elif ECB_C99
242 #define ecb_inline static inline 339 #define ecb_inline static inline
… …
256 353
257#define ECB_CONCAT_(a, b) a ## b 354#define ECB_CONCAT_(a, b) a ## b
258#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
259#define ECB_STRINGIFY_(a) # a 356#define ECB_STRINGIFY_(a) # a
260#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 357#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
261 359
262#define ecb_function_ ecb_inline 360#define ecb_function_ ecb_inline
263 361
264#if ECB_GCC_VERSION(3,1) 362#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
265 #define ecb_attribute(attrlist) __attribute__(attrlist) 363 #define ecb_attribute(attrlist) __attribute__ (attrlist)
364#else
365 #define ecb_attribute(attrlist)
366#endif
367
368#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
266 #define ecb_is_constant(expr) __builtin_constant_p (expr) 369 #define ecb_is_constant(expr) __builtin_constant_p (expr)
370#else
371 /* possible C11 impl for integral types
372 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
373 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
374
375 #define ecb_is_constant(expr) 0
376#endif
377
378#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
267 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 379 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
380#else
381 #define ecb_expect(expr,value) (expr)
382#endif
383
384#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
268 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 385 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
269#else 386#else
270 #define ecb_attribute(attrlist)
271 #define ecb_is_constant(expr) 0
272 #define ecb_expect(expr,value) (expr)
273 #define ecb_prefetch(addr,rw,locality) 387 #define ecb_prefetch(addr,rw,locality)
274#endif 388#endif
275 389
276/* no emulation for ecb_decltype */ 390/* no emulation for ecb_decltype */
277#if ECB_GCC_VERSION(4,5) 391#if ECB_CPP11
392 // older implementations might have problems with decltype(x)::type, work around it
393 template<class T> struct ecb_decltype_t { typedef T type; };
278 #define ecb_decltype(x) __decltype(x) 394 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
279#elif ECB_GCC_VERSION(3,0) 395#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
280 #define ecb_decltype(x) __typeof(x) 396 #define ecb_decltype(x) __typeof__ (x)
281#endif 397#endif
282 398
399#if _MSC_VER >= 1300
400 #define ecb_deprecated __declspec (deprecated)
401#else
402 #define ecb_deprecated ecb_attribute ((__deprecated__))
403#endif
404
405#if _MSC_VER >= 1500
406 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
407#elif ECB_GCC_VERSION(4,5)
408 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
409#else
410 #define ecb_deprecated_message(msg) ecb_deprecated
411#endif
412
413#if _MSC_VER >= 1400
414 #define ecb_noinline __declspec (noinline)
415#else
283#define ecb_noinline ecb_attribute ((__noinline__)) 416 #define ecb_noinline ecb_attribute ((__noinline__))
417#endif
418
284#define ecb_unused ecb_attribute ((__unused__)) 419#define ecb_unused ecb_attribute ((__unused__))
285#define ecb_const ecb_attribute ((__const__)) 420#define ecb_const ecb_attribute ((__const__))
286#define ecb_pure ecb_attribute ((__pure__)) 421#define ecb_pure ecb_attribute ((__pure__))
287 422
288#if ECB_C11 423#if ECB_C11 || __IBMC_NORETURN
424 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
289 #define ecb_noreturn _Noreturn 425 #define ecb_noreturn _Noreturn
426#elif ECB_CPP11
427 #define ecb_noreturn [[noreturn]]
428#elif _MSC_VER >= 1200
429 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
430 #define ecb_noreturn __declspec (noreturn)
290#else 431#else
291 #define ecb_noreturn ecb_attribute ((__noreturn__)) 432 #define ecb_noreturn ecb_attribute ((__noreturn__))
292#endif 433#endif
293 434
294#if ECB_GCC_VERSION(4,3) 435#if ECB_GCC_VERSION(4,3)
… …
309/* for compatibility to the rest of the world */ 450/* for compatibility to the rest of the world */
310#define ecb_likely(expr) ecb_expect_true (expr) 451#define ecb_likely(expr) ecb_expect_true (expr)
311#define ecb_unlikely(expr) ecb_expect_false (expr) 452#define ecb_unlikely(expr) ecb_expect_false (expr)
312 453
313/* count trailing zero bits and count # of one bits */ 454/* count trailing zero bits and count # of one bits */
314#if ECB_GCC_VERSION(3,4) 455#if ECB_GCC_VERSION(3,4) \
456 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
457 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
458 && ECB_CLANG_BUILTIN(__builtin_popcount))
315 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ 459 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
316 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 460 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
317 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 461 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
318 #define ecb_ctz32(x) __builtin_ctz (x) 462 #define ecb_ctz32(x) __builtin_ctz (x)
319 #define ecb_ctz64(x) __builtin_ctzll (x) 463 #define ecb_ctz64(x) __builtin_ctzll (x)
320 #define ecb_popcount32(x) __builtin_popcount (x) 464 #define ecb_popcount32(x) __builtin_popcount (x)
321 /* no popcountll */ 465 /* no popcountll */
322#else 466#else
323 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 467 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
324 ecb_function_ int 468 ecb_function_ ecb_const int
325 ecb_ctz32 (uint32_t x) 469 ecb_ctz32 (uint32_t x)
326 { 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
327 int r = 0; 476 int r = 0;
328 477
329 x &= ~x + 1; /* this isolates the lowest bit */ 478 x &= ~x + 1; /* this isolates the lowest bit */
330 479
331#if ECB_branchless_on_i386 480#if ECB_branchless_on_i386
… …
341 if (x & 0xff00ff00) r += 8; 490 if (x & 0xff00ff00) r += 8;
342 if (x & 0xffff0000) r += 16; 491 if (x & 0xffff0000) r += 16;
343#endif 492#endif
344 493
345 return r; 494 return r;
495#endif
346 } 496 }
347 497
348 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 498 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
349 ecb_function_ int 499 ecb_function_ ecb_const int
350 ecb_ctz64 (uint64_t x) 500 ecb_ctz64 (uint64_t x)
351 { 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
352 int shift = x & 0xffffffffU ? 0 : 32; 507 int shift = x & 0xffffffff ? 0 : 32;
353 return ecb_ctz32 (x >> shift) + shift; 508 return ecb_ctz32 (x >> shift) + shift;
509#endif
354 } 510 }
355 511
356 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 512 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
357 ecb_function_ int 513 ecb_function_ ecb_const int
358 ecb_popcount32 (uint32_t x) 514 ecb_popcount32 (uint32_t x)
359 { 515 {
360 x -= (x >> 1) & 0x55555555; 516 x -= (x >> 1) & 0x55555555;
361 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 517 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
362 x = ((x >> 4) + x) & 0x0f0f0f0f; 518 x = ((x >> 4) + x) & 0x0f0f0f0f;
363 x *= 0x01010101; 519 x *= 0x01010101;
364 520
365 return x >> 24; 521 return x >> 24;
366 } 522 }
367 523
368 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 524 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
369 ecb_function_ int ecb_ld32 (uint32_t x) 525 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
370 { 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
371 int r = 0; 532 int r = 0;
372 533
373 if (x >> 16) { x >>= 16; r += 16; } 534 if (x >> 16) { x >>= 16; r += 16; }
374 if (x >> 8) { x >>= 8; r += 8; } 535 if (x >> 8) { x >>= 8; r += 8; }
375 if (x >> 4) { x >>= 4; r += 4; } 536 if (x >> 4) { x >>= 4; r += 4; }
376 if (x >> 2) { x >>= 2; r += 2; } 537 if (x >> 2) { x >>= 2; r += 2; }
377 if (x >> 1) { r += 1; } 538 if (x >> 1) { r += 1; }
378 539
379 return r; 540 return r;
541#endif
380 } 542 }
381 543
382 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 544 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
383 ecb_function_ int ecb_ld64 (uint64_t x) 545 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
384 { 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
385 int r = 0; 552 int r = 0;
386 553
387 if (x >> 32) { x >>= 32; r += 32; } 554 if (x >> 32) { x >>= 32; r += 32; }
388 555
389 return r + ecb_ld32 (x); 556 return r + ecb_ld32 (x);
390 }
391#endif 557#endif
558 }
559#endif
392 560
393ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 561ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
394ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 562ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
395ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 563ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
396ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 564ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
397 565
398ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 566ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
399ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 567ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
400{ 568{
401 return ( (x * 0x0802U & 0x22110U) 569 return ( (x * 0x0802U & 0x22110U)
402 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 570 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
403} 571}
404 572
405ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 573ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
406ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 574ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
407{ 575{
408 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 576 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
409 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 577 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
410 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 578 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
411 x = ( x >> 8 ) | ( x << 8); 579 x = ( x >> 8 ) | ( x << 8);
412 580
413 return x; 581 return x;
414} 582}
415 583
416ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 584ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
417ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 585ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
418{ 586{
419 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 587 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
420 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 588 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
421 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 589 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
422 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 590 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
… …
425 return x; 593 return x;
426} 594}
427 595
428/* popcount64 is only available on 64 bit cpus as gcc builtin */ 596/* popcount64 is only available on 64 bit cpus as gcc builtin */
429/* so for this version we are lazy */ 597/* so for this version we are lazy */
430ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 598ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
431ecb_function_ int 599ecb_function_ ecb_const int
432ecb_popcount64 (uint64_t x) 600ecb_popcount64 (uint64_t x)
433{ 601{
434 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 602 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
435} 603}
436 604
437ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 605ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
438ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 606ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
439ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 607ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
440ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 608ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
441ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 609ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
442ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 610ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
443ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 611ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
444ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 612ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
445 613
446ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 614ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
447ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 615ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
448ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 616ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
449ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 617ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
450ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 618ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
451ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 619ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
452ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 620ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
453ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 621ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
454 622
455#if ECB_GCC_VERSION(4,3) 623#if ECB_CPP
624
625inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
626inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
627inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
628inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
629
630inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
631inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
632inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
633inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
634
635inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
636inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
637inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
638inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
639
640inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
641inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
642inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
643inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
644
645inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
646inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
647inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
648
649inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
650inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
651inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
652inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
653
654inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
655inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
656inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
657inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
658
659#endif
660
661#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
662 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
663 #define ecb_bswap16(x) __builtin_bswap16 (x)
664 #else
456 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 665 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
666 #endif
457 #define ecb_bswap32(x) __builtin_bswap32 (x) 667 #define ecb_bswap32(x) __builtin_bswap32 (x)
458 #define ecb_bswap64(x) __builtin_bswap64 (x) 668 #define ecb_bswap64(x) __builtin_bswap64 (x)
669#elif _MSC_VER
670 #include <stdlib.h>
671 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
672 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
673 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
459#else 674#else
460 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 675 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
461 ecb_function_ uint16_t 676 ecb_function_ ecb_const uint16_t
462 ecb_bswap16 (uint16_t x) 677 ecb_bswap16 (uint16_t x)
463 { 678 {
464 return ecb_rotl16 (x, 8); 679 return ecb_rotl16 (x, 8);
465 } 680 }
466 681
467 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 682 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
468 ecb_function_ uint32_t 683 ecb_function_ ecb_const uint32_t
469 ecb_bswap32 (uint32_t x) 684 ecb_bswap32 (uint32_t x)
470 { 685 {
471 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 686 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
472 } 687 }
473 688
474 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 689 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
475 ecb_function_ uint64_t 690 ecb_function_ ecb_const uint64_t
476 ecb_bswap64 (uint64_t x) 691 ecb_bswap64 (uint64_t x)
477 { 692 {
478 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 693 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
479 } 694 }
480#endif 695#endif
481 696
482#if ECB_GCC_VERSION(4,5) 697#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
483 #define ecb_unreachable() __builtin_unreachable () 698 #define ecb_unreachable() __builtin_unreachable ()
484#else 699#else
485 /* this seems to work fine, but gcc always emits a warning for it :/ */ 700 /* this seems to work fine, but gcc always emits a warning for it :/ */
486 ecb_inline void ecb_unreachable (void) ecb_noreturn; 701 ecb_inline ecb_noreturn void ecb_unreachable (void);
487 ecb_inline void ecb_unreachable (void) { } 702 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
488#endif 703#endif
489 704
490/* try to tell the compiler that some condition is definitely true */ 705/* try to tell the compiler that some condition is definitely true */
491#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 706#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
492 707
493ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 708ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
494ecb_inline unsigned char 709ecb_inline ecb_const uint32_t
495ecb_byteorder_helper (void) 710ecb_byteorder_helper (void)
496{ 711{
497 /* the union code still generates code under pressure in gcc, */ 712 /* the union code still generates code under pressure in gcc, */
498 /* but less than using pointers, and always seems to */ 713 /* but less than using pointers, and always seems to */
499 /* successfully return a constant. */ 714 /* successfully return a constant. */
500 /* the reason why we have this horrible preprocessor mess */ 715 /* the reason why we have this horrible preprocessor mess */
501 /* 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 */
502 /* or when using a recent enough gcc version (>= 4.6) */ 717 /* or when using a recent enough gcc version (>= 4.6) */
503#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
504 return 0x44;
505#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
506 return 0x44; 721 return 0x44332211;
507#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
508 return 0x11; 725 return 0x11223344;
509#else 726#else
510 union 727 union
511 { 728 {
729 uint8_t c[4];
512 uint32_t i; 730 uint32_t u;
513 uint8_t c;
514 } u = { 0x11223344 }; 731 } u = { 0x11, 0x22, 0x33, 0x44 };
515 return u.c; 732 return u.u;
516#endif 733#endif
517} 734}
518 735
519ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 736ecb_inline ecb_const ecb_bool ecb_big_endian (void);
520ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 737ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
521ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 738ecb_inline ecb_const ecb_bool ecb_little_endian (void);
522ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 739ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
740
741/*****************************************************************************/
742/* unaligned load/store */
743
744ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
745ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
746ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
747
748ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
749ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
750ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
751
752ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
754ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
755
756ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
757ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
758ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
759
760ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
761ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
762ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
763
764ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
765ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
766ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
767
768ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
769ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
770ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
771
772ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
773ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
774ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
775
776ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
777ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
778ecb_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
780ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
781ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
782ecb_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
786inline uint8_t ecb_bswap (uint8_t v) { return v; }
787inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
788inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
789inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
790
791template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
792template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
793template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
794template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
795template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
796template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
797template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
798template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
799
800template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
801template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
802template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
803template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
804template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
805template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
806template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
807template<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 */
523 813
524#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 */
525 #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))
526#else 817#else
527 #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)))
528#endif 819#endif
529 820
530#if __cplusplus 821#if ECB_CPP
531 template<typename T> 822 template<typename T>
532 static inline T ecb_div_rd (T val, T div) 823 static inline T ecb_div_rd (T val, T div)
533 { 824 {
534 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 825 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
535 } 826 }
… …
540 } 831 }
541#else 832#else
542 #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))
543 #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))
544#endif 835#endif
836
837/*****************************************************************************/
838/* array length */
545 839
546#if ecb_cplusplus_does_not_suck 840#if ecb_cplusplus_does_not_suck
547 /* 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) */
548 template<typename T, int N> 842 template<typename T, int N>
549 static inline int ecb_array_length (const T (&arr)[N]) 843 static inline int ecb_array_length (const T (&arr)[N])
… …
552 } 846 }
553#else 847#else
554 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 848 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
555#endif 849#endif
556 850
851/*****************************************************************************/
852/* IEEE 754-2008 half float conversions */
853
854ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
855ecb_function_ ecb_const uint32_t
856ecb_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
886ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
887ecb_function_ ecb_const uint16_t
888ecb_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/*
954 * This code is pretty complicated because it is general. The idea behind it,
955 * however, is pretty simple: first, the number is multiplied with a scaling
956 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
957 * number with the first digit in the upper bits.
958 * Then this digit is converted to text and masked out. The resulting number
959 * is then multiplied by 10, by multiplying the fixed point representation
960 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
961 * format becomes 5.27, 6.26 and so on.
962 * The rest involves only advancing the pointer if we already generated a
963 * non-zero digit, so leading zeroes are overwritten.
964 */
965
966// simply return a mask with "bits" bits set
967#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
968
969// oputput a single digit. maskvalue is 10**digitidx
970#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
971 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
972 { \
973 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
974 *ptr = digit + '0'; /* output it */ \
975 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
976 ptr += nz; /* output digit only if non-zero digit seen */ \
977 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
978 }
979
980// convert integer to fixed point format and multiply out digits, highest first
981// requires magic constants: max. digits and number of bits after the decimal point
982#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
983ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
984{ \
985 char nz = lz; /* non-zero digit seen? */ \
986 /* convert to x.bits fixed-point */ \
987 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
988 /* output up to 10 digits */ \
989 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
990 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
991 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
992 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
993 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
994 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
995 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
996 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
997 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
998 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
999 return ptr; \
1000}
1001
1002// predefined versions of the above, for various digits
1003// ecb_i2a_xN = almost N digits, limit defined by macro
1004// ecb_i2a_N = up to N digits, leading zeroes suppressed
1005// ecb_i2a_0N = exactly N digits, including leading zeroes
1006
1007// non-leading-zero versions, limited range
1008#define ECB_I2A_MAX_X5 59074 // limit for ecb_i2a_x5
1009#define ECB_I2A_MAX_X10 2932500665 // limit for ecb_i2a_x10
1010ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1011ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1012
1013// non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit
1014ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1015ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1016ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1017ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1018ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1019ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1020ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1021ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1022
1023// leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit
1024ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1025ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1026ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1027ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1028ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1029ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1030ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1031ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1032
1033#define ECB_I2A_I32_DIGITS 11
1034#define ECB_I2A_U32_DIGITS 10
1035#define ECB_I2A_I64_DIGITS 20
1036#define ECB_I2A_U64_DIGITS 21
1037#define ECB_I2A_MAX_DIGITS 21
1038
1039ecb_inline char *
1040ecb_i2a_u32 (char *ptr, uint32_t u)
1041{
1042 #if ECB_64BIT_NATIVE
1043 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1044 ptr = ecb_i2a_x10 (ptr, u);
1045 else // x10 almost, but not fully, covers 32 bit
1046 {
1047 uint32_t u1 = u % 1000000000;
1048 uint32_t u2 = u / 1000000000;
1049
1050 *ptr++ = u2 + '0';
1051 ptr = ecb_i2a_09 (ptr, u1);
1052 }
1053 #else
1054 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1055 ecb_i2a_x5 (ptr, u);
1056 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1057 {
1058 uint32_t u1 = u % 10000;
1059 uint32_t u2 = u / 10000;
1060
1061 ptr = ecb_i2a_x5 (ptr, u2);
1062 ptr = ecb_i2a_04 (ptr, u1);
1063 }
1064 else
1065 {
1066 uint32_t u1 = u % 10000;
1067 uint32_t ua = u / 10000;
1068 uint32_t u2 = ua % 10000;
1069 uint32_t u3 = ua / 10000;
1070
1071 ptr = ecb_i2a_2 (ptr, u3);
1072 ptr = ecb_i2a_04 (ptr, u2);
1073 ptr = ecb_i2a_04 (ptr, u1);
1074 }
1075 #endif
1076
1077 return ptr;
1078}
1079
1080ecb_inline char *
1081ecb_i2a_i32 (char *ptr, int32_t v)
1082{
1083 *ptr = '-'; ptr += v < 0;
1084 uint32_t u = v < 0 ? -(uint32_t)v : v;
1085
1086 #if ECB_64BIT_NATIVE
1087 ptr = ecb_i2a_x10 (ptr, u); // x10 fully covers 31 bit
1088 #else
1089 ptr = ecb_i2a_u32 (ptr, u);
1090 #endif
1091
1092 return ptr;
1093}
1094
1095ecb_inline char *
1096ecb_i2a_u64 (char *ptr, uint64_t u)
1097{
1098 #if ECB_64BIT_NATIVE
1099 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1100 ptr = ecb_i2a_x10 (ptr, u);
1101 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1102 {
1103 uint64_t u1 = u % 1000000000;
1104 uint64_t u2 = u / 1000000000;
1105
1106 ptr = ecb_i2a_x10 (ptr, u2);
1107 ptr = ecb_i2a_09 (ptr, u1);
1108 }
1109 else
1110 {
1111 uint64_t u1 = u % 1000000000;
1112 uint64_t ua = u / 1000000000;
1113 uint64_t u2 = ua % 1000000000;
1114 uint64_t u3 = ua / 1000000000;
1115
1116 ptr = ecb_i2a_2 (ptr, u3);
1117 ptr = ecb_i2a_09 (ptr, u2);
1118 ptr = ecb_i2a_09 (ptr, u1);
1119 }
1120 #else
1121 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1122 ptr = ecb_i2a_x5 (ptr, u);
1123 else
1124 {
1125 uint64_t u1 = u % 10000;
1126 uint64_t u2 = u / 10000;
1127
1128 ptr = ecb_i2a_u64 (ptr, u2);
1129 ptr = ecb_i2a_04 (ptr, u1);
1130 }
1131 #endif
1132
1133 return ptr;
1134}
1135
1136ecb_inline char *
1137ecb_i2a_i64 (char *ptr, int64_t v)
1138{
1139 *ptr = '-'; ptr += v < 0;
1140 uint64_t u = v < 0 ? -(uint64_t)v : v;
1141
1142 #if ECB_64BIT_NATIVE
1143 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1144 ptr = ecb_i2a_x10 (ptr, u);
1145 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1146 {
1147 uint64_t u1 = u % 1000000000;
1148 uint64_t u2 = u / 1000000000;
1149
1150 ptr = ecb_i2a_x10 (ptr, u2);
1151 ptr = ecb_i2a_09 (ptr, u1);
1152 }
1153 else
1154 {
1155 uint64_t u1 = u % 1000000000;
1156 uint64_t ua = u / 1000000000;
1157 uint64_t u2 = ua % 1000000000;
1158 uint64_t u3 = ua / 1000000000;
1159
1160 // 2**31 is 19 digits, so the top is exactly one digit
1161 *ptr++ = u3 + '0';
1162 ptr = ecb_i2a_09 (ptr, u2);
1163 ptr = ecb_i2a_09 (ptr, u1);
1164 }
1165 #else
1166 ptr = ecb_i2a_u64 (ptr, u);
1167 #endif
1168
1169 return ptr;
1170}
1171
557/*******************************************************************************/ 1172/*******************************************************************************/
558/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 1173/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
559 1174
560/* basically, everything uses "ieee pure-endian" floating point numbers */ 1175/* basically, everything uses "ieee pure-endian" floating point numbers */
561/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 1176/* the only noteworthy exception is ancient armle, which uses order 43218765 */
562#if 0 \ 1177#if 0 \
563 || __i386 || __i386__ \ 1178 || __i386 || __i386__ \
564 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 1179 || ECB_GCC_AMD64 \
565 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 1180 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
566 || defined __arm__ && defined __ARM_EABI__ \
567 || defined __s390__ || defined __s390x__ \ 1181 || defined __s390__ || defined __s390x__ \
568 || defined __mips__ \ 1182 || defined __mips__ \
569 || defined __alpha__ \ 1183 || defined __alpha__ \
570 || defined __hppa__ \ 1184 || defined __hppa__ \
571 || defined __ia64__ \ 1185 || defined __ia64__ \
1186 || defined __m68k__ \
1187 || defined __m88k__ \
1188 || defined __sh__ \
572 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 1189 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1190 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1191 || defined __aarch64__
573 #define ECB_STDFP 1 1192 #define ECB_STDFP 1
574 #include <string.h> /* for memcpy */
575#else 1193#else
576 #define ECB_STDFP 0 1194 #define ECB_STDFP 0
577 #include <math.h> /* for frexp*, ldexp* */
578#endif 1195#endif
579 1196
580#ifndef ECB_NO_LIBM 1197#ifndef ECB_NO_LIBM
581 1198
1199 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
1200
1201 /* only the oldest of old doesn't have this one. solaris. */
1202 #ifdef INFINITY
1203 #define ECB_INFINITY INFINITY
1204 #else
1205 #define ECB_INFINITY HUGE_VAL
1206 #endif
1207
1208 #ifdef NAN
1209 #define ECB_NAN NAN
1210 #else
1211 #define ECB_NAN ECB_INFINITY
1212 #endif
1213
1214 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
1215 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1216 #define ecb_frexpf(x,e) frexpf ((x), (e))
1217 #else
1218 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1219 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1220 #endif
1221
582 /* convert a float to ieee single/binary32 */ 1222 /* convert a float to ieee single/binary32 */
583 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 1223 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
584 ecb_function_ uint32_t 1224 ecb_function_ ecb_const uint32_t
585 ecb_float_to_binary32 (float x) 1225 ecb_float_to_binary32 (float x)
586 { 1226 {
587 uint32_t r; 1227 uint32_t r;
588 1228
589 #if ECB_STDFP 1229 #if ECB_STDFP
… …
596 if (x == 0e0f ) return 0x00000000U; 1236 if (x == 0e0f ) return 0x00000000U;
597 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 1237 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
598 if (x < -3.40282346638528860e+38f) return 0xff800000U; 1238 if (x < -3.40282346638528860e+38f) return 0xff800000U;
599 if (x != x ) return 0x7fbfffffU; 1239 if (x != x ) return 0x7fbfffffU;
600 1240
601 m = frexpf (x, &e) * 0x1000000U; 1241 m = ecb_frexpf (x, &e) * 0x1000000U;
602 1242
603 r = m & 0x80000000U; 1243 r = m & 0x80000000U;
604 1244
605 if (r) 1245 if (r)
606 m = -m; 1246 m = -m;
… …
618 1258
619 return r; 1259 return r;
620 } 1260 }
621 1261
622 /* converts an ieee single/binary32 to a float */ 1262 /* converts an ieee single/binary32 to a float */
623 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 1263 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
624 ecb_function_ float 1264 ecb_function_ ecb_const float
625 ecb_binary32_to_float (uint32_t x) 1265 ecb_binary32_to_float (uint32_t x)
626 { 1266 {
627 float r; 1267 float r;
628 1268
629 #if ECB_STDFP 1269 #if ECB_STDFP
… …
639 x |= 0x800000U; 1279 x |= 0x800000U;
640 else 1280 else
641 e = 1; 1281 e = 1;
642 1282
643 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ 1283 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
644 r = ldexpf (x * (0.5f / 0x800000U), e - 126); 1284 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
645 1285
646 r = neg ? -r : r; 1286 r = neg ? -r : r;
647 #endif 1287 #endif
648 1288
649 return r; 1289 return r;
650 } 1290 }
651 1291
652 /* convert a double to ieee double/binary64 */ 1292 /* convert a double to ieee double/binary64 */
653 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 1293 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
654 ecb_function_ uint64_t 1294 ecb_function_ ecb_const uint64_t
655 ecb_double_to_binary64 (double x) 1295 ecb_double_to_binary64 (double x)
656 { 1296 {
657 uint64_t r; 1297 uint64_t r;
658 1298
659 #if ECB_STDFP 1299 #if ECB_STDFP
… …
688 1328
689 return r; 1329 return r;
690 } 1330 }
691 1331
692 /* converts an ieee double/binary64 to a double */ 1332 /* converts an ieee double/binary64 to a double */
693 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 1333 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
694 ecb_function_ double 1334 ecb_function_ ecb_const double
695 ecb_binary64_to_double (uint64_t x) 1335 ecb_binary64_to_double (uint64_t x)
696 { 1336 {
697 double r; 1337 double r;
698 1338
699 #if ECB_STDFP 1339 #if ECB_STDFP
… …
717 #endif 1357 #endif
718 1358
719 return r; 1359 return r;
720 } 1360 }
721 1361
722#endif 1362 /* convert a float to ieee half/binary16 */
1363 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1364 ecb_function_ ecb_const uint16_t
1365 ecb_float_to_binary16 (float x)
1366 {
1367 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1368 }
723 1369
724#endif 1370 /* convert an ieee half/binary16 to float */
1371 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1372 ecb_function_ ecb_const float
1373 ecb_binary16_to_float (uint16_t x)
1374 {
1375 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1376 }
725 1377
1378#endif
1379
1380#endif
1381

Diff Legend

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