1 |
/* |
2 |
* libecb - http://software.schmorp.de/pkg/libecb |
3 |
* |
4 |
* Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de> |
5 |
* Copyright (©) 2011 Emanuele Giaquinta |
6 |
* 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 |
* |
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 |
*/ |
40 |
|
41 |
#ifndef ECB_H |
42 |
#define ECB_H |
43 |
|
44 |
/* 16 bits major, 16 bits minor */ |
45 |
#define ECB_VERSION 0x00010004 |
46 |
|
47 |
#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 |
#else /* _MSC_VER || __BORLANDC__ */ |
58 |
typedef signed __int64 int64_t; |
59 |
typedef unsigned __int64 uint64_t; |
60 |
#endif |
61 |
#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 |
#else |
71 |
#include <inttypes.h> |
72 |
#if UINTMAX_MAX > 0xffffffffU |
73 |
#define ECB_PTRSIZE 8 |
74 |
#else |
75 |
#define ECB_PTRSIZE 4 |
76 |
#endif |
77 |
#endif |
78 |
|
79 |
#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
80 |
#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
81 |
|
82 |
/* work around x32 idiocy by defining proper macros */ |
83 |
#if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
84 |
#if _ILP32 |
85 |
#define ECB_AMD64_X32 1 |
86 |
#else |
87 |
#define ECB_AMD64 1 |
88 |
#endif |
89 |
#endif |
90 |
|
91 |
/* many compilers define _GNUC_ to some versions but then only implement |
92 |
* what their idiot authors think are the "more important" extensions, |
93 |
* causing enormous grief in return for some better fake benchmark numbers. |
94 |
* or so. |
95 |
* 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 |
#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 |
#endif |
103 |
|
104 |
#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
105 |
|
106 |
#if __clang__ && defined __has_builtin |
107 |
#define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
108 |
#else |
109 |
#define ECB_CLANG_BUILTIN(x) 0 |
110 |
#endif |
111 |
|
112 |
#if __clang__ && defined __has_extension |
113 |
#define ECB_CLANG_EXTENSION(x) __has_extension (x) |
114 |
#else |
115 |
#define ECB_CLANG_EXTENSION(x) 0 |
116 |
#endif |
117 |
|
118 |
#define ECB_CPP (__cplusplus+0) |
119 |
#define ECB_CPP11 (__cplusplus >= 201103L) |
120 |
|
121 |
#if ECB_CPP |
122 |
#define ECB_C 0 |
123 |
#define ECB_STDC_VERSION 0 |
124 |
#else |
125 |
#define ECB_C 1 |
126 |
#define ECB_STDC_VERSION __STDC_VERSION__ |
127 |
#endif |
128 |
|
129 |
#define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
130 |
#define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
131 |
|
132 |
#if ECB_CPP |
133 |
#define ECB_EXTERN_C extern "C" |
134 |
#define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
135 |
#define ECB_EXTERN_C_END } |
136 |
#else |
137 |
#define ECB_EXTERN_C extern |
138 |
#define ECB_EXTERN_C_BEG |
139 |
#define ECB_EXTERN_C_END |
140 |
#endif |
141 |
|
142 |
/*****************************************************************************/ |
143 |
|
144 |
/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
145 |
/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
146 |
|
147 |
#if ECB_NO_THREADS |
148 |
#define ECB_NO_SMP 1 |
149 |
#endif |
150 |
|
151 |
#if ECB_NO_SMP |
152 |
#define ECB_MEMORY_FENCE do { } while (0) |
153 |
#endif |
154 |
|
155 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
156 |
#if __xlC__ && ECB_CPP |
157 |
#include <builtins.h> |
158 |
#endif |
159 |
|
160 |
#ifndef ECB_MEMORY_FENCE |
161 |
#if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
162 |
#if __i386 || __i386__ |
163 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
164 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
165 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
166 |
#elif ECB_GCC_AMD64 |
167 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
168 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
169 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
170 |
#elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
171 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
172 |
#elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
173 |
|| defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
174 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
175 |
#elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
176 |
|| defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
177 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
178 |
#elif __aarch64__ |
179 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
180 |
#elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
181 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
182 |
#define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
183 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
184 |
#elif defined __s390__ || defined __s390x__ |
185 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
186 |
#elif defined __mips__ |
187 |
/* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
188 |
/* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
189 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
190 |
#elif defined __alpha__ |
191 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
192 |
#elif defined __hppa__ |
193 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
194 |
#define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
195 |
#elif defined __ia64__ |
196 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
197 |
#elif defined __m68k__ |
198 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
199 |
#elif defined __m88k__ |
200 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
201 |
#elif defined __sh__ |
202 |
#define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
203 |
#endif |
204 |
#endif |
205 |
#endif |
206 |
|
207 |
#ifndef ECB_MEMORY_FENCE |
208 |
#if ECB_GCC_VERSION(4,7) |
209 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
210 |
#define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
211 |
#define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
212 |
#define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
213 |
|
214 |
#elif ECB_CLANG_EXTENSION(c_atomic) |
215 |
/* see comment below (stdatomic.h) about the C11 memory model. */ |
216 |
#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
217 |
#define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
218 |
#define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
219 |
|
220 |
#elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
221 |
#define ECB_MEMORY_FENCE __sync_synchronize () |
222 |
#elif _MSC_VER >= 1500 /* VC++ 2008 */ |
223 |
/* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
224 |
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
225 |
#define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
226 |
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
227 |
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
228 |
#elif _MSC_VER >= 1400 /* VC++ 2005 */ |
229 |
#pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
230 |
#define ECB_MEMORY_FENCE _ReadWriteBarrier () |
231 |
#define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
232 |
#define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
233 |
#elif defined _WIN32 |
234 |
#include <WinNT.h> |
235 |
#define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
236 |
#elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
237 |
#include <mbarrier.h> |
238 |
#define ECB_MEMORY_FENCE __machine_rw_barrier () |
239 |
#define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
240 |
#define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
241 |
#elif __xlC__ |
242 |
#define ECB_MEMORY_FENCE __sync () |
243 |
#endif |
244 |
#endif |
245 |
|
246 |
#ifndef ECB_MEMORY_FENCE |
247 |
#if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
248 |
/* we assume that these memory fences work on all variables/all memory accesses, */ |
249 |
/* not just C11 atomics and atomic accesses */ |
250 |
#include <stdatomic.h> |
251 |
/* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
252 |
/* any fence other than seq_cst, which isn't very efficient for us. */ |
253 |
/* Why that is, we don't know - either the C11 memory model is quite useless */ |
254 |
/* for most usages, or gcc and clang have a bug */ |
255 |
/* I *currently* lean towards the latter, and inefficiently implement */ |
256 |
/* all three of ecb's fences as a seq_cst fence */ |
257 |
/* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
258 |
/* for all __atomic_thread_fence's except seq_cst */ |
259 |
#define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
260 |
#endif |
261 |
#endif |
262 |
|
263 |
#ifndef ECB_MEMORY_FENCE |
264 |
#if !ECB_AVOID_PTHREADS |
265 |
/* |
266 |
* if you get undefined symbol references to pthread_mutex_lock, |
267 |
* or failure to find pthread.h, then you should implement |
268 |
* the ECB_MEMORY_FENCE operations for your cpu/compiler |
269 |
* OR provide pthread.h and link against the posix thread library |
270 |
* of your system. |
271 |
*/ |
272 |
#include <pthread.h> |
273 |
#define ECB_NEEDS_PTHREADS 1 |
274 |
#define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
275 |
|
276 |
static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
277 |
#define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
278 |
#endif |
279 |
#endif |
280 |
|
281 |
#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
282 |
#define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
283 |
#endif |
284 |
|
285 |
#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
286 |
#define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
287 |
#endif |
288 |
|
289 |
/*****************************************************************************/ |
290 |
|
291 |
#if ECB_CPP |
292 |
#define ecb_inline static inline |
293 |
#elif ECB_GCC_VERSION(2,5) |
294 |
#define ecb_inline static __inline__ |
295 |
#elif ECB_C99 |
296 |
#define ecb_inline static inline |
297 |
#else |
298 |
#define ecb_inline static |
299 |
#endif |
300 |
|
301 |
#if ECB_GCC_VERSION(3,3) |
302 |
#define ecb_restrict __restrict__ |
303 |
#elif ECB_C99 |
304 |
#define ecb_restrict restrict |
305 |
#else |
306 |
#define ecb_restrict |
307 |
#endif |
308 |
|
309 |
typedef int ecb_bool; |
310 |
|
311 |
#define ECB_CONCAT_(a, b) a ## b |
312 |
#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
313 |
#define ECB_STRINGIFY_(a) # a |
314 |
#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
315 |
#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
316 |
|
317 |
#define ecb_function_ ecb_inline |
318 |
|
319 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
320 |
#define ecb_attribute(attrlist) __attribute__ (attrlist) |
321 |
#else |
322 |
#define ecb_attribute(attrlist) |
323 |
#endif |
324 |
|
325 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
326 |
#define ecb_is_constant(expr) __builtin_constant_p (expr) |
327 |
#else |
328 |
/* possible C11 impl for integral types |
329 |
typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
330 |
#define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
331 |
|
332 |
#define ecb_is_constant(expr) 0 |
333 |
#endif |
334 |
|
335 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
336 |
#define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
337 |
#else |
338 |
#define ecb_expect(expr,value) (expr) |
339 |
#endif |
340 |
|
341 |
#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
342 |
#define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
343 |
#else |
344 |
#define ecb_prefetch(addr,rw,locality) |
345 |
#endif |
346 |
|
347 |
/* no emulation for ecb_decltype */ |
348 |
#if ECB_CPP11 |
349 |
// older implementations might have problems with decltype(x)::type, work around it |
350 |
template<class T> struct ecb_decltype_t { typedef T type; }; |
351 |
#define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
352 |
#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
353 |
#define ecb_decltype(x) __typeof__ (x) |
354 |
#endif |
355 |
|
356 |
#if _MSC_VER >= 1300 |
357 |
#define ecb_deprecated __declspec (deprecated) |
358 |
#else |
359 |
#define ecb_deprecated ecb_attribute ((__deprecated__)) |
360 |
#endif |
361 |
|
362 |
#if _MSC_VER >= 1500 |
363 |
#define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
364 |
#elif ECB_GCC_VERSION(4,5) |
365 |
#define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
366 |
#else |
367 |
#define ecb_deprecated_message(msg) ecb_deprecated |
368 |
#endif |
369 |
|
370 |
#if _MSC_VER >= 1400 |
371 |
#define ecb_noinline __declspec (noinline) |
372 |
#else |
373 |
#define ecb_noinline ecb_attribute ((__noinline__)) |
374 |
#endif |
375 |
|
376 |
#define ecb_unused ecb_attribute ((__unused__)) |
377 |
#define ecb_const ecb_attribute ((__const__)) |
378 |
#define ecb_pure ecb_attribute ((__pure__)) |
379 |
|
380 |
#if ECB_C11 || __IBMC_NORETURN |
381 |
/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
382 |
#define ecb_noreturn _Noreturn |
383 |
#elif ECB_CPP11 |
384 |
#define ecb_noreturn [[noreturn]] |
385 |
#elif _MSC_VER >= 1200 |
386 |
/* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
387 |
#define ecb_noreturn __declspec (noreturn) |
388 |
#else |
389 |
#define ecb_noreturn ecb_attribute ((__noreturn__)) |
390 |
#endif |
391 |
|
392 |
#if ECB_GCC_VERSION(4,3) |
393 |
#define ecb_artificial ecb_attribute ((__artificial__)) |
394 |
#define ecb_hot ecb_attribute ((__hot__)) |
395 |
#define ecb_cold ecb_attribute ((__cold__)) |
396 |
#else |
397 |
#define ecb_artificial |
398 |
#define ecb_hot |
399 |
#define ecb_cold |
400 |
#endif |
401 |
|
402 |
/* put around conditional expressions if you are very sure that the */ |
403 |
/* expression is mostly true or mostly false. note that these return */ |
404 |
/* booleans, not the expression. */ |
405 |
#define ecb_expect_false(expr) ecb_expect (!!(expr), 0) |
406 |
#define ecb_expect_true(expr) ecb_expect (!!(expr), 1) |
407 |
/* for compatibility to the rest of the world */ |
408 |
#define ecb_likely(expr) ecb_expect_true (expr) |
409 |
#define ecb_unlikely(expr) ecb_expect_false (expr) |
410 |
|
411 |
/* count trailing zero bits and count # of one bits */ |
412 |
#if ECB_GCC_VERSION(3,4) \ |
413 |
|| (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
414 |
&& ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
415 |
&& ECB_CLANG_BUILTIN(__builtin_popcount)) |
416 |
/* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
417 |
#define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
418 |
#define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
419 |
#define ecb_ctz32(x) __builtin_ctz (x) |
420 |
#define ecb_ctz64(x) __builtin_ctzll (x) |
421 |
#define ecb_popcount32(x) __builtin_popcount (x) |
422 |
/* no popcountll */ |
423 |
#else |
424 |
ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
425 |
ecb_function_ ecb_const int |
426 |
ecb_ctz32 (uint32_t x) |
427 |
{ |
428 |
int r = 0; |
429 |
|
430 |
x &= ~x + 1; /* this isolates the lowest bit */ |
431 |
|
432 |
#if ECB_branchless_on_i386 |
433 |
r += !!(x & 0xaaaaaaaa) << 0; |
434 |
r += !!(x & 0xcccccccc) << 1; |
435 |
r += !!(x & 0xf0f0f0f0) << 2; |
436 |
r += !!(x & 0xff00ff00) << 3; |
437 |
r += !!(x & 0xffff0000) << 4; |
438 |
#else |
439 |
if (x & 0xaaaaaaaa) r += 1; |
440 |
if (x & 0xcccccccc) r += 2; |
441 |
if (x & 0xf0f0f0f0) r += 4; |
442 |
if (x & 0xff00ff00) r += 8; |
443 |
if (x & 0xffff0000) r += 16; |
444 |
#endif |
445 |
|
446 |
return r; |
447 |
} |
448 |
|
449 |
ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
450 |
ecb_function_ ecb_const int |
451 |
ecb_ctz64 (uint64_t x) |
452 |
{ |
453 |
int shift = x & 0xffffffffU ? 0 : 32; |
454 |
return ecb_ctz32 (x >> shift) + shift; |
455 |
} |
456 |
|
457 |
ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
458 |
ecb_function_ ecb_const int |
459 |
ecb_popcount32 (uint32_t x) |
460 |
{ |
461 |
x -= (x >> 1) & 0x55555555; |
462 |
x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
463 |
x = ((x >> 4) + x) & 0x0f0f0f0f; |
464 |
x *= 0x01010101; |
465 |
|
466 |
return x >> 24; |
467 |
} |
468 |
|
469 |
ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
470 |
ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
471 |
{ |
472 |
int r = 0; |
473 |
|
474 |
if (x >> 16) { x >>= 16; r += 16; } |
475 |
if (x >> 8) { x >>= 8; r += 8; } |
476 |
if (x >> 4) { x >>= 4; r += 4; } |
477 |
if (x >> 2) { x >>= 2; r += 2; } |
478 |
if (x >> 1) { r += 1; } |
479 |
|
480 |
return r; |
481 |
} |
482 |
|
483 |
ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
484 |
ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
485 |
{ |
486 |
int r = 0; |
487 |
|
488 |
if (x >> 32) { x >>= 32; r += 32; } |
489 |
|
490 |
return r + ecb_ld32 (x); |
491 |
} |
492 |
#endif |
493 |
|
494 |
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
495 |
ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
496 |
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
497 |
ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
498 |
|
499 |
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
500 |
ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
501 |
{ |
502 |
return ( (x * 0x0802U & 0x22110U) |
503 |
| (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
504 |
} |
505 |
|
506 |
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
507 |
ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
508 |
{ |
509 |
x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
510 |
x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
511 |
x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
512 |
x = ( x >> 8 ) | ( x << 8); |
513 |
|
514 |
return x; |
515 |
} |
516 |
|
517 |
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
518 |
ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
519 |
{ |
520 |
x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
521 |
x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
522 |
x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
523 |
x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
524 |
x = ( x >> 16 ) | ( x << 16); |
525 |
|
526 |
return x; |
527 |
} |
528 |
|
529 |
/* popcount64 is only available on 64 bit cpus as gcc builtin */ |
530 |
/* so for this version we are lazy */ |
531 |
ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
532 |
ecb_function_ ecb_const int |
533 |
ecb_popcount64 (uint64_t x) |
534 |
{ |
535 |
return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
536 |
} |
537 |
|
538 |
ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
539 |
ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
540 |
ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
541 |
ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
542 |
ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
543 |
ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
544 |
ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
545 |
ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
546 |
|
547 |
ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
548 |
ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
549 |
ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
550 |
ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
551 |
ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
552 |
ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
553 |
ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
554 |
ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
555 |
|
556 |
#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
557 |
#if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
558 |
#define ecb_bswap16(x) __builtin_bswap16 (x) |
559 |
#else |
560 |
#define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
561 |
#endif |
562 |
#define ecb_bswap32(x) __builtin_bswap32 (x) |
563 |
#define ecb_bswap64(x) __builtin_bswap64 (x) |
564 |
#elif _MSC_VER |
565 |
#include <stdlib.h> |
566 |
#define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
567 |
#define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
568 |
#define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
569 |
#else |
570 |
ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
571 |
ecb_function_ ecb_const uint16_t |
572 |
ecb_bswap16 (uint16_t x) |
573 |
{ |
574 |
return ecb_rotl16 (x, 8); |
575 |
} |
576 |
|
577 |
ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
578 |
ecb_function_ ecb_const uint32_t |
579 |
ecb_bswap32 (uint32_t x) |
580 |
{ |
581 |
return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
582 |
} |
583 |
|
584 |
ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
585 |
ecb_function_ ecb_const uint64_t |
586 |
ecb_bswap64 (uint64_t x) |
587 |
{ |
588 |
return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
589 |
} |
590 |
#endif |
591 |
|
592 |
#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
593 |
#define ecb_unreachable() __builtin_unreachable () |
594 |
#else |
595 |
/* this seems to work fine, but gcc always emits a warning for it :/ */ |
596 |
ecb_inline ecb_noreturn void ecb_unreachable (void); |
597 |
ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
598 |
#endif |
599 |
|
600 |
/* try to tell the compiler that some condition is definitely true */ |
601 |
#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
602 |
|
603 |
ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); |
604 |
ecb_inline ecb_const unsigned char |
605 |
ecb_byteorder_helper (void) |
606 |
{ |
607 |
/* the union code still generates code under pressure in gcc, */ |
608 |
/* but less than using pointers, and always seems to */ |
609 |
/* successfully return a constant. */ |
610 |
/* the reason why we have this horrible preprocessor mess */ |
611 |
/* is to avoid it in all cases, at least on common architectures */ |
612 |
/* or when using a recent enough gcc version (>= 4.6) */ |
613 |
#if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
614 |
return 0x44; |
615 |
#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
616 |
return 0x44; |
617 |
#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
618 |
return 0x11; |
619 |
#else |
620 |
union |
621 |
{ |
622 |
uint32_t i; |
623 |
uint8_t c; |
624 |
} u = { 0x11223344 }; |
625 |
return u.c; |
626 |
#endif |
627 |
} |
628 |
|
629 |
ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
630 |
ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
631 |
ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
632 |
ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
633 |
|
634 |
#if ECB_GCC_VERSION(3,0) || ECB_C99 |
635 |
#define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
636 |
#else |
637 |
#define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
638 |
#endif |
639 |
|
640 |
#if ECB_CPP |
641 |
template<typename T> |
642 |
static inline T ecb_div_rd (T val, T div) |
643 |
{ |
644 |
return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
645 |
} |
646 |
template<typename T> |
647 |
static inline T ecb_div_ru (T val, T div) |
648 |
{ |
649 |
return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
650 |
} |
651 |
#else |
652 |
#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
653 |
#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
654 |
#endif |
655 |
|
656 |
#if ecb_cplusplus_does_not_suck |
657 |
/* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
658 |
template<typename T, int N> |
659 |
static inline int ecb_array_length (const T (&arr)[N]) |
660 |
{ |
661 |
return N; |
662 |
} |
663 |
#else |
664 |
#define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
665 |
#endif |
666 |
|
667 |
/*******************************************************************************/ |
668 |
/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
669 |
|
670 |
/* basically, everything uses "ieee pure-endian" floating point numbers */ |
671 |
/* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
672 |
#if 0 \ |
673 |
|| __i386 || __i386__ \ |
674 |
|| ECB_GCC_AMD64 \ |
675 |
|| __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
676 |
|| defined __s390__ || defined __s390x__ \ |
677 |
|| defined __mips__ \ |
678 |
|| defined __alpha__ \ |
679 |
|| defined __hppa__ \ |
680 |
|| defined __ia64__ \ |
681 |
|| defined __m68k__ \ |
682 |
|| defined __m88k__ \ |
683 |
|| defined __sh__ \ |
684 |
|| defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
685 |
|| (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
686 |
|| defined __aarch64__ |
687 |
#define ECB_STDFP 1 |
688 |
#include <string.h> /* for memcpy */ |
689 |
#else |
690 |
#define ECB_STDFP 0 |
691 |
#endif |
692 |
|
693 |
#ifndef ECB_NO_LIBM |
694 |
|
695 |
#include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
696 |
|
697 |
/* only the oldest of old doesn't have this one. solaris. */ |
698 |
#ifdef INFINITY |
699 |
#define ECB_INFINITY INFINITY |
700 |
#else |
701 |
#define ECB_INFINITY HUGE_VAL |
702 |
#endif |
703 |
|
704 |
#ifdef NAN |
705 |
#define ECB_NAN NAN |
706 |
#else |
707 |
#define ECB_NAN ECB_INFINITY |
708 |
#endif |
709 |
|
710 |
#if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
711 |
#define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
712 |
#define ecb_frexpf(x,e) frexpf ((x), (e)) |
713 |
#else |
714 |
#define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
715 |
#define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
716 |
#endif |
717 |
|
718 |
/* converts an ieee half/binary16 to a float */ |
719 |
ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
720 |
ecb_function_ ecb_const float |
721 |
ecb_binary16_to_float (uint16_t x) |
722 |
{ |
723 |
int e = (x >> 10) & 0x1f; |
724 |
int m = x & 0x3ff; |
725 |
float r; |
726 |
|
727 |
if (!e ) r = ecb_ldexpf (m , -24); |
728 |
else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25); |
729 |
else if (m ) r = ECB_NAN; |
730 |
else r = ECB_INFINITY; |
731 |
|
732 |
return x & 0x8000 ? -r : r; |
733 |
} |
734 |
|
735 |
/* convert a float to ieee single/binary32 */ |
736 |
ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
737 |
ecb_function_ ecb_const uint32_t |
738 |
ecb_float_to_binary32 (float x) |
739 |
{ |
740 |
uint32_t r; |
741 |
|
742 |
#if ECB_STDFP |
743 |
memcpy (&r, &x, 4); |
744 |
#else |
745 |
/* slow emulation, works for anything but -0 */ |
746 |
uint32_t m; |
747 |
int e; |
748 |
|
749 |
if (x == 0e0f ) return 0x00000000U; |
750 |
if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
751 |
if (x < -3.40282346638528860e+38f) return 0xff800000U; |
752 |
if (x != x ) return 0x7fbfffffU; |
753 |
|
754 |
m = ecb_frexpf (x, &e) * 0x1000000U; |
755 |
|
756 |
r = m & 0x80000000U; |
757 |
|
758 |
if (r) |
759 |
m = -m; |
760 |
|
761 |
if (e <= -126) |
762 |
{ |
763 |
m &= 0xffffffU; |
764 |
m >>= (-125 - e); |
765 |
e = -126; |
766 |
} |
767 |
|
768 |
r |= (e + 126) << 23; |
769 |
r |= m & 0x7fffffU; |
770 |
#endif |
771 |
|
772 |
return r; |
773 |
} |
774 |
|
775 |
/* converts an ieee single/binary32 to a float */ |
776 |
ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
777 |
ecb_function_ ecb_const float |
778 |
ecb_binary32_to_float (uint32_t x) |
779 |
{ |
780 |
float r; |
781 |
|
782 |
#if ECB_STDFP |
783 |
memcpy (&r, &x, 4); |
784 |
#else |
785 |
/* emulation, only works for normals and subnormals and +0 */ |
786 |
int neg = x >> 31; |
787 |
int e = (x >> 23) & 0xffU; |
788 |
|
789 |
x &= 0x7fffffU; |
790 |
|
791 |
if (e) |
792 |
x |= 0x800000U; |
793 |
else |
794 |
e = 1; |
795 |
|
796 |
/* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
797 |
r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
798 |
|
799 |
r = neg ? -r : r; |
800 |
#endif |
801 |
|
802 |
return r; |
803 |
} |
804 |
|
805 |
/* convert a double to ieee double/binary64 */ |
806 |
ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
807 |
ecb_function_ ecb_const uint64_t |
808 |
ecb_double_to_binary64 (double x) |
809 |
{ |
810 |
uint64_t r; |
811 |
|
812 |
#if ECB_STDFP |
813 |
memcpy (&r, &x, 8); |
814 |
#else |
815 |
/* slow emulation, works for anything but -0 */ |
816 |
uint64_t m; |
817 |
int e; |
818 |
|
819 |
if (x == 0e0 ) return 0x0000000000000000U; |
820 |
if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
821 |
if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
822 |
if (x != x ) return 0X7ff7ffffffffffffU; |
823 |
|
824 |
m = frexp (x, &e) * 0x20000000000000U; |
825 |
|
826 |
r = m & 0x8000000000000000;; |
827 |
|
828 |
if (r) |
829 |
m = -m; |
830 |
|
831 |
if (e <= -1022) |
832 |
{ |
833 |
m &= 0x1fffffffffffffU; |
834 |
m >>= (-1021 - e); |
835 |
e = -1022; |
836 |
} |
837 |
|
838 |
r |= ((uint64_t)(e + 1022)) << 52; |
839 |
r |= m & 0xfffffffffffffU; |
840 |
#endif |
841 |
|
842 |
return r; |
843 |
} |
844 |
|
845 |
/* converts an ieee double/binary64 to a double */ |
846 |
ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
847 |
ecb_function_ ecb_const double |
848 |
ecb_binary64_to_double (uint64_t x) |
849 |
{ |
850 |
double r; |
851 |
|
852 |
#if ECB_STDFP |
853 |
memcpy (&r, &x, 8); |
854 |
#else |
855 |
/* emulation, only works for normals and subnormals and +0 */ |
856 |
int neg = x >> 63; |
857 |
int e = (x >> 52) & 0x7ffU; |
858 |
|
859 |
x &= 0xfffffffffffffU; |
860 |
|
861 |
if (e) |
862 |
x |= 0x10000000000000U; |
863 |
else |
864 |
e = 1; |
865 |
|
866 |
/* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
867 |
r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
868 |
|
869 |
r = neg ? -r : r; |
870 |
#endif |
871 |
|
872 |
return r; |
873 |
} |
874 |
|
875 |
#endif |
876 |
|
877 |
#endif |
878 |
|