1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2014 Marc Alexander Lehmann <libecb@schmorp.de> |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
6 | * All rights reserved. |
6 | * All rights reserved. |
7 | * |
7 | * |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
9 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
… | |
… | |
23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
28 | * |
|
|
29 | * Alternatively, the contents of this file may be used under the terms of |
|
|
30 | * the GNU General Public License ("GPL") version 2 or any later version, |
|
|
31 | * in which case the provisions of the GPL are applicable instead of |
|
|
32 | * the above. If you wish to allow the use of your version of this file |
|
|
33 | * only under the terms of the GPL and not to allow others to use your |
|
|
34 | * version of this file under the BSD license, indicate your decision |
|
|
35 | * by deleting the provisions above and replace them with the notice |
|
|
36 | * and other provisions required by the GPL. If you do not delete the |
|
|
37 | * provisions above, a recipient may use your version of this file under |
|
|
38 | * either the BSD or the GPL. |
28 | */ |
39 | */ |
29 | |
40 | |
30 | #ifndef ECB_H |
41 | #ifndef ECB_H |
31 | #define ECB_H |
42 | #define ECB_H |
32 | |
43 | |
33 | /* 16 bits major, 16 bits minor */ |
44 | /* 16 bits major, 16 bits minor */ |
34 | #define ECB_VERSION 0x00010001 |
45 | #define ECB_VERSION 0x00010004 |
35 | |
46 | |
36 | #ifdef _WIN32 |
47 | #ifdef _WIN32 |
37 | typedef signed char int8_t; |
48 | typedef signed char int8_t; |
38 | typedef unsigned char uint8_t; |
49 | typedef unsigned char uint8_t; |
39 | typedef signed short int16_t; |
50 | typedef signed short int16_t; |
… | |
… | |
54 | #else |
65 | #else |
55 | #define ECB_PTRSIZE 4 |
66 | #define ECB_PTRSIZE 4 |
56 | typedef uint32_t uintptr_t; |
67 | typedef uint32_t uintptr_t; |
57 | typedef int32_t intptr_t; |
68 | typedef int32_t intptr_t; |
58 | #endif |
69 | #endif |
59 | typedef intptr_t ptrdiff_t; |
|
|
60 | #else |
70 | #else |
61 | #include <inttypes.h> |
71 | #include <inttypes.h> |
62 | #if UINTMAX_MAX > 0xffffffffU |
72 | #if UINTMAX_MAX > 0xffffffffU |
63 | #define ECB_PTRSIZE 8 |
73 | #define ECB_PTRSIZE 8 |
64 | #else |
74 | #else |
65 | #define ECB_PTRSIZE 4 |
75 | #define ECB_PTRSIZE 4 |
|
|
76 | #endif |
|
|
77 | #endif |
|
|
78 | |
|
|
79 | /* work around x32 idiocy by defining proper macros */ |
|
|
80 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
|
|
81 | #if _ILP32 |
|
|
82 | #define ECB_AMD64_X32 1 |
|
|
83 | #else |
|
|
84 | #define ECB_AMD64 1 |
66 | #endif |
85 | #endif |
67 | #endif |
86 | #endif |
68 | |
87 | |
69 | /* many compilers define _GNUC_ to some versions but then only implement |
88 | /* many compilers define _GNUC_ to some versions but then only implement |
70 | * what their idiot authors think are the "more important" extensions, |
89 | * what their idiot authors think are the "more important" extensions, |
71 | * causing enormous grief in return for some better fake benchmark numbers. |
90 | * causing enormous grief in return for some better fake benchmark numbers. |
72 | * or so. |
91 | * or so. |
73 | * we try to detect these and simply assume they are not gcc - if they have |
92 | * we try to detect these and simply assume they are not gcc - if they have |
74 | * an issue with that they should have done it right in the first place. |
93 | * an issue with that they should have done it right in the first place. |
75 | */ |
94 | */ |
76 | #ifndef ECB_GCC_VERSION |
|
|
77 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
95 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
78 | #define ECB_GCC_VERSION(major,minor) 0 |
96 | #define ECB_GCC_VERSION(major,minor) 0 |
79 | #else |
97 | #else |
80 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
98 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
81 | #endif |
99 | #endif |
82 | #endif |
|
|
83 | |
100 | |
84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
101 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
102 | |
86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
103 | #if __clang__ && defined(__has_builtin) |
|
|
104 | #define ECB_CLANG_BUILTIN(x) __has_builtin(x) |
|
|
105 | #else |
|
|
106 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
107 | #endif |
|
|
108 | |
|
|
109 | #if __clang__ && defined(__has_extension) |
|
|
110 | #define ECB_CLANG_EXTENSION(x) __has_extension(x) |
|
|
111 | #else |
|
|
112 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
113 | #endif |
|
|
114 | |
87 | #define ECB_CPP (__cplusplus+0) |
115 | #define ECB_CPP (__cplusplus+0) |
88 | #define ECB_CPP98 (__cplusplus >= 199711L) |
|
|
89 | #define ECB_CPP11 (__cplusplus >= 201103L) |
116 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
117 | |
|
|
118 | #if ECB_CPP |
|
|
119 | #define ECB_C 0 |
|
|
120 | #define ECB_STDC_VERSION 0 |
|
|
121 | #else |
|
|
122 | #define ECB_C 1 |
|
|
123 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
124 | #endif |
|
|
125 | |
|
|
126 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
127 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
128 | |
|
|
129 | #if ECB_CPP |
|
|
130 | #define ECB_EXTERN_C extern "C" |
|
|
131 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
|
|
132 | #define ECB_EXTERN_C_END } |
|
|
133 | #else |
|
|
134 | #define ECB_EXTERN_C extern |
|
|
135 | #define ECB_EXTERN_C_BEG |
|
|
136 | #define ECB_EXTERN_C_END |
|
|
137 | #endif |
90 | |
138 | |
91 | /*****************************************************************************/ |
139 | /*****************************************************************************/ |
92 | |
140 | |
93 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
141 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
94 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
142 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
95 | |
143 | |
96 | #if ECB_NO_THREADS |
144 | #if ECB_NO_THREADS |
97 | # define ECB_NO_SMP 1 |
145 | #define ECB_NO_SMP 1 |
98 | #endif |
146 | #endif |
99 | |
147 | |
100 | #if ECB_NO_SMP |
148 | #if ECB_NO_SMP |
101 | #define ECB_MEMORY_FENCE do { } while (0) |
149 | #define ECB_MEMORY_FENCE do { } while (0) |
102 | #endif |
150 | #endif |
… | |
… | |
117 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
165 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
118 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
166 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
119 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
167 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
120 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
168 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
121 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
170 | #elif __aarch64__ |
|
|
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
122 | #elif __sparc || __sparc__ |
172 | #elif (__sparc || __sparc__) && !__sparcv8 |
123 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
173 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
124 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
174 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
125 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
175 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
126 | #elif defined __s390__ || defined __s390x__ |
176 | #elif defined __s390__ || defined __s390x__ |
127 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
177 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
128 | #elif defined __mips__ |
178 | #elif defined __mips__ |
|
|
179 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
180 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
181 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
130 | #elif defined __alpha__ |
182 | #elif defined __alpha__ |
131 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
183 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
132 | #elif defined __hppa__ |
184 | #elif defined __hppa__ |
133 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
185 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
134 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
186 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
135 | #elif defined __ia64__ |
187 | #elif defined __ia64__ |
136 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
188 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
189 | #elif defined __m68k__ |
|
|
190 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
191 | #elif defined __m88k__ |
|
|
192 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
193 | #elif defined __sh__ |
|
|
194 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
137 | #endif |
195 | #endif |
138 | #endif |
196 | #endif |
139 | #endif |
197 | #endif |
140 | |
198 | |
141 | #ifndef ECB_MEMORY_FENCE |
199 | #ifndef ECB_MEMORY_FENCE |
142 | #if ECB_GCC_VERSION(4,7) |
200 | #if ECB_GCC_VERSION(4,7) |
143 | /* see comment below about the C11 memory model. in short - avoid */ |
201 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
144 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
202 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
145 | #elif defined __clang && __has_feature (cxx_atomic) |
203 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
146 | /* see above */ |
204 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
205 | |
|
|
206 | #elif ECB_CLANG_EXTENSION(c_atomic) |
|
|
207 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
147 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
208 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
209 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
210 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
211 | |
148 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
212 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
149 | #define ECB_MEMORY_FENCE __sync_synchronize () |
213 | #define ECB_MEMORY_FENCE __sync_synchronize () |
150 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
214 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
151 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
215 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
216 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
217 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
218 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
219 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
152 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
220 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
153 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
221 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
154 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
222 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
155 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
223 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
156 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
224 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
170 | #ifndef ECB_MEMORY_FENCE |
238 | #ifndef ECB_MEMORY_FENCE |
171 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
239 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
172 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
240 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
173 | /* not just C11 atomics and atomic accesses */ |
241 | /* not just C11 atomics and atomic accesses */ |
174 | #include <stdatomic.h> |
242 | #include <stdatomic.h> |
175 | /* unfortunately, the C11 memory model seems to be very limited, and unable to express */ |
243 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
176 | /* simple barrier semantics. That means we need to take out thor's hammer. */ |
244 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
245 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
246 | /* for most usages, or gcc and clang have a bug */ |
|
|
247 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
248 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
249 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
250 | /* for all __atomic_thread_fence's except seq_cst */ |
177 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
251 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
178 | #endif |
|
|
179 | #endif |
252 | #endif |
180 | #endif |
253 | #endif |
181 | |
254 | |
182 | #ifndef ECB_MEMORY_FENCE |
255 | #ifndef ECB_MEMORY_FENCE |
183 | #if !ECB_AVOID_PTHREADS |
256 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
232 | #define ECB_STRINGIFY_(a) # a |
305 | #define ECB_STRINGIFY_(a) # a |
233 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
306 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
234 | |
307 | |
235 | #define ecb_function_ ecb_inline |
308 | #define ecb_function_ ecb_inline |
236 | |
309 | |
237 | #if ECB_GCC_VERSION(3,1) |
310 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
238 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
311 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
312 | #else |
|
|
313 | #define ecb_attribute(attrlist) |
|
|
314 | #endif |
|
|
315 | |
|
|
316 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
239 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
317 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
318 | #else |
|
|
319 | /* possible C11 impl for integral types |
|
|
320 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
321 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
322 | |
|
|
323 | #define ecb_is_constant(expr) 0 |
|
|
324 | #endif |
|
|
325 | |
|
|
326 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
240 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
327 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
328 | #else |
|
|
329 | #define ecb_expect(expr,value) (expr) |
|
|
330 | #endif |
|
|
331 | |
|
|
332 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
241 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
333 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
242 | #else |
334 | #else |
243 | #define ecb_attribute(attrlist) |
|
|
244 | #define ecb_is_constant(expr) 0 |
|
|
245 | #define ecb_expect(expr,value) (expr) |
|
|
246 | #define ecb_prefetch(addr,rw,locality) |
335 | #define ecb_prefetch(addr,rw,locality) |
247 | #endif |
336 | #endif |
248 | |
337 | |
249 | /* no emulation for ecb_decltype */ |
338 | /* no emulation for ecb_decltype */ |
250 | #if ECB_GCC_VERSION(4,5) |
339 | #if ECB_CPP11 |
|
|
340 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
341 | struct ecb_decltype_t { typedef T type; }; |
251 | #define ecb_decltype(x) __decltype(x) |
342 | #define ecb_decltype(x) ecb_decltype_t<decltype(x)>::type |
252 | #elif ECB_GCC_VERSION(3,0) |
343 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
253 | #define ecb_decltype(x) __typeof(x) |
344 | #define ecb_decltype(x) __typeof__ (x) |
|
|
345 | #endif |
|
|
346 | |
|
|
347 | #if _MSC_VER >= 1300 |
|
|
348 | #define ecb_deprecated __declspec(deprecated) |
|
|
349 | #else |
|
|
350 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
254 | #endif |
351 | #endif |
255 | |
352 | |
256 | #define ecb_noinline ecb_attribute ((__noinline__)) |
353 | #define ecb_noinline ecb_attribute ((__noinline__)) |
257 | #define ecb_unused ecb_attribute ((__unused__)) |
354 | #define ecb_unused ecb_attribute ((__unused__)) |
258 | #define ecb_const ecb_attribute ((__const__)) |
355 | #define ecb_const ecb_attribute ((__const__)) |
259 | #define ecb_pure ecb_attribute ((__pure__)) |
356 | #define ecb_pure ecb_attribute ((__pure__)) |
260 | |
357 | |
|
|
358 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */ |
261 | #if ECB_C11 |
359 | #if ECB_C11 |
262 | #define ecb_noreturn _Noreturn |
360 | #define ecb_noreturn _Noreturn |
263 | #else |
361 | #else |
264 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
362 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
265 | #endif |
363 | #endif |
… | |
… | |
282 | /* for compatibility to the rest of the world */ |
380 | /* for compatibility to the rest of the world */ |
283 | #define ecb_likely(expr) ecb_expect_true (expr) |
381 | #define ecb_likely(expr) ecb_expect_true (expr) |
284 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
382 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
285 | |
383 | |
286 | /* count trailing zero bits and count # of one bits */ |
384 | /* count trailing zero bits and count # of one bits */ |
287 | #if ECB_GCC_VERSION(3,4) |
385 | #if ECB_GCC_VERSION(3,4) \ |
|
|
386 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
387 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
388 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
288 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
389 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
289 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
390 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
290 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
391 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
291 | #define ecb_ctz32(x) __builtin_ctz (x) |
392 | #define ecb_ctz32(x) __builtin_ctz (x) |
292 | #define ecb_ctz64(x) __builtin_ctzll (x) |
393 | #define ecb_ctz64(x) __builtin_ctzll (x) |
… | |
… | |
423 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
524 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
424 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
525 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
425 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
526 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
426 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
527 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
427 | |
528 | |
428 | #if ECB_GCC_VERSION(4,3) |
529 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
429 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
530 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
430 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
531 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
431 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
532 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
432 | #else |
533 | #else |
433 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
534 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
… | |
… | |
450 | { |
551 | { |
451 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
552 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
452 | } |
553 | } |
453 | #endif |
554 | #endif |
454 | |
555 | |
455 | #if ECB_GCC_VERSION(4,5) |
556 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
456 | #define ecb_unreachable() __builtin_unreachable () |
557 | #define ecb_unreachable() __builtin_unreachable () |
457 | #else |
558 | #else |
458 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
559 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
459 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
560 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
460 | ecb_inline void ecb_unreachable (void) { } |
561 | ecb_inline void ecb_unreachable (void) { } |
461 | #endif |
562 | #endif |
462 | |
563 | |
463 | /* try to tell the compiler that some condition is definitely true */ |
564 | /* try to tell the compiler that some condition is definitely true */ |
464 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
565 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
465 | |
566 | |
466 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
567 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
467 | ecb_inline unsigned char |
568 | ecb_inline unsigned char |
468 | ecb_byteorder_helper (void) |
569 | ecb_byteorder_helper (void) |
469 | { |
570 | { |
470 | const uint32_t u = 0x11223344; |
571 | /* the union code still generates code under pressure in gcc, */ |
471 | return *(unsigned char *)&u; |
572 | /* but less than using pointers, and always seems to */ |
|
|
573 | /* successfully return a constant. */ |
|
|
574 | /* the reason why we have this horrible preprocessor mess */ |
|
|
575 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
576 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
577 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
578 | return 0x44; |
|
|
579 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
580 | return 0x44; |
|
|
581 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
582 | return 0x11; |
|
|
583 | #else |
|
|
584 | union |
|
|
585 | { |
|
|
586 | uint32_t i; |
|
|
587 | uint8_t c; |
|
|
588 | } u = { 0x11223344 }; |
|
|
589 | return u.c; |
|
|
590 | #endif |
472 | } |
591 | } |
473 | |
592 | |
474 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
593 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
475 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
594 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
476 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
595 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
… | |
… | |
507 | } |
626 | } |
508 | #else |
627 | #else |
509 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
628 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
510 | #endif |
629 | #endif |
511 | |
630 | |
|
|
631 | /*******************************************************************************/ |
|
|
632 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
633 | |
|
|
634 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
635 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
636 | #if 0 \ |
|
|
637 | || __i386 || __i386__ \ |
|
|
638 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
|
|
639 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
640 | || defined __s390__ || defined __s390x__ \ |
|
|
641 | || defined __mips__ \ |
|
|
642 | || defined __alpha__ \ |
|
|
643 | || defined __hppa__ \ |
|
|
644 | || defined __ia64__ \ |
|
|
645 | || defined __m68k__ \ |
|
|
646 | || defined __m88k__ \ |
|
|
647 | || defined __sh__ \ |
|
|
648 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
649 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
650 | || defined __aarch64__ |
|
|
651 | #define ECB_STDFP 1 |
|
|
652 | #include <string.h> /* for memcpy */ |
|
|
653 | #else |
|
|
654 | #define ECB_STDFP 0 |
|
|
655 | #endif |
|
|
656 | |
|
|
657 | #ifndef ECB_NO_LIBM |
|
|
658 | |
|
|
659 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
660 | |
|
|
661 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
662 | #ifdef INFINITY |
|
|
663 | #define ECB_INFINITY INFINITY |
|
|
664 | #else |
|
|
665 | #define ECB_INFINITY HUGE_VAL |
512 | #endif |
666 | #endif |
513 | |
667 | |
|
|
668 | #ifdef NAN |
|
|
669 | #define ECB_NAN NAN |
|
|
670 | #else |
|
|
671 | #define ECB_NAN ECB_INFINITY |
|
|
672 | #endif |
|
|
673 | |
|
|
674 | /* converts an ieee half/binary16 to a float */ |
|
|
675 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
676 | ecb_function_ float |
|
|
677 | ecb_binary16_to_float (uint16_t x) |
|
|
678 | { |
|
|
679 | int e = (x >> 10) & 0x1f; |
|
|
680 | int m = x & 0x3ff; |
|
|
681 | float r; |
|
|
682 | |
|
|
683 | if (!e ) r = ldexpf (m , -24); |
|
|
684 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
685 | else if (m ) r = ECB_NAN; |
|
|
686 | else r = ECB_INFINITY; |
|
|
687 | |
|
|
688 | return x & 0x8000 ? -r : r; |
|
|
689 | } |
|
|
690 | |
|
|
691 | /* convert a float to ieee single/binary32 */ |
|
|
692 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
|
|
693 | ecb_function_ uint32_t |
|
|
694 | ecb_float_to_binary32 (float x) |
|
|
695 | { |
|
|
696 | uint32_t r; |
|
|
697 | |
|
|
698 | #if ECB_STDFP |
|
|
699 | memcpy (&r, &x, 4); |
|
|
700 | #else |
|
|
701 | /* slow emulation, works for anything but -0 */ |
|
|
702 | uint32_t m; |
|
|
703 | int e; |
|
|
704 | |
|
|
705 | if (x == 0e0f ) return 0x00000000U; |
|
|
706 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
707 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
708 | if (x != x ) return 0x7fbfffffU; |
|
|
709 | |
|
|
710 | m = frexpf (x, &e) * 0x1000000U; |
|
|
711 | |
|
|
712 | r = m & 0x80000000U; |
|
|
713 | |
|
|
714 | if (r) |
|
|
715 | m = -m; |
|
|
716 | |
|
|
717 | if (e <= -126) |
|
|
718 | { |
|
|
719 | m &= 0xffffffU; |
|
|
720 | m >>= (-125 - e); |
|
|
721 | e = -126; |
|
|
722 | } |
|
|
723 | |
|
|
724 | r |= (e + 126) << 23; |
|
|
725 | r |= m & 0x7fffffU; |
|
|
726 | #endif |
|
|
727 | |
|
|
728 | return r; |
|
|
729 | } |
|
|
730 | |
|
|
731 | /* converts an ieee single/binary32 to a float */ |
|
|
732 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
|
|
733 | ecb_function_ float |
|
|
734 | ecb_binary32_to_float (uint32_t x) |
|
|
735 | { |
|
|
736 | float r; |
|
|
737 | |
|
|
738 | #if ECB_STDFP |
|
|
739 | memcpy (&r, &x, 4); |
|
|
740 | #else |
|
|
741 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
742 | int neg = x >> 31; |
|
|
743 | int e = (x >> 23) & 0xffU; |
|
|
744 | |
|
|
745 | x &= 0x7fffffU; |
|
|
746 | |
|
|
747 | if (e) |
|
|
748 | x |= 0x800000U; |
|
|
749 | else |
|
|
750 | e = 1; |
|
|
751 | |
|
|
752 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
753 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
754 | |
|
|
755 | r = neg ? -r : r; |
|
|
756 | #endif |
|
|
757 | |
|
|
758 | return r; |
|
|
759 | } |
|
|
760 | |
|
|
761 | /* convert a double to ieee double/binary64 */ |
|
|
762 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
|
|
763 | ecb_function_ uint64_t |
|
|
764 | ecb_double_to_binary64 (double x) |
|
|
765 | { |
|
|
766 | uint64_t r; |
|
|
767 | |
|
|
768 | #if ECB_STDFP |
|
|
769 | memcpy (&r, &x, 8); |
|
|
770 | #else |
|
|
771 | /* slow emulation, works for anything but -0 */ |
|
|
772 | uint64_t m; |
|
|
773 | int e; |
|
|
774 | |
|
|
775 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
776 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
777 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
778 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
779 | |
|
|
780 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
781 | |
|
|
782 | r = m & 0x8000000000000000;; |
|
|
783 | |
|
|
784 | if (r) |
|
|
785 | m = -m; |
|
|
786 | |
|
|
787 | if (e <= -1022) |
|
|
788 | { |
|
|
789 | m &= 0x1fffffffffffffU; |
|
|
790 | m >>= (-1021 - e); |
|
|
791 | e = -1022; |
|
|
792 | } |
|
|
793 | |
|
|
794 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
795 | r |= m & 0xfffffffffffffU; |
|
|
796 | #endif |
|
|
797 | |
|
|
798 | return r; |
|
|
799 | } |
|
|
800 | |
|
|
801 | /* converts an ieee double/binary64 to a double */ |
|
|
802 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
|
|
803 | ecb_function_ double |
|
|
804 | ecb_binary64_to_double (uint64_t x) |
|
|
805 | { |
|
|
806 | double r; |
|
|
807 | |
|
|
808 | #if ECB_STDFP |
|
|
809 | memcpy (&r, &x, 8); |
|
|
810 | #else |
|
|
811 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
812 | int neg = x >> 63; |
|
|
813 | int e = (x >> 52) & 0x7ffU; |
|
|
814 | |
|
|
815 | x &= 0xfffffffffffffU; |
|
|
816 | |
|
|
817 | if (e) |
|
|
818 | x |= 0x10000000000000U; |
|
|
819 | else |
|
|
820 | e = 1; |
|
|
821 | |
|
|
822 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
823 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
824 | |
|
|
825 | r = neg ? -r : r; |
|
|
826 | #endif |
|
|
827 | |
|
|
828 | return r; |
|
|
829 | } |
|
|
830 | |
|
|
831 | #endif |
|
|
832 | |
|
|
833 | #endif |
|
|
834 | |