1 |
root |
1.1 |
/* |
2 |
root |
1.17 |
* libecb - http://software.schmorp.de/pkg/libecb |
3 |
root |
1.1 |
* |
4 |
root |
1.152 |
* Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de> |
5 |
root |
1.7 |
* Copyright (©) 2011 Emanuele Giaquinta |
6 |
root |
1.1 |
* All rights reserved. |
7 |
|
|
* |
8 |
|
|
* Redistribution and use in source and binary forms, with or without modifica- |
9 |
|
|
* tion, are permitted provided that the following conditions are met: |
10 |
|
|
* |
11 |
|
|
* 1. Redistributions of source code must retain the above copyright notice, |
12 |
|
|
* this list of conditions and the following disclaimer. |
13 |
|
|
* |
14 |
|
|
* 2. Redistributions in binary form must reproduce the above copyright |
15 |
|
|
* notice, this list of conditions and the following disclaimer in the |
16 |
|
|
* documentation and/or other materials provided with the distribution. |
17 |
|
|
* |
18 |
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED |
19 |
|
|
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
20 |
|
|
* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
21 |
|
|
* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- |
22 |
|
|
* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
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, |
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 |
27 |
|
|
* OF THE POSSIBILITY OF SUCH DAMAGE. |
28 |
root |
1.133 |
* |
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. |
39 |
root |
1.1 |
*/ |
40 |
|
|
|
41 |
|
|
#ifndef ECB_H |
42 |
|
|
#define ECB_H |
43 |
|
|
|
44 |
root |
1.87 |
/* 16 bits major, 16 bits minor */ |
45 |
root |
1.167 |
#define ECB_VERSION 0x00010005 |
46 |
root |
1.87 |
|
47 |
root |
1.44 |
#ifdef _WIN32 |
48 |
|
|
typedef signed char int8_t; |
49 |
|
|
typedef unsigned char uint8_t; |
50 |
|
|
typedef signed short int16_t; |
51 |
|
|
typedef unsigned short uint16_t; |
52 |
|
|
typedef signed int int32_t; |
53 |
|
|
typedef unsigned int uint32_t; |
54 |
|
|
#if __GNUC__ |
55 |
|
|
typedef signed long long int64_t; |
56 |
|
|
typedef unsigned long long uint64_t; |
57 |
root |
1.51 |
#else /* _MSC_VER || __BORLANDC__ */ |
58 |
root |
1.44 |
typedef signed __int64 int64_t; |
59 |
|
|
typedef unsigned __int64 uint64_t; |
60 |
|
|
#endif |
61 |
root |
1.87 |
#ifdef _WIN64 |
62 |
|
|
#define ECB_PTRSIZE 8 |
63 |
|
|
typedef uint64_t uintptr_t; |
64 |
|
|
typedef int64_t intptr_t; |
65 |
|
|
#else |
66 |
|
|
#define ECB_PTRSIZE 4 |
67 |
|
|
typedef uint32_t uintptr_t; |
68 |
|
|
typedef int32_t intptr_t; |
69 |
|
|
#endif |
70 |
root |
1.44 |
#else |
71 |
|
|
#include <inttypes.h> |
72 |
root |
1.173 |
#if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
73 |
root |
1.87 |
#define ECB_PTRSIZE 8 |
74 |
|
|
#else |
75 |
|
|
#define ECB_PTRSIZE 4 |
76 |
|
|
#endif |
77 |
root |
1.44 |
#endif |
78 |
root |
1.6 |
|
79 |
sf-exg |
1.159 |
#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
80 |
|
|
#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
81 |
|
|
|
82 |
root |
1.114 |
/* work around x32 idiocy by defining proper macros */ |
83 |
sf-exg |
1.159 |
#if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
84 |
root |
1.119 |
#if _ILP32 |
85 |
root |
1.115 |
#define ECB_AMD64_X32 1 |
86 |
root |
1.114 |
#else |
87 |
root |
1.115 |
#define ECB_AMD64 1 |
88 |
root |
1.114 |
#endif |
89 |
|
|
#endif |
90 |
|
|
|
91 |
root |
1.12 |
/* many compilers define _GNUC_ to some versions but then only implement |
92 |
|
|
* what their idiot authors think are the "more important" extensions, |
93 |
sf-exg |
1.59 |
* causing enormous grief in return for some better fake benchmark numbers. |
94 |
root |
1.18 |
* or so. |
95 |
root |
1.12 |
* we try to detect these and simply assume they are not gcc - if they have |
96 |
|
|
* an issue with that they should have done it right in the first place. |
97 |
|
|
*/ |
98 |
root |
1.137 |
#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
99 |
|
|
#define ECB_GCC_VERSION(major,minor) 0 |
100 |
|
|
#else |
101 |
|
|
#define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
102 |
root |
1.12 |
#endif |
103 |
root |
1.1 |
|
104 |
sf-exg |
1.138 |
#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
105 |
|
|
|
106 |
root |
1.147 |
#if __clang__ && defined __has_builtin |
107 |
|
|
#define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
108 |
sf-exg |
1.138 |
#else |
109 |
|
|
#define ECB_CLANG_BUILTIN(x) 0 |
110 |
|
|
#endif |
111 |
|
|
|
112 |
root |
1.147 |
#if __clang__ && defined __has_extension |
113 |
|
|
#define ECB_CLANG_EXTENSION(x) __has_extension (x) |
114 |
sf-exg |
1.140 |
#else |
115 |
|
|
#define ECB_CLANG_EXTENSION(x) 0 |
116 |
|
|
#endif |
117 |
|
|
|
118 |
root |
1.91 |
#define ECB_CPP (__cplusplus+0) |
119 |
|
|
#define ECB_CPP11 (__cplusplus >= 201103L) |
120 |
root |
1.177 |
#define ECB_CPP14 (__cplusplus >= 201402L) |
121 |
|
|
#define ECB_CPP17 (__cplusplus >= 201703L) |
122 |
root |
1.90 |
|
123 |
root |
1.102 |
#if ECB_CPP |
124 |
root |
1.127 |
#define ECB_C 0 |
125 |
|
|
#define ECB_STDC_VERSION 0 |
126 |
|
|
#else |
127 |
|
|
#define ECB_C 1 |
128 |
|
|
#define ECB_STDC_VERSION __STDC_VERSION__ |
129 |
|
|
#endif |
130 |
|
|
|
131 |
|
|
#define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
132 |
|
|
#define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
133 |
root |
1.177 |
#define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
134 |
root |
1.127 |
|
135 |
|
|
#if ECB_CPP |
136 |
root |
1.102 |
#define ECB_EXTERN_C extern "C" |
137 |
|
|
#define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
138 |
|
|
#define ECB_EXTERN_C_END } |
139 |
|
|
#else |
140 |
|
|
#define ECB_EXTERN_C extern |
141 |
|
|
#define ECB_EXTERN_C_BEG |
142 |
|
|
#define ECB_EXTERN_C_END |
143 |
|
|
#endif |
144 |
|
|
|
145 |
root |
1.52 |
/*****************************************************************************/ |
146 |
|
|
|
147 |
root |
1.58 |
/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
148 |
|
|
/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
149 |
|
|
|
150 |
root |
1.79 |
#if ECB_NO_THREADS |
151 |
root |
1.95 |
#define ECB_NO_SMP 1 |
152 |
root |
1.79 |
#endif |
153 |
|
|
|
154 |
root |
1.93 |
#if ECB_NO_SMP |
155 |
root |
1.64 |
#define ECB_MEMORY_FENCE do { } while (0) |
156 |
root |
1.58 |
#endif |
157 |
|
|
|
158 |
sf-exg |
1.165 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
159 |
|
|
#if __xlC__ && ECB_CPP |
160 |
|
|
#include <builtins.h> |
161 |
|
|
#endif |
162 |
|
|
|
163 |
root |
1.171 |
#if 1400 <= _MSC_VER |
164 |
|
|
#include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
165 |
|
|
#endif |
166 |
|
|
|
167 |
root |
1.52 |
#ifndef ECB_MEMORY_FENCE |
168 |
root |
1.85 |
#if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
169 |
root |
1.73 |
#if __i386 || __i386__ |
170 |
root |
1.54 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
171 |
root |
1.94 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
172 |
root |
1.176 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
173 |
sf-exg |
1.159 |
#elif ECB_GCC_AMD64 |
174 |
root |
1.94 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
175 |
|
|
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
176 |
root |
1.176 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
177 |
root |
1.63 |
#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
178 |
root |
1.94 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
179 |
root |
1.175 |
#elif defined __ARM_ARCH_2__ \ |
180 |
|
|
|| defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
181 |
|
|
|| defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
182 |
|
|
|| defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
183 |
|
|
|| defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
184 |
|
|
|| defined __ARM_ARCH_5TEJ__ |
185 |
|
|
/* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
186 |
root |
1.85 |
#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
187 |
root |
1.175 |
|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
188 |
|
|
|| defined __ARM_ARCH_6T2__ |
189 |
root |
1.84 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
190 |
root |
1.85 |
#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
191 |
root |
1.175 |
|| defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
192 |
root |
1.94 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
193 |
root |
1.129 |
#elif __aarch64__ |
194 |
|
|
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
195 |
root |
1.166 |
#elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
196 |
root |
1.94 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
197 |
|
|
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
198 |
|
|
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
199 |
root |
1.85 |
#elif defined __s390__ || defined __s390x__ |
200 |
root |
1.77 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
201 |
root |
1.85 |
#elif defined __mips__ |
202 |
root |
1.118 |
/* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
203 |
root |
1.116 |
/* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
204 |
|
|
#define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
205 |
root |
1.86 |
#elif defined __alpha__ |
206 |
root |
1.94 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
207 |
|
|
#elif defined __hppa__ |
208 |
|
|
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
209 |
|
|
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
210 |
|
|
#elif defined __ia64__ |
211 |
|
|
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
212 |
root |
1.117 |
#elif defined __m68k__ |
213 |
|
|
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
214 |
|
|
#elif defined __m88k__ |
215 |
|
|
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
216 |
|
|
#elif defined __sh__ |
217 |
|
|
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
218 |
root |
1.52 |
#endif |
219 |
|
|
#endif |
220 |
|
|
#endif |
221 |
|
|
|
222 |
|
|
#ifndef ECB_MEMORY_FENCE |
223 |
root |
1.93 |
#if ECB_GCC_VERSION(4,7) |
224 |
root |
1.97 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
225 |
root |
1.93 |
#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
226 |
root |
1.128 |
#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
227 |
|
|
#define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
228 |
root |
1.110 |
|
229 |
sf-exg |
1.140 |
#elif ECB_CLANG_EXTENSION(c_atomic) |
230 |
|
|
/* see comment below (stdatomic.h) about the C11 memory model. */ |
231 |
|
|
#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
232 |
|
|
#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
233 |
|
|
#define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
234 |
root |
1.110 |
|
235 |
root |
1.93 |
#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
236 |
root |
1.52 |
#define ECB_MEMORY_FENCE __sync_synchronize () |
237 |
root |
1.126 |
#elif _MSC_VER >= 1500 /* VC++ 2008 */ |
238 |
|
|
/* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
239 |
|
|
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
240 |
|
|
#define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
241 |
|
|
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
242 |
|
|
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
243 |
root |
1.57 |
#elif _MSC_VER >= 1400 /* VC++ 2005 */ |
244 |
|
|
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
245 |
|
|
#define ECB_MEMORY_FENCE _ReadWriteBarrier () |
246 |
|
|
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
247 |
|
|
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
248 |
root |
1.85 |
#elif defined _WIN32 |
249 |
root |
1.55 |
#include <WinNT.h> |
250 |
root |
1.57 |
#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
251 |
root |
1.72 |
#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
252 |
|
|
#include <mbarrier.h> |
253 |
|
|
#define ECB_MEMORY_FENCE __machine_rw_barrier () |
254 |
|
|
#define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
255 |
|
|
#define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
256 |
root |
1.82 |
#elif __xlC__ |
257 |
root |
1.83 |
#define ECB_MEMORY_FENCE __sync () |
258 |
root |
1.52 |
#endif |
259 |
|
|
#endif |
260 |
|
|
|
261 |
root |
1.53 |
#ifndef ECB_MEMORY_FENCE |
262 |
root |
1.94 |
#if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
263 |
|
|
/* we assume that these memory fences work on all variables/all memory accesses, */ |
264 |
|
|
/* not just C11 atomics and atomic accesses */ |
265 |
|
|
#include <stdatomic.h> |
266 |
root |
1.97 |
/* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
267 |
|
|
/* any fence other than seq_cst, which isn't very efficient for us. */ |
268 |
|
|
/* Why that is, we don't know - either the C11 memory model is quite useless */ |
269 |
|
|
/* for most usages, or gcc and clang have a bug */ |
270 |
|
|
/* I *currently* lean towards the latter, and inefficiently implement */ |
271 |
|
|
/* all three of ecb's fences as a seq_cst fence */ |
272 |
root |
1.128 |
/* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
273 |
|
|
/* for all __atomic_thread_fence's except seq_cst */ |
274 |
root |
1.94 |
#define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
275 |
|
|
#endif |
276 |
|
|
#endif |
277 |
|
|
|
278 |
|
|
#ifndef ECB_MEMORY_FENCE |
279 |
root |
1.62 |
#if !ECB_AVOID_PTHREADS |
280 |
|
|
/* |
281 |
|
|
* if you get undefined symbol references to pthread_mutex_lock, |
282 |
|
|
* or failure to find pthread.h, then you should implement |
283 |
|
|
* the ECB_MEMORY_FENCE operations for your cpu/compiler |
284 |
|
|
* OR provide pthread.h and link against the posix thread library |
285 |
|
|
* of your system. |
286 |
|
|
*/ |
287 |
|
|
#include <pthread.h> |
288 |
|
|
#define ECB_NEEDS_PTHREADS 1 |
289 |
|
|
#define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
290 |
root |
1.52 |
|
291 |
root |
1.62 |
static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
292 |
|
|
#define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
293 |
|
|
#endif |
294 |
|
|
#endif |
295 |
|
|
|
296 |
root |
1.85 |
#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
297 |
root |
1.52 |
#define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
298 |
root |
1.62 |
#endif |
299 |
|
|
|
300 |
root |
1.85 |
#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
301 |
root |
1.52 |
#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
302 |
|
|
#endif |
303 |
|
|
|
304 |
|
|
/*****************************************************************************/ |
305 |
|
|
|
306 |
root |
1.149 |
#if ECB_CPP |
307 |
root |
1.46 |
#define ecb_inline static inline |
308 |
root |
1.38 |
#elif ECB_GCC_VERSION(2,5) |
309 |
root |
1.46 |
#define ecb_inline static __inline__ |
310 |
root |
1.39 |
#elif ECB_C99 |
311 |
root |
1.46 |
#define ecb_inline static inline |
312 |
root |
1.29 |
#else |
313 |
root |
1.46 |
#define ecb_inline static |
314 |
root |
1.38 |
#endif |
315 |
|
|
|
316 |
|
|
#if ECB_GCC_VERSION(3,3) |
317 |
|
|
#define ecb_restrict __restrict__ |
318 |
root |
1.39 |
#elif ECB_C99 |
319 |
root |
1.38 |
#define ecb_restrict restrict |
320 |
|
|
#else |
321 |
|
|
#define ecb_restrict |
322 |
root |
1.4 |
#endif |
323 |
|
|
|
324 |
root |
1.38 |
typedef int ecb_bool; |
325 |
|
|
|
326 |
root |
1.8 |
#define ECB_CONCAT_(a, b) a ## b |
327 |
|
|
#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
328 |
|
|
#define ECB_STRINGIFY_(a) # a |
329 |
|
|
#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
330 |
root |
1.155 |
#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
331 |
root |
1.8 |
|
332 |
root |
1.46 |
#define ecb_function_ ecb_inline |
333 |
root |
1.3 |
|
334 |
sf-exg |
1.138 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
335 |
root |
1.142 |
#define ecb_attribute(attrlist) __attribute__ (attrlist) |
336 |
root |
1.37 |
#else |
337 |
|
|
#define ecb_attribute(attrlist) |
338 |
sf-exg |
1.138 |
#endif |
339 |
root |
1.127 |
|
340 |
sf-exg |
1.138 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
341 |
|
|
#define ecb_is_constant(expr) __builtin_constant_p (expr) |
342 |
|
|
#else |
343 |
root |
1.127 |
/* possible C11 impl for integral types |
344 |
|
|
typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
345 |
|
|
#define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
346 |
|
|
|
347 |
root |
1.37 |
#define ecb_is_constant(expr) 0 |
348 |
sf-exg |
1.138 |
#endif |
349 |
|
|
|
350 |
|
|
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
351 |
|
|
#define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
352 |
|
|
#else |
353 |
root |
1.37 |
#define ecb_expect(expr,value) (expr) |
354 |
sf-exg |
1.138 |
#endif |
355 |
|
|
|
356 |
|
|
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
357 |
|
|
#define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
358 |
|
|
#else |
359 |
root |
1.37 |
#define ecb_prefetch(addr,rw,locality) |
360 |
root |
1.1 |
#endif |
361 |
|
|
|
362 |
root |
1.2 |
/* no emulation for ecb_decltype */ |
363 |
root |
1.143 |
#if ECB_CPP11 |
364 |
root |
1.144 |
// older implementations might have problems with decltype(x)::type, work around it |
365 |
root |
1.146 |
template<class T> struct ecb_decltype_t { typedef T type; }; |
366 |
|
|
#define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
367 |
root |
1.143 |
#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
368 |
|
|
#define ecb_decltype(x) __typeof__ (x) |
369 |
root |
1.1 |
#endif |
370 |
|
|
|
371 |
root |
1.135 |
#if _MSC_VER >= 1300 |
372 |
root |
1.149 |
#define ecb_deprecated __declspec (deprecated) |
373 |
root |
1.135 |
#else |
374 |
|
|
#define ecb_deprecated ecb_attribute ((__deprecated__)) |
375 |
|
|
#endif |
376 |
|
|
|
377 |
sf-exg |
1.162 |
#if _MSC_VER >= 1500 |
378 |
root |
1.154 |
#define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
379 |
|
|
#elif ECB_GCC_VERSION(4,5) |
380 |
|
|
#define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
381 |
|
|
#else |
382 |
|
|
#define ecb_deprecated_message(msg) ecb_deprecated |
383 |
|
|
#endif |
384 |
|
|
|
385 |
|
|
#if _MSC_VER >= 1400 |
386 |
|
|
#define ecb_noinline __declspec (noinline) |
387 |
|
|
#else |
388 |
|
|
#define ecb_noinline ecb_attribute ((__noinline__)) |
389 |
|
|
#endif |
390 |
|
|
|
391 |
root |
1.24 |
#define ecb_unused ecb_attribute ((__unused__)) |
392 |
|
|
#define ecb_const ecb_attribute ((__const__)) |
393 |
|
|
#define ecb_pure ecb_attribute ((__pure__)) |
394 |
root |
1.35 |
|
395 |
root |
1.145 |
#if ECB_C11 || __IBMC_NORETURN |
396 |
sf-exg |
1.165 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
397 |
root |
1.90 |
#define ecb_noreturn _Noreturn |
398 |
root |
1.153 |
#elif ECB_CPP11 |
399 |
|
|
#define ecb_noreturn [[noreturn]] |
400 |
|
|
#elif _MSC_VER >= 1200 |
401 |
sf-exg |
1.156 |
/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
402 |
root |
1.153 |
#define ecb_noreturn __declspec (noreturn) |
403 |
root |
1.90 |
#else |
404 |
|
|
#define ecb_noreturn ecb_attribute ((__noreturn__)) |
405 |
|
|
#endif |
406 |
|
|
|
407 |
root |
1.35 |
#if ECB_GCC_VERSION(4,3) |
408 |
root |
1.39 |
#define ecb_artificial ecb_attribute ((__artificial__)) |
409 |
|
|
#define ecb_hot ecb_attribute ((__hot__)) |
410 |
|
|
#define ecb_cold ecb_attribute ((__cold__)) |
411 |
root |
1.35 |
#else |
412 |
|
|
#define ecb_artificial |
413 |
|
|
#define ecb_hot |
414 |
|
|
#define ecb_cold |
415 |
|
|
#endif |
416 |
root |
1.1 |
|
417 |
root |
1.39 |
/* put around conditional expressions if you are very sure that the */ |
418 |
|
|
/* expression is mostly true or mostly false. note that these return */ |
419 |
|
|
/* booleans, not the expression. */ |
420 |
root |
1.33 |
#define ecb_expect_false(expr) ecb_expect (!!(expr), 0) |
421 |
|
|
#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) |
422 |
root |
1.36 |
/* for compatibility to the rest of the world */ |
423 |
root |
1.33 |
#define ecb_likely(expr) ecb_expect_true (expr) |
424 |
|
|
#define ecb_unlikely(expr) ecb_expect_false (expr) |
425 |
root |
1.1 |
|
426 |
root |
1.3 |
/* count trailing zero bits and count # of one bits */ |
427 |
root |
1.139 |
#if ECB_GCC_VERSION(3,4) \ |
428 |
|
|
|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
429 |
|
|
&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
430 |
|
|
&& ECB_CLANG_BUILTIN(__builtin_popcount)) |
431 |
root |
1.49 |
/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
432 |
|
|
#define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
433 |
|
|
#define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
434 |
root |
1.35 |
#define ecb_ctz32(x) __builtin_ctz (x) |
435 |
root |
1.49 |
#define ecb_ctz64(x) __builtin_ctzll (x) |
436 |
root |
1.35 |
#define ecb_popcount32(x) __builtin_popcount (x) |
437 |
root |
1.49 |
/* no popcountll */ |
438 |
root |
1.1 |
#else |
439 |
root |
1.151 |
ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
440 |
|
|
ecb_function_ ecb_const int |
441 |
root |
1.35 |
ecb_ctz32 (uint32_t x) |
442 |
|
|
{ |
443 |
root |
1.171 |
#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
444 |
root |
1.172 |
unsigned long r; |
445 |
root |
1.171 |
_BitScanForward (&r, x); |
446 |
|
|
return (int)r; |
447 |
|
|
#else |
448 |
root |
1.35 |
int r = 0; |
449 |
|
|
|
450 |
root |
1.48 |
x &= ~x + 1; /* this isolates the lowest bit */ |
451 |
root |
1.35 |
|
452 |
root |
1.50 |
#if ECB_branchless_on_i386 |
453 |
|
|
r += !!(x & 0xaaaaaaaa) << 0; |
454 |
|
|
r += !!(x & 0xcccccccc) << 1; |
455 |
|
|
r += !!(x & 0xf0f0f0f0) << 2; |
456 |
|
|
r += !!(x & 0xff00ff00) << 3; |
457 |
|
|
r += !!(x & 0xffff0000) << 4; |
458 |
|
|
#else |
459 |
root |
1.35 |
if (x & 0xaaaaaaaa) r += 1; |
460 |
|
|
if (x & 0xcccccccc) r += 2; |
461 |
|
|
if (x & 0xf0f0f0f0) r += 4; |
462 |
|
|
if (x & 0xff00ff00) r += 8; |
463 |
|
|
if (x & 0xffff0000) r += 16; |
464 |
root |
1.50 |
#endif |
465 |
root |
1.35 |
|
466 |
|
|
return r; |
467 |
root |
1.171 |
#endif |
468 |
root |
1.35 |
} |
469 |
|
|
|
470 |
root |
1.151 |
ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
471 |
|
|
ecb_function_ ecb_const int |
472 |
root |
1.49 |
ecb_ctz64 (uint64_t x) |
473 |
|
|
{ |
474 |
root |
1.171 |
#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
475 |
root |
1.172 |
unsigned long r; |
476 |
root |
1.171 |
_BitScanForward64 (&r, x); |
477 |
|
|
return (int)r; |
478 |
|
|
#else |
479 |
root |
1.168 |
int shift = x & 0xffffffff ? 0 : 32; |
480 |
root |
1.50 |
return ecb_ctz32 (x >> shift) + shift; |
481 |
root |
1.171 |
#endif |
482 |
root |
1.49 |
} |
483 |
|
|
|
484 |
root |
1.151 |
ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
485 |
|
|
ecb_function_ ecb_const int |
486 |
root |
1.35 |
ecb_popcount32 (uint32_t x) |
487 |
|
|
{ |
488 |
|
|
x -= (x >> 1) & 0x55555555; |
489 |
|
|
x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
490 |
|
|
x = ((x >> 4) + x) & 0x0f0f0f0f; |
491 |
|
|
x *= 0x01010101; |
492 |
root |
1.1 |
|
493 |
root |
1.35 |
return x >> 24; |
494 |
|
|
} |
495 |
root |
1.49 |
|
496 |
root |
1.151 |
ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
497 |
|
|
ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
498 |
root |
1.49 |
{ |
499 |
root |
1.171 |
#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
500 |
root |
1.172 |
unsigned long r; |
501 |
root |
1.171 |
_BitScanReverse (&r, x); |
502 |
|
|
return (int)r; |
503 |
|
|
#else |
504 |
root |
1.50 |
int r = 0; |
505 |
root |
1.49 |
|
506 |
root |
1.50 |
if (x >> 16) { x >>= 16; r += 16; } |
507 |
|
|
if (x >> 8) { x >>= 8; r += 8; } |
508 |
|
|
if (x >> 4) { x >>= 4; r += 4; } |
509 |
|
|
if (x >> 2) { x >>= 2; r += 2; } |
510 |
|
|
if (x >> 1) { r += 1; } |
511 |
root |
1.49 |
|
512 |
|
|
return r; |
513 |
root |
1.171 |
#endif |
514 |
root |
1.49 |
} |
515 |
|
|
|
516 |
root |
1.151 |
ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
517 |
|
|
ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
518 |
root |
1.49 |
{ |
519 |
root |
1.171 |
#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
520 |
root |
1.172 |
unsigned long r; |
521 |
root |
1.171 |
_BitScanReverse64 (&r, x); |
522 |
|
|
return (int)r; |
523 |
|
|
#else |
524 |
root |
1.50 |
int r = 0; |
525 |
root |
1.49 |
|
526 |
root |
1.50 |
if (x >> 32) { x >>= 32; r += 32; } |
527 |
root |
1.49 |
|
528 |
root |
1.50 |
return r + ecb_ld32 (x); |
529 |
root |
1.171 |
#endif |
530 |
root |
1.49 |
} |
531 |
root |
1.1 |
#endif |
532 |
|
|
|
533 |
root |
1.151 |
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
534 |
|
|
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
535 |
|
|
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
536 |
|
|
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
537 |
root |
1.88 |
|
538 |
root |
1.151 |
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
539 |
|
|
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
540 |
root |
1.70 |
{ |
541 |
|
|
return ( (x * 0x0802U & 0x22110U) |
542 |
root |
1.151 |
| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
543 |
root |
1.70 |
} |
544 |
|
|
|
545 |
root |
1.151 |
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
546 |
|
|
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
547 |
root |
1.70 |
{ |
548 |
|
|
x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
549 |
|
|
x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
550 |
|
|
x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
551 |
|
|
x = ( x >> 8 ) | ( x << 8); |
552 |
|
|
|
553 |
|
|
return x; |
554 |
|
|
} |
555 |
|
|
|
556 |
root |
1.151 |
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
557 |
|
|
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
558 |
root |
1.70 |
{ |
559 |
|
|
x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
560 |
|
|
x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
561 |
|
|
x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
562 |
|
|
x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
563 |
|
|
x = ( x >> 16 ) | ( x << 16); |
564 |
|
|
|
565 |
|
|
return x; |
566 |
|
|
} |
567 |
|
|
|
568 |
root |
1.49 |
/* popcount64 is only available on 64 bit cpus as gcc builtin */ |
569 |
|
|
/* so for this version we are lazy */ |
570 |
root |
1.151 |
ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
571 |
|
|
ecb_function_ ecb_const int |
572 |
root |
1.49 |
ecb_popcount64 (uint64_t x) |
573 |
|
|
{ |
574 |
|
|
return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
575 |
|
|
} |
576 |
|
|
|
577 |
root |
1.151 |
ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
578 |
|
|
ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
579 |
|
|
ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
580 |
|
|
ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
581 |
|
|
ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
582 |
|
|
ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
583 |
|
|
ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
584 |
|
|
ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
585 |
|
|
|
586 |
|
|
ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
587 |
|
|
ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
588 |
|
|
ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
589 |
|
|
ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
590 |
|
|
ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
591 |
|
|
ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
592 |
|
|
ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
593 |
|
|
ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
594 |
root |
1.50 |
|
595 |
sf-exg |
1.138 |
#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
596 |
root |
1.164 |
#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
597 |
|
|
#define ecb_bswap16(x) __builtin_bswap16 (x) |
598 |
|
|
#else |
599 |
root |
1.49 |
#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
600 |
root |
1.164 |
#endif |
601 |
root |
1.49 |
#define ecb_bswap32(x) __builtin_bswap32 (x) |
602 |
|
|
#define ecb_bswap64(x) __builtin_bswap64 (x) |
603 |
root |
1.164 |
#elif _MSC_VER |
604 |
|
|
#include <stdlib.h> |
605 |
|
|
#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
606 |
|
|
#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
607 |
|
|
#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
608 |
root |
1.13 |
#else |
609 |
root |
1.151 |
ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
610 |
|
|
ecb_function_ ecb_const uint16_t |
611 |
root |
1.50 |
ecb_bswap16 (uint16_t x) |
612 |
root |
1.49 |
{ |
613 |
root |
1.50 |
return ecb_rotl16 (x, 8); |
614 |
root |
1.49 |
} |
615 |
|
|
|
616 |
root |
1.151 |
ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
617 |
|
|
ecb_function_ ecb_const uint32_t |
618 |
root |
1.35 |
ecb_bswap32 (uint32_t x) |
619 |
|
|
{ |
620 |
root |
1.50 |
return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
621 |
root |
1.35 |
} |
622 |
|
|
|
623 |
root |
1.151 |
ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
624 |
|
|
ecb_function_ ecb_const uint64_t |
625 |
root |
1.49 |
ecb_bswap64 (uint64_t x) |
626 |
root |
1.35 |
{ |
627 |
root |
1.50 |
return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
628 |
root |
1.35 |
} |
629 |
root |
1.13 |
#endif |
630 |
|
|
|
631 |
sf-exg |
1.138 |
#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
632 |
root |
1.35 |
#define ecb_unreachable() __builtin_unreachable () |
633 |
root |
1.13 |
#else |
634 |
root |
1.35 |
/* this seems to work fine, but gcc always emits a warning for it :/ */ |
635 |
root |
1.151 |
ecb_inline ecb_noreturn void ecb_unreachable (void); |
636 |
|
|
ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
637 |
root |
1.13 |
#endif |
638 |
|
|
|
639 |
root |
1.41 |
/* try to tell the compiler that some condition is definitely true */ |
640 |
root |
1.100 |
#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
641 |
root |
1.41 |
|
642 |
root |
1.174 |
ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
643 |
|
|
ecb_inline ecb_const uint32_t |
644 |
root |
1.23 |
ecb_byteorder_helper (void) |
645 |
root |
1.3 |
{ |
646 |
root |
1.98 |
/* the union code still generates code under pressure in gcc, */ |
647 |
sf-exg |
1.111 |
/* but less than using pointers, and always seems to */ |
648 |
root |
1.98 |
/* successfully return a constant. */ |
649 |
|
|
/* the reason why we have this horrible preprocessor mess */ |
650 |
root |
1.99 |
/* is to avoid it in all cases, at least on common architectures */ |
651 |
sf-exg |
1.111 |
/* or when using a recent enough gcc version (>= 4.6) */ |
652 |
root |
1.174 |
#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
653 |
|
|
|| ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
654 |
|
|
#define ECB_LITTLE_ENDIAN 1 |
655 |
|
|
return 0x44332211; |
656 |
|
|
#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
657 |
|
|
|| ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
658 |
|
|
#define ECB_BIG_ENDIAN 1 |
659 |
|
|
return 0x11223344; |
660 |
root |
1.98 |
#else |
661 |
|
|
union |
662 |
|
|
{ |
663 |
root |
1.174 |
uint8_t c[4]; |
664 |
|
|
uint32_t u; |
665 |
|
|
} u = { 0x11, 0x22, 0x33, 0x44 }; |
666 |
|
|
return u.u; |
667 |
root |
1.98 |
#endif |
668 |
root |
1.3 |
} |
669 |
|
|
|
670 |
root |
1.151 |
ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
671 |
root |
1.174 |
ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
672 |
root |
1.151 |
ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
673 |
root |
1.174 |
ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
674 |
root |
1.3 |
|
675 |
root |
1.39 |
#if ECB_GCC_VERSION(3,0) || ECB_C99 |
676 |
root |
1.35 |
#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
677 |
root |
1.31 |
#else |
678 |
root |
1.35 |
#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
679 |
root |
1.31 |
#endif |
680 |
root |
1.21 |
|
681 |
root |
1.149 |
#if ECB_CPP |
682 |
sf-exg |
1.68 |
template<typename T> |
683 |
|
|
static inline T ecb_div_rd (T val, T div) |
684 |
|
|
{ |
685 |
|
|
return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
686 |
|
|
} |
687 |
|
|
template<typename T> |
688 |
|
|
static inline T ecb_div_ru (T val, T div) |
689 |
|
|
{ |
690 |
|
|
return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
691 |
|
|
} |
692 |
|
|
#else |
693 |
|
|
#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
694 |
|
|
#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
695 |
|
|
#endif |
696 |
sf-exg |
1.67 |
|
697 |
root |
1.5 |
#if ecb_cplusplus_does_not_suck |
698 |
root |
1.40 |
/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
699 |
root |
1.35 |
template<typename T, int N> |
700 |
|
|
static inline int ecb_array_length (const T (&arr)[N]) |
701 |
|
|
{ |
702 |
|
|
return N; |
703 |
|
|
} |
704 |
root |
1.5 |
#else |
705 |
root |
1.35 |
#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
706 |
root |
1.5 |
#endif |
707 |
|
|
|
708 |
root |
1.170 |
ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
709 |
root |
1.167 |
ecb_function_ ecb_const uint32_t |
710 |
root |
1.170 |
ecb_binary16_to_binary32 (uint32_t x) |
711 |
root |
1.167 |
{ |
712 |
|
|
unsigned int s = (x & 0x8000) << (31 - 15); |
713 |
|
|
int e = (x >> 10) & 0x001f; |
714 |
|
|
unsigned int m = x & 0x03ff; |
715 |
|
|
|
716 |
|
|
if (ecb_expect_false (e == 31)) |
717 |
|
|
/* infinity or NaN */ |
718 |
|
|
e = 255 - (127 - 15); |
719 |
|
|
else if (ecb_expect_false (!e)) |
720 |
|
|
{ |
721 |
|
|
if (ecb_expect_true (!m)) |
722 |
|
|
/* zero, handled by code below by forcing e to 0 */ |
723 |
|
|
e = 0 - (127 - 15); |
724 |
|
|
else |
725 |
|
|
{ |
726 |
|
|
/* subnormal, renormalise */ |
727 |
|
|
unsigned int s = 10 - ecb_ld32 (m); |
728 |
|
|
|
729 |
|
|
m = (m << s) & 0x3ff; /* mask implicit bit */ |
730 |
|
|
e -= s - 1; |
731 |
|
|
} |
732 |
|
|
} |
733 |
|
|
|
734 |
|
|
/* e and m now are normalised, or zero, (or inf or nan) */ |
735 |
|
|
e += 127 - 15; |
736 |
|
|
|
737 |
|
|
return s | (e << 23) | (m << (23 - 10)); |
738 |
|
|
} |
739 |
|
|
|
740 |
|
|
ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
741 |
|
|
ecb_function_ ecb_const uint16_t |
742 |
|
|
ecb_binary32_to_binary16 (uint32_t x) |
743 |
|
|
{ |
744 |
|
|
unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
745 |
|
|
unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
746 |
|
|
unsigned int m = x & 0x007fffff; |
747 |
|
|
|
748 |
|
|
x &= 0x7fffffff; |
749 |
|
|
|
750 |
|
|
/* if it's within range of binary16 normals, use fast path */ |
751 |
|
|
if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
752 |
|
|
{ |
753 |
|
|
/* mantissa round-to-even */ |
754 |
|
|
m += 0x00000fff + ((m >> (23 - 10)) & 1); |
755 |
|
|
|
756 |
|
|
/* handle overflow */ |
757 |
|
|
if (ecb_expect_false (m >= 0x00800000)) |
758 |
|
|
{ |
759 |
|
|
m >>= 1; |
760 |
|
|
e += 1; |
761 |
|
|
} |
762 |
|
|
|
763 |
|
|
return s | (e << 10) | (m >> (23 - 10)); |
764 |
|
|
} |
765 |
|
|
|
766 |
|
|
/* handle large numbers and infinity */ |
767 |
|
|
if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
768 |
|
|
return s | 0x7c00; |
769 |
|
|
|
770 |
root |
1.169 |
/* handle zero, subnormals and small numbers */ |
771 |
root |
1.167 |
if (ecb_expect_true (x < 0x38800000)) |
772 |
|
|
{ |
773 |
|
|
/* zero */ |
774 |
|
|
if (ecb_expect_true (!x)) |
775 |
|
|
return s; |
776 |
|
|
|
777 |
|
|
/* handle subnormals */ |
778 |
|
|
|
779 |
root |
1.169 |
/* too small, will be zero */ |
780 |
|
|
if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
781 |
|
|
return s; |
782 |
|
|
|
783 |
root |
1.167 |
m |= 0x00800000; /* make implicit bit explicit */ |
784 |
|
|
|
785 |
|
|
/* very tricky - we need to round to the nearest e (+10) bit value */ |
786 |
|
|
{ |
787 |
|
|
unsigned int bits = 14 - e; |
788 |
|
|
unsigned int half = (1 << (bits - 1)) - 1; |
789 |
|
|
unsigned int even = (m >> bits) & 1; |
790 |
|
|
|
791 |
|
|
/* if this overflows, we will end up with a normalised number */ |
792 |
|
|
m = (m + half + even) >> bits; |
793 |
|
|
} |
794 |
|
|
|
795 |
|
|
return s | m; |
796 |
|
|
} |
797 |
|
|
|
798 |
|
|
/* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
799 |
|
|
m >>= 13; |
800 |
|
|
|
801 |
|
|
return s | 0x7c00 | m | !m; |
802 |
|
|
} |
803 |
|
|
|
804 |
root |
1.104 |
/*******************************************************************************/ |
805 |
|
|
/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
806 |
|
|
|
807 |
|
|
/* basically, everything uses "ieee pure-endian" floating point numbers */ |
808 |
|
|
/* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
809 |
|
|
#if 0 \ |
810 |
|
|
|| __i386 || __i386__ \ |
811 |
sf-exg |
1.159 |
|| ECB_GCC_AMD64 \ |
812 |
root |
1.104 |
|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
813 |
|
|
|| defined __s390__ || defined __s390x__ \ |
814 |
|
|
|| defined __mips__ \ |
815 |
|
|
|| defined __alpha__ \ |
816 |
|
|
|| defined __hppa__ \ |
817 |
|
|
|| defined __ia64__ \ |
818 |
root |
1.117 |
|| defined __m68k__ \ |
819 |
|
|
|| defined __m88k__ \ |
820 |
|
|
|| defined __sh__ \ |
821 |
sf-exg |
1.159 |
|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
822 |
root |
1.131 |
|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
823 |
root |
1.132 |
|| defined __aarch64__ |
824 |
root |
1.104 |
#define ECB_STDFP 1 |
825 |
|
|
#include <string.h> /* for memcpy */ |
826 |
root |
1.102 |
#else |
827 |
root |
1.104 |
#define ECB_STDFP 0 |
828 |
root |
1.102 |
#endif |
829 |
|
|
|
830 |
root |
1.104 |
#ifndef ECB_NO_LIBM |
831 |
root |
1.103 |
|
832 |
root |
1.121 |
#include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
833 |
|
|
|
834 |
root |
1.122 |
/* only the oldest of old doesn't have this one. solaris. */ |
835 |
|
|
#ifdef INFINITY |
836 |
|
|
#define ECB_INFINITY INFINITY |
837 |
|
|
#else |
838 |
|
|
#define ECB_INFINITY HUGE_VAL |
839 |
|
|
#endif |
840 |
|
|
|
841 |
|
|
#ifdef NAN |
842 |
root |
1.121 |
#define ECB_NAN NAN |
843 |
|
|
#else |
844 |
root |
1.122 |
#define ECB_NAN ECB_INFINITY |
845 |
root |
1.121 |
#endif |
846 |
root |
1.120 |
|
847 |
root |
1.148 |
#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
848 |
root |
1.150 |
#define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
849 |
sf-exg |
1.163 |
#define ecb_frexpf(x,e) frexpf ((x), (e)) |
850 |
root |
1.148 |
#else |
851 |
sf-exg |
1.161 |
#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
852 |
sf-exg |
1.163 |
#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
853 |
root |
1.148 |
#endif |
854 |
|
|
|
855 |
root |
1.104 |
/* convert a float to ieee single/binary32 */ |
856 |
root |
1.151 |
ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
857 |
|
|
ecb_function_ ecb_const uint32_t |
858 |
root |
1.103 |
ecb_float_to_binary32 (float x) |
859 |
|
|
{ |
860 |
|
|
uint32_t r; |
861 |
|
|
|
862 |
|
|
#if ECB_STDFP |
863 |
root |
1.104 |
memcpy (&r, &x, 4); |
864 |
root |
1.103 |
#else |
865 |
root |
1.105 |
/* slow emulation, works for anything but -0 */ |
866 |
root |
1.103 |
uint32_t m; |
867 |
|
|
int e; |
868 |
|
|
|
869 |
root |
1.108 |
if (x == 0e0f ) return 0x00000000U; |
870 |
root |
1.103 |
if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
871 |
|
|
if (x < -3.40282346638528860e+38f) return 0xff800000U; |
872 |
root |
1.105 |
if (x != x ) return 0x7fbfffffU; |
873 |
root |
1.103 |
|
874 |
sf-exg |
1.163 |
m = ecb_frexpf (x, &e) * 0x1000000U; |
875 |
root |
1.103 |
|
876 |
|
|
r = m & 0x80000000U; |
877 |
|
|
|
878 |
|
|
if (r) |
879 |
|
|
m = -m; |
880 |
|
|
|
881 |
root |
1.108 |
if (e <= -126) |
882 |
root |
1.103 |
{ |
883 |
|
|
m &= 0xffffffU; |
884 |
|
|
m >>= (-125 - e); |
885 |
|
|
e = -126; |
886 |
|
|
} |
887 |
|
|
|
888 |
|
|
r |= (e + 126) << 23; |
889 |
|
|
r |= m & 0x7fffffU; |
890 |
|
|
#endif |
891 |
|
|
|
892 |
|
|
return r; |
893 |
|
|
} |
894 |
|
|
|
895 |
root |
1.104 |
/* converts an ieee single/binary32 to a float */ |
896 |
root |
1.151 |
ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
897 |
|
|
ecb_function_ ecb_const float |
898 |
root |
1.103 |
ecb_binary32_to_float (uint32_t x) |
899 |
|
|
{ |
900 |
|
|
float r; |
901 |
|
|
|
902 |
|
|
#if ECB_STDFP |
903 |
root |
1.104 |
memcpy (&r, &x, 4); |
904 |
root |
1.103 |
#else |
905 |
|
|
/* emulation, only works for normals and subnormals and +0 */ |
906 |
|
|
int neg = x >> 31; |
907 |
|
|
int e = (x >> 23) & 0xffU; |
908 |
|
|
|
909 |
|
|
x &= 0x7fffffU; |
910 |
|
|
|
911 |
|
|
if (e) |
912 |
|
|
x |= 0x800000U; |
913 |
root |
1.104 |
else |
914 |
|
|
e = 1; |
915 |
root |
1.103 |
|
916 |
|
|
/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
917 |
root |
1.148 |
r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
918 |
root |
1.103 |
|
919 |
|
|
r = neg ? -r : r; |
920 |
|
|
#endif |
921 |
|
|
|
922 |
|
|
return r; |
923 |
|
|
} |
924 |
|
|
|
925 |
root |
1.104 |
/* convert a double to ieee double/binary64 */ |
926 |
root |
1.151 |
ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
927 |
|
|
ecb_function_ ecb_const uint64_t |
928 |
root |
1.103 |
ecb_double_to_binary64 (double x) |
929 |
|
|
{ |
930 |
root |
1.104 |
uint64_t r; |
931 |
|
|
|
932 |
|
|
#if ECB_STDFP |
933 |
|
|
memcpy (&r, &x, 8); |
934 |
|
|
#else |
935 |
root |
1.105 |
/* slow emulation, works for anything but -0 */ |
936 |
root |
1.104 |
uint64_t m; |
937 |
|
|
int e; |
938 |
|
|
|
939 |
root |
1.108 |
if (x == 0e0 ) return 0x0000000000000000U; |
940 |
root |
1.104 |
if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
941 |
|
|
if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
942 |
root |
1.105 |
if (x != x ) return 0X7ff7ffffffffffffU; |
943 |
root |
1.104 |
|
944 |
|
|
m = frexp (x, &e) * 0x20000000000000U; |
945 |
|
|
|
946 |
|
|
r = m & 0x8000000000000000;; |
947 |
|
|
|
948 |
|
|
if (r) |
949 |
|
|
m = -m; |
950 |
|
|
|
951 |
root |
1.108 |
if (e <= -1022) |
952 |
root |
1.104 |
{ |
953 |
|
|
m &= 0x1fffffffffffffU; |
954 |
|
|
m >>= (-1021 - e); |
955 |
|
|
e = -1022; |
956 |
|
|
} |
957 |
|
|
|
958 |
|
|
r |= ((uint64_t)(e + 1022)) << 52; |
959 |
|
|
r |= m & 0xfffffffffffffU; |
960 |
|
|
#endif |
961 |
|
|
|
962 |
|
|
return r; |
963 |
|
|
} |
964 |
|
|
|
965 |
|
|
/* converts an ieee double/binary64 to a double */ |
966 |
root |
1.151 |
ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
967 |
|
|
ecb_function_ ecb_const double |
968 |
root |
1.104 |
ecb_binary64_to_double (uint64_t x) |
969 |
|
|
{ |
970 |
|
|
double r; |
971 |
|
|
|
972 |
|
|
#if ECB_STDFP |
973 |
|
|
memcpy (&r, &x, 8); |
974 |
|
|
#else |
975 |
|
|
/* emulation, only works for normals and subnormals and +0 */ |
976 |
|
|
int neg = x >> 63; |
977 |
|
|
int e = (x >> 52) & 0x7ffU; |
978 |
|
|
|
979 |
|
|
x &= 0xfffffffffffffU; |
980 |
|
|
|
981 |
|
|
if (e) |
982 |
|
|
x |= 0x10000000000000U; |
983 |
|
|
else |
984 |
|
|
e = 1; |
985 |
|
|
|
986 |
root |
1.107 |
/* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
987 |
root |
1.108 |
r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
988 |
root |
1.104 |
|
989 |
|
|
r = neg ? -r : r; |
990 |
|
|
#endif |
991 |
|
|
|
992 |
|
|
return r; |
993 |
root |
1.103 |
} |
994 |
|
|
|
995 |
root |
1.167 |
/* convert a float to ieee half/binary16 */ |
996 |
|
|
ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
997 |
|
|
ecb_function_ ecb_const uint16_t |
998 |
|
|
ecb_float_to_binary16 (float x) |
999 |
|
|
{ |
1000 |
|
|
return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
1001 |
|
|
} |
1002 |
|
|
|
1003 |
|
|
/* convert an ieee half/binary16 to float */ |
1004 |
|
|
ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
1005 |
|
|
ecb_function_ ecb_const float |
1006 |
|
|
ecb_binary16_to_float (uint16_t x) |
1007 |
|
|
{ |
1008 |
|
|
return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
1009 |
|
|
} |
1010 |
|
|
|
1011 |
root |
1.103 |
#endif |
1012 |
root |
1.102 |
|
1013 |
root |
1.1 |
#endif |
1014 |
|
|
|