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

Comparing libecb/ecb.h (file contents):
Revision 1.67 by sf-exg, Wed Aug 24 23:28:47 2011 UTC vs.
Revision 1.189 by root, Mon Jun 21 21:26:48 2021 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2011 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#ifdef _WIN32 44/* 16 bits major, 16 bits minor */
45#define ECB_VERSION 0x00010009
46
47#include <string.h> /* for memcpy */
48
49#if defined (_WIN32) && !defined (__MINGW32__)
34 typedef signed char int8_t; 50 typedef signed char int8_t;
35 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;
36 typedef signed short int16_t; 54 typedef signed short int16_t;
37 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;
38 typedef signed int int32_t; 58 typedef signed int int32_t;
39 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;
40 #if __GNUC__ 62 #if __GNUC__
41 typedef signed long long int64_t; 63 typedef signed long long int64_t;
42 typedef unsigned long long uint64_t; 64 typedef unsigned long long uint64_t;
43 #else /* _MSC_VER || __BORLANDC__ */ 65 #else /* _MSC_VER || __BORLANDC__ */
44 typedef signed __int64 int64_t; 66 typedef signed __int64 int64_t;
45 typedef unsigned __int64 uint64_t; 67 typedef unsigned __int64 uint64_t;
46 #endif 68 #endif
69 typedef int64_t int_fast64_t;
70 typedef uint64_t uint_fast64_t;
71 #ifdef _WIN64
72 #define ECB_PTRSIZE 8
73 typedef uint64_t uintptr_t;
74 typedef int64_t intptr_t;
75 #else
76 #define ECB_PTRSIZE 4
77 typedef uint32_t uintptr_t;
78 typedef int32_t intptr_t;
79 #endif
47#else 80#else
48 #include <inttypes.h> 81 #include <inttypes.h>
82 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
83 #define ECB_PTRSIZE 8
84 #else
85 #define ECB_PTRSIZE 4
86 #endif
87#endif
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
100/* work around x32 idiocy by defining proper macros */
101#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
102 #if _ILP32
103 #define ECB_AMD64_X32 1
104 #else
105 #define ECB_AMD64 1
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
49#endif 113#endif
50 114
51/* many compilers define _GNUC_ to some versions but then only implement 115/* many compilers define _GNUC_ to some versions but then only implement
52 * what their idiot authors think are the "more important" extensions, 116 * what their idiot authors think are the "more important" extensions,
53 * causing enormous grief in return for some better fake benchmark numbers. 117 * causing enormous grief in return for some better fake benchmark numbers.
54 * or so. 118 * or so.
55 * 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
56 * 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.
57 */ 121 */
58#ifndef ECB_GCC_VERSION
59 #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__
60 #define ECB_GCC_VERSION(major,minor) 0 123 #define ECB_GCC_VERSION(major,minor) 0
61 #else 124#else
62 #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)))
63 #endif 126#endif
127
128#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
129
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
142#define ECB_CPP (__cplusplus+0)
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)
158
159#if ECB_CPP
160 #define ECB_EXTERN_C extern "C"
161 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
162 #define ECB_EXTERN_C_END }
163#else
164 #define ECB_EXTERN_C extern
165 #define ECB_EXTERN_C_BEG
166 #define ECB_EXTERN_C_END
64#endif 167#endif
65 168
66/*****************************************************************************/ 169/*****************************************************************************/
67 170
68/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ 171/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
69/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ 172/* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
70 173
71#if ECB_NO_THREADS || ECB_NO_SMP 174#if ECB_NO_THREADS
175 #define ECB_NO_SMP 1
176#endif
177
178#if ECB_NO_SMP
72 #define ECB_MEMORY_FENCE do { } while (0) 179 #define ECB_MEMORY_FENCE do { } while (0)
73#endif 180#endif
74 181
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
75#ifndef ECB_MEMORY_FENCE 191#ifndef ECB_MEMORY_FENCE
76 #if ECB_GCC_VERSION(2,5) 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")
77 #if __i386__ 194 #if __i386 || __i386__
78 #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")
79 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ 196 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
80 #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ 197 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
81 #elif __amd64 198 #elif ECB_GCC_AMD64
82 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 199 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
83 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") 200 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
84 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ 201 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
85 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 202 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
86 #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 */
87 #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ 211 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
88 || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) 212 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213 || defined __ARM_ARCH_6T2__
89 #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")
90 #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ 215 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
91 || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) 216 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
92 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
218 #elif __aarch64__
219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
220 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
221 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
222 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
223 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
224 #elif defined __s390__ || defined __s390x__
225 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
226 #elif defined __mips__
227 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
228 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
229 #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
230 #elif defined __alpha__
231 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
232 #elif defined __hppa__
233 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
234 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
235 #elif defined __ia64__
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")
93 #endif 243 #endif
94 #endif 244 #endif
95#endif 245#endif
96 246
97#ifndef ECB_MEMORY_FENCE 247#ifndef ECB_MEMORY_FENCE
248 #if ECB_GCC_VERSION(4,7)
249 /* see comment below (stdatomic.h) about the C11 memory model. */
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 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
254
255 #elif ECB_CLANG_EXTENSION(c_atomic)
256 /* see comment below (stdatomic.h) about the C11 memory model. */
257 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
258 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
259 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
260 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
261
98 #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) 262 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
99 #define ECB_MEMORY_FENCE __sync_synchronize () 263 #define ECB_MEMORY_FENCE __sync_synchronize ()
100 /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ 264 #elif _MSC_VER >= 1500 /* VC++ 2008 */
101 /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ 265 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
266 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
267 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
268 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
269 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
102 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 270 #elif _MSC_VER >= 1400 /* VC++ 2005 */
103 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 271 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
104 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 272 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
105 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 273 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
106 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 274 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
107 #elif defined(_WIN32) 275 #elif defined _WIN32
108 #include <WinNT.h> 276 #include <WinNT.h>
109 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 277 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
278 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
279 #include <mbarrier.h>
280 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
281 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
282 #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
283 #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
284 #elif __xlC__
285 #define ECB_MEMORY_FENCE __sync ()
286 #endif
287#endif
288
289#ifndef ECB_MEMORY_FENCE
290 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
291 /* we assume that these memory fences work on all variables/all memory accesses, */
292 /* not just C11 atomics and atomic accesses */
293 #include <stdatomic.h>
294 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
295 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
296 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
110 #endif 297 #endif
111#endif 298#endif
112 299
113#ifndef ECB_MEMORY_FENCE 300#ifndef ECB_MEMORY_FENCE
114 #if !ECB_AVOID_PTHREADS 301 #if !ECB_AVOID_PTHREADS
126 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; 313 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
127 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) 314 #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
128 #endif 315 #endif
129#endif 316#endif
130 317
131#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) 318#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
132 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE 319 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
133#endif 320#endif
134 321
135#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) 322#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
136 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 323 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
137#endif 324#endif
138 325
326#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
327 #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
328#endif
329
139/*****************************************************************************/ 330/*****************************************************************************/
140 331
141#define ECB_C99 (__STDC_VERSION__ >= 199901L) 332#if ECB_CPP
142
143#if __cplusplus
144 #define ecb_inline static inline 333 #define ecb_inline static inline
145#elif ECB_GCC_VERSION(2,5) 334#elif ECB_GCC_VERSION(2,5)
146 #define ecb_inline static __inline__ 335 #define ecb_inline static __inline__
147#elif ECB_C99 336#elif ECB_C99
148 #define ecb_inline static inline 337 #define ecb_inline static inline
162 351
163#define ECB_CONCAT_(a, b) a ## b 352#define ECB_CONCAT_(a, b) a ## b
164#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 353#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
165#define ECB_STRINGIFY_(a) # a 354#define ECB_STRINGIFY_(a) # a
166#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 355#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
356#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
167 357
168#define ecb_function_ ecb_inline 358#define ecb_function_ ecb_inline
169 359
170#if ECB_GCC_VERSION(3,1) 360#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
171 #define ecb_attribute(attrlist) __attribute__(attrlist) 361 #define ecb_attribute(attrlist) __attribute__ (attrlist)
362#else
363 #define ecb_attribute(attrlist)
364#endif
365
366#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
172 #define ecb_is_constant(expr) __builtin_constant_p (expr) 367 #define ecb_is_constant(expr) __builtin_constant_p (expr)
368#else
369 /* possible C11 impl for integral types
370 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
371 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
372
373 #define ecb_is_constant(expr) 0
374#endif
375
376#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
173 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 377 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
378#else
379 #define ecb_expect(expr,value) (expr)
380#endif
381
382#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
174 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 383 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
175#else 384#else
176 #define ecb_attribute(attrlist)
177 #define ecb_is_constant(expr) 0
178 #define ecb_expect(expr,value) (expr)
179 #define ecb_prefetch(addr,rw,locality) 385 #define ecb_prefetch(addr,rw,locality)
180#endif 386#endif
181 387
182/* no emulation for ecb_decltype */ 388/* no emulation for ecb_decltype */
183#if ECB_GCC_VERSION(4,5) 389#if ECB_CPP11
390 // older implementations might have problems with decltype(x)::type, work around it
391 template<class T> struct ecb_decltype_t { typedef T type; };
184 #define ecb_decltype(x) __decltype(x) 392 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
185#elif ECB_GCC_VERSION(3,0) 393#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
186 #define ecb_decltype(x) __typeof(x) 394 #define ecb_decltype(x) __typeof__ (x)
187#endif 395#endif
188 396
397#if _MSC_VER >= 1300
398 #define ecb_deprecated __declspec (deprecated)
399#else
400 #define ecb_deprecated ecb_attribute ((__deprecated__))
401#endif
402
403#if _MSC_VER >= 1500
404 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
405#elif ECB_GCC_VERSION(4,5)
406 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
407#else
408 #define ecb_deprecated_message(msg) ecb_deprecated
409#endif
410
411#if _MSC_VER >= 1400
412 #define ecb_noinline __declspec (noinline)
413#else
189#define ecb_noinline ecb_attribute ((__noinline__)) 414 #define ecb_noinline ecb_attribute ((__noinline__))
190#define ecb_noreturn ecb_attribute ((__noreturn__)) 415#endif
416
191#define ecb_unused ecb_attribute ((__unused__)) 417#define ecb_unused ecb_attribute ((__unused__))
192#define ecb_const ecb_attribute ((__const__)) 418#define ecb_const ecb_attribute ((__const__))
193#define ecb_pure ecb_attribute ((__pure__)) 419#define ecb_pure ecb_attribute ((__pure__))
420
421#if ECB_C11 || __IBMC_NORETURN
422 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
423 #define ecb_noreturn _Noreturn
424#elif ECB_CPP11
425 #define ecb_noreturn [[noreturn]]
426#elif _MSC_VER >= 1200
427 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
428 #define ecb_noreturn __declspec (noreturn)
429#else
430 #define ecb_noreturn ecb_attribute ((__noreturn__))
431#endif
194 432
195#if ECB_GCC_VERSION(4,3) 433#if ECB_GCC_VERSION(4,3)
196 #define ecb_artificial ecb_attribute ((__artificial__)) 434 #define ecb_artificial ecb_attribute ((__artificial__))
197 #define ecb_hot ecb_attribute ((__hot__)) 435 #define ecb_hot ecb_attribute ((__hot__))
198 #define ecb_cold ecb_attribute ((__cold__)) 436 #define ecb_cold ecb_attribute ((__cold__))
210/* for compatibility to the rest of the world */ 448/* for compatibility to the rest of the world */
211#define ecb_likely(expr) ecb_expect_true (expr) 449#define ecb_likely(expr) ecb_expect_true (expr)
212#define ecb_unlikely(expr) ecb_expect_false (expr) 450#define ecb_unlikely(expr) ecb_expect_false (expr)
213 451
214/* count trailing zero bits and count # of one bits */ 452/* count trailing zero bits and count # of one bits */
215#if ECB_GCC_VERSION(3,4) 453#if ECB_GCC_VERSION(3,4) \
454 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
455 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
456 && ECB_CLANG_BUILTIN(__builtin_popcount))
216 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ 457 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
217 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 458 #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
218 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 459 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
219 #define ecb_ctz32(x) __builtin_ctz (x) 460 #define ecb_ctz32(x) __builtin_ctz (x)
220 #define ecb_ctz64(x) __builtin_ctzll (x) 461 #define ecb_ctz64(x) __builtin_ctzll (x)
221 #define ecb_popcount32(x) __builtin_popcount (x) 462 #define ecb_popcount32(x) __builtin_popcount (x)
222 /* no popcountll */ 463 /* no popcountll */
223#else 464#else
224 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 465 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
225 ecb_function_ int 466 ecb_function_ ecb_const int
226 ecb_ctz32 (uint32_t x) 467 ecb_ctz32 (uint32_t x)
227 { 468 {
469#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
470 unsigned long r;
471 _BitScanForward (&r, x);
472 return (int)r;
473#else
228 int r = 0; 474 int r = 0;
229 475
230 x &= ~x + 1; /* this isolates the lowest bit */ 476 x &= ~x + 1; /* this isolates the lowest bit */
231 477
232#if ECB_branchless_on_i386 478#if ECB_branchless_on_i386
242 if (x & 0xff00ff00) r += 8; 488 if (x & 0xff00ff00) r += 8;
243 if (x & 0xffff0000) r += 16; 489 if (x & 0xffff0000) r += 16;
244#endif 490#endif
245 491
246 return r; 492 return r;
493#endif
247 } 494 }
248 495
249 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 496 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
250 ecb_function_ int 497 ecb_function_ ecb_const int
251 ecb_ctz64 (uint64_t x) 498 ecb_ctz64 (uint64_t x)
252 { 499 {
500#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
501 unsigned long r;
502 _BitScanForward64 (&r, x);
503 return (int)r;
504#else
253 int shift = x & 0xffffffffU ? 0 : 32; 505 int shift = x & 0xffffffff ? 0 : 32;
254 return ecb_ctz32 (x >> shift) + shift; 506 return ecb_ctz32 (x >> shift) + shift;
507#endif
255 } 508 }
256 509
257 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 510 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
258 ecb_function_ int 511 ecb_function_ ecb_const int
259 ecb_popcount32 (uint32_t x) 512 ecb_popcount32 (uint32_t x)
260 { 513 {
261 x -= (x >> 1) & 0x55555555; 514 x -= (x >> 1) & 0x55555555;
262 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 515 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
263 x = ((x >> 4) + x) & 0x0f0f0f0f; 516 x = ((x >> 4) + x) & 0x0f0f0f0f;
264 x *= 0x01010101; 517 x *= 0x01010101;
265 518
266 return x >> 24; 519 return x >> 24;
267 } 520 }
268 521
269 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 522 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
270 ecb_function_ int ecb_ld32 (uint32_t x) 523 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
271 { 524 {
525#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
526 unsigned long r;
527 _BitScanReverse (&r, x);
528 return (int)r;
529#else
272 int r = 0; 530 int r = 0;
273 531
274 if (x >> 16) { x >>= 16; r += 16; } 532 if (x >> 16) { x >>= 16; r += 16; }
275 if (x >> 8) { x >>= 8; r += 8; } 533 if (x >> 8) { x >>= 8; r += 8; }
276 if (x >> 4) { x >>= 4; r += 4; } 534 if (x >> 4) { x >>= 4; r += 4; }
277 if (x >> 2) { x >>= 2; r += 2; } 535 if (x >> 2) { x >>= 2; r += 2; }
278 if (x >> 1) { r += 1; } 536 if (x >> 1) { r += 1; }
279 537
280 return r; 538 return r;
539#endif
281 } 540 }
282 541
283 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 542 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
284 ecb_function_ int ecb_ld64 (uint64_t x) 543 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
285 { 544 {
545#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
546 unsigned long r;
547 _BitScanReverse64 (&r, x);
548 return (int)r;
549#else
286 int r = 0; 550 int r = 0;
287 551
288 if (x >> 32) { x >>= 32; r += 32; } 552 if (x >> 32) { x >>= 32; r += 32; }
289 553
290 return r + ecb_ld32 (x); 554 return r + ecb_ld32 (x);
291 }
292#endif 555#endif
556 }
557#endif
558
559ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
560ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
561ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
562ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
563
564ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
565ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
566{
567 return ( (x * 0x0802U & 0x22110U)
568 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
569}
570
571ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
572ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
573{
574 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
575 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
576 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
577 x = ( x >> 8 ) | ( x << 8);
578
579 return x;
580}
581
582ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
583ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
584{
585 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
586 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
587 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
588 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
589 x = ( x >> 16 ) | ( x << 16);
590
591 return x;
592}
293 593
294/* popcount64 is only available on 64 bit cpus as gcc builtin */ 594/* popcount64 is only available on 64 bit cpus as gcc builtin */
295/* so for this version we are lazy */ 595/* so for this version we are lazy */
296ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 596ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
297ecb_function_ int 597ecb_function_ ecb_const int
298ecb_popcount64 (uint64_t x) 598ecb_popcount64 (uint64_t x)
299{ 599{
300 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 600 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
301} 601}
302 602
303ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 603ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
304ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 604ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
305ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 605ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
306ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 606ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
307ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 607ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
308ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 608ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
309ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 609ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
310ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 610ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
311 611
312ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 612ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
313ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 613ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
314ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 614ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
315ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 615ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
316ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 616ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
317ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 617ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
318ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 618ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
319ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 619ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
320 620
321#if ECB_GCC_VERSION(4,3) 621#if ECB_CPP
622
623inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
624inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
625inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
626inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
627
628inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
629inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
630inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
631inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
632
633inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
634inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
635inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
636inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
637
638inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
639inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
640inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
641inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
642
643inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
644inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
645inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
646
647inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
648inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
649inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
650inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
651
652inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
653inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
654inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
655inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
656
657#endif
658
659#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
660 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
661 #define ecb_bswap16(x) __builtin_bswap16 (x)
662 #else
322 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 663 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
664 #endif
323 #define ecb_bswap32(x) __builtin_bswap32 (x) 665 #define ecb_bswap32(x) __builtin_bswap32 (x)
324 #define ecb_bswap64(x) __builtin_bswap64 (x) 666 #define ecb_bswap64(x) __builtin_bswap64 (x)
667#elif _MSC_VER
668 #include <stdlib.h>
669 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
670 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
671 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
325#else 672#else
326 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 673 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
327 ecb_function_ uint16_t 674 ecb_function_ ecb_const uint16_t
328 ecb_bswap16 (uint16_t x) 675 ecb_bswap16 (uint16_t x)
329 { 676 {
330 return ecb_rotl16 (x, 8); 677 return ecb_rotl16 (x, 8);
331 } 678 }
332 679
333 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 680 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
334 ecb_function_ uint32_t 681 ecb_function_ ecb_const uint32_t
335 ecb_bswap32 (uint32_t x) 682 ecb_bswap32 (uint32_t x)
336 { 683 {
337 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 684 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
338 } 685 }
339 686
340 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 687 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
341 ecb_function_ uint64_t 688 ecb_function_ ecb_const uint64_t
342 ecb_bswap64 (uint64_t x) 689 ecb_bswap64 (uint64_t x)
343 { 690 {
344 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 691 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
345 } 692 }
346#endif 693#endif
347 694
348#if ECB_GCC_VERSION(4,5) 695#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
349 #define ecb_unreachable() __builtin_unreachable () 696 #define ecb_unreachable() __builtin_unreachable ()
350#else 697#else
351 /* this seems to work fine, but gcc always emits a warning for it :/ */ 698 /* this seems to work fine, but gcc always emits a warning for it :/ */
352 ecb_function_ void ecb_unreachable (void) ecb_noreturn; 699 ecb_inline ecb_noreturn void ecb_unreachable (void);
353 ecb_function_ void ecb_unreachable (void) { } 700 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
354#endif 701#endif
355 702
356/* try to tell the compiler that some condition is definitely true */ 703/* try to tell the compiler that some condition is definitely true */
357#define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) 704#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
358 705
359ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; 706ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
360ecb_function_ unsigned char 707ecb_inline ecb_const uint32_t
361ecb_byteorder_helper (void) 708ecb_byteorder_helper (void)
362{ 709{
363 const uint32_t u = 0x11223344; 710 /* the union code still generates code under pressure in gcc, */
364 return *(unsigned char *)&u; 711 /* but less than using pointers, and always seems to */
712 /* successfully return a constant. */
713 /* the reason why we have this horrible preprocessor mess */
714 /* is to avoid it in all cases, at least on common architectures */
715 /* or when using a recent enough gcc version (>= 4.6) */
716#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
717 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
718 #define ECB_LITTLE_ENDIAN 1
719 return 0x44332211;
720#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
721 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
722 #define ECB_BIG_ENDIAN 1
723 return 0x11223344;
724#else
725 union
726 {
727 uint8_t c[4];
728 uint32_t u;
729 } u = { 0x11, 0x22, 0x33, 0x44 };
730 return u.u;
731#endif
365} 732}
366 733
367ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; 734ecb_inline ecb_const ecb_bool ecb_big_endian (void);
368ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 735ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
369ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; 736ecb_inline ecb_const ecb_bool ecb_little_endian (void);
370ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 737ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
738
739/*****************************************************************************/
740/* unaligned load/store */
741
742ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
743ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
744ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
745
746ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
747ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
748ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
749
750ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
751ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
752ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753
754ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
755ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
756ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
757
758ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
759ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
760ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
761
762ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
763ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
764ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
765
766ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
767ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
768ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
769
770ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
771ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
772ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
773
774ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
775ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
776ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
777
778ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
779ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
780ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
781
782#if ECB_CPP
783
784inline uint8_t ecb_bswap (uint8_t v) { return v; }
785inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
786inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
787inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
788
789template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
790template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
791template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
792template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
793template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
794template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
795template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
796template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
797
798template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
799template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
800template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
801template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
802template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
803template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
804template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
805template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
806
807#endif
808
809/*****************************************************************************/
810/* division */
371 811
372#if ECB_GCC_VERSION(3,0) || ECB_C99 812#if ECB_GCC_VERSION(3,0) || ECB_C99
813 /* C99 tightened the definition of %, so we can use a more efficient version */
373 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 814 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
374#else 815#else
375 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 816 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
376#endif 817#endif
377 818
819#if ECB_CPP
820 template<typename T>
821 static inline T ecb_div_rd (T val, T div)
822 {
823 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
824 }
825 template<typename T>
826 static inline T ecb_div_ru (T val, T div)
827 {
828 return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
829 }
830#else
378#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) 831 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
379#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) 832 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
833#endif
834
835/*****************************************************************************/
836/* array length */
380 837
381#if ecb_cplusplus_does_not_suck 838#if ecb_cplusplus_does_not_suck
382 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 839 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
383 template<typename T, int N> 840 template<typename T, int N>
384 static inline int ecb_array_length (const T (&arr)[N]) 841 static inline int ecb_array_length (const T (&arr)[N])
387 } 844 }
388#else 845#else
389 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 846 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
390#endif 847#endif
391 848
849/*****************************************************************************/
850/* IEEE 754-2008 half float conversions */
851
852ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
853ecb_function_ ecb_const uint32_t
854ecb_binary16_to_binary32 (uint32_t x)
855{
856 unsigned int s = (x & 0x8000) << (31 - 15);
857 int e = (x >> 10) & 0x001f;
858 unsigned int m = x & 0x03ff;
859
860 if (ecb_expect_false (e == 31))
861 /* infinity or NaN */
862 e = 255 - (127 - 15);
863 else if (ecb_expect_false (!e))
864 {
865 if (ecb_expect_true (!m))
866 /* zero, handled by code below by forcing e to 0 */
867 e = 0 - (127 - 15);
868 else
869 {
870 /* subnormal, renormalise */
871 unsigned int s = 10 - ecb_ld32 (m);
872
873 m = (m << s) & 0x3ff; /* mask implicit bit */
874 e -= s - 1;
875 }
876 }
877
878 /* e and m now are normalised, or zero, (or inf or nan) */
879 e += 127 - 15;
880
881 return s | (e << 23) | (m << (23 - 10));
882}
883
884ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
885ecb_function_ ecb_const uint16_t
886ecb_binary32_to_binary16 (uint32_t x)
887{
888 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
889 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
890 unsigned int m = x & 0x007fffff;
891
892 x &= 0x7fffffff;
893
894 /* if it's within range of binary16 normals, use fast path */
895 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
896 {
897 /* mantissa round-to-even */
898 m += 0x00000fff + ((m >> (23 - 10)) & 1);
899
900 /* handle overflow */
901 if (ecb_expect_false (m >= 0x00800000))
902 {
903 m >>= 1;
904 e += 1;
905 }
906
907 return s | (e << 10) | (m >> (23 - 10));
908 }
909
910 /* handle large numbers and infinity */
911 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
912 return s | 0x7c00;
913
914 /* handle zero, subnormals and small numbers */
915 if (ecb_expect_true (x < 0x38800000))
916 {
917 /* zero */
918 if (ecb_expect_true (!x))
919 return s;
920
921 /* handle subnormals */
922
923 /* too small, will be zero */
924 if (e < (14 - 24)) /* might not be sharp, but is good enough */
925 return s;
926
927 m |= 0x00800000; /* make implicit bit explicit */
928
929 /* very tricky - we need to round to the nearest e (+10) bit value */
930 {
931 unsigned int bits = 14 - e;
932 unsigned int half = (1 << (bits - 1)) - 1;
933 unsigned int even = (m >> bits) & 1;
934
935 /* if this overflows, we will end up with a normalised number */
936 m = (m + half + even) >> bits;
937 }
938
939 return s | m;
940 }
941
942 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
943 m >>= 13;
944
945 return s | 0x7c00 | m | !m;
946}
947
948/*******************************************************************************/
949/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
950
951/* basically, everything uses "ieee pure-endian" floating point numbers */
952/* the only noteworthy exception is ancient armle, which uses order 43218765 */
953#if 0 \
954 || __i386 || __i386__ \
955 || ECB_GCC_AMD64 \
956 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
957 || defined __s390__ || defined __s390x__ \
958 || defined __mips__ \
959 || defined __alpha__ \
960 || defined __hppa__ \
961 || defined __ia64__ \
962 || defined __m68k__ \
963 || defined __m88k__ \
964 || defined __sh__ \
965 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
966 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
967 || defined __aarch64__
968 #define ECB_STDFP 1
969#else
970 #define ECB_STDFP 0
971#endif
972
973#ifndef ECB_NO_LIBM
974
975 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
976
977 /* only the oldest of old doesn't have this one. solaris. */
978 #ifdef INFINITY
979 #define ECB_INFINITY INFINITY
980 #else
981 #define ECB_INFINITY HUGE_VAL
392#endif 982 #endif
393 983
984 #ifdef NAN
985 #define ECB_NAN NAN
986 #else
987 #define ECB_NAN ECB_INFINITY
988 #endif
989
990 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
991 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
992 #define ecb_frexpf(x,e) frexpf ((x), (e))
993 #else
994 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
995 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
996 #endif
997
998 /* convert a float to ieee single/binary32 */
999 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
1000 ecb_function_ ecb_const uint32_t
1001 ecb_float_to_binary32 (float x)
1002 {
1003 uint32_t r;
1004
1005 #if ECB_STDFP
1006 memcpy (&r, &x, 4);
1007 #else
1008 /* slow emulation, works for anything but -0 */
1009 uint32_t m;
1010 int e;
1011
1012 if (x == 0e0f ) return 0x00000000U;
1013 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
1014 if (x < -3.40282346638528860e+38f) return 0xff800000U;
1015 if (x != x ) return 0x7fbfffffU;
1016
1017 m = ecb_frexpf (x, &e) * 0x1000000U;
1018
1019 r = m & 0x80000000U;
1020
1021 if (r)
1022 m = -m;
1023
1024 if (e <= -126)
1025 {
1026 m &= 0xffffffU;
1027 m >>= (-125 - e);
1028 e = -126;
1029 }
1030
1031 r |= (e + 126) << 23;
1032 r |= m & 0x7fffffU;
1033 #endif
1034
1035 return r;
1036 }
1037
1038 /* converts an ieee single/binary32 to a float */
1039 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
1040 ecb_function_ ecb_const float
1041 ecb_binary32_to_float (uint32_t x)
1042 {
1043 float r;
1044
1045 #if ECB_STDFP
1046 memcpy (&r, &x, 4);
1047 #else
1048 /* emulation, only works for normals and subnormals and +0 */
1049 int neg = x >> 31;
1050 int e = (x >> 23) & 0xffU;
1051
1052 x &= 0x7fffffU;
1053
1054 if (e)
1055 x |= 0x800000U;
1056 else
1057 e = 1;
1058
1059 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
1060 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
1061
1062 r = neg ? -r : r;
1063 #endif
1064
1065 return r;
1066 }
1067
1068 /* convert a double to ieee double/binary64 */
1069 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
1070 ecb_function_ ecb_const uint64_t
1071 ecb_double_to_binary64 (double x)
1072 {
1073 uint64_t r;
1074
1075 #if ECB_STDFP
1076 memcpy (&r, &x, 8);
1077 #else
1078 /* slow emulation, works for anything but -0 */
1079 uint64_t m;
1080 int e;
1081
1082 if (x == 0e0 ) return 0x0000000000000000U;
1083 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
1084 if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
1085 if (x != x ) return 0X7ff7ffffffffffffU;
1086
1087 m = frexp (x, &e) * 0x20000000000000U;
1088
1089 r = m & 0x8000000000000000;;
1090
1091 if (r)
1092 m = -m;
1093
1094 if (e <= -1022)
1095 {
1096 m &= 0x1fffffffffffffU;
1097 m >>= (-1021 - e);
1098 e = -1022;
1099 }
1100
1101 r |= ((uint64_t)(e + 1022)) << 52;
1102 r |= m & 0xfffffffffffffU;
1103 #endif
1104
1105 return r;
1106 }
1107
1108 /* converts an ieee double/binary64 to a double */
1109 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
1110 ecb_function_ ecb_const double
1111 ecb_binary64_to_double (uint64_t x)
1112 {
1113 double r;
1114
1115 #if ECB_STDFP
1116 memcpy (&r, &x, 8);
1117 #else
1118 /* emulation, only works for normals and subnormals and +0 */
1119 int neg = x >> 63;
1120 int e = (x >> 52) & 0x7ffU;
1121
1122 x &= 0xfffffffffffffU;
1123
1124 if (e)
1125 x |= 0x10000000000000U;
1126 else
1127 e = 1;
1128
1129 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
1130 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
1131
1132 r = neg ? -r : r;
1133 #endif
1134
1135 return r;
1136 }
1137
1138 /* convert a float to ieee half/binary16 */
1139 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1140 ecb_function_ ecb_const uint16_t
1141 ecb_float_to_binary16 (float x)
1142 {
1143 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1144 }
1145
1146 /* convert an ieee half/binary16 to float */
1147 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1148 ecb_function_ ecb_const float
1149 ecb_binary16_to_float (uint16_t x)
1150 {
1151 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1152 }
1153
1154#endif
1155
1156#endif
1157

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines