1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2015 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 0x00010005 |
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 | #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 |
66 | #endif |
88 | #endif |
67 | #endif |
89 | #endif |
68 | |
90 | |
69 | /* many compilers define _GNUC_ to some versions but then only implement |
91 | /* many compilers define _GNUC_ to some versions but then only implement |
70 | * what their idiot authors think are the "more important" extensions, |
92 | * what their idiot authors think are the "more important" extensions, |
71 | * causing enormous grief in return for some better fake benchmark numbers. |
93 | * causing enormous grief in return for some better fake benchmark numbers. |
72 | * or so. |
94 | * or so. |
73 | * we try to detect these and simply assume they are not gcc - if they have |
95 | * 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. |
96 | * an issue with that they should have done it right in the first place. |
75 | */ |
97 | */ |
76 | #ifndef ECB_GCC_VERSION |
|
|
77 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
98 | #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 |
99 | #define ECB_GCC_VERSION(major,minor) 0 |
79 | #else |
100 | #else |
80 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
101 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
81 | #endif |
102 | #endif |
82 | #endif |
|
|
83 | |
103 | |
84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
104 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
105 | |
86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
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 | |
87 | #define ECB_CPP (__cplusplus+0) |
118 | #define ECB_CPP (__cplusplus+0) |
88 | #define ECB_CPP98 (__cplusplus >= 199711L) |
|
|
89 | #define ECB_CPP11 (__cplusplus >= 201103L) |
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 |
90 | |
141 | |
91 | /*****************************************************************************/ |
142 | /*****************************************************************************/ |
92 | |
143 | |
93 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
144 | /* 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 */ |
145 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
95 | |
146 | |
96 | #if ECB_NO_THREADS |
147 | #if ECB_NO_THREADS |
97 | # define ECB_NO_SMP 1 |
148 | #define ECB_NO_SMP 1 |
98 | #endif |
149 | #endif |
99 | |
150 | |
100 | #if ECB_NO_THREADS || ECB_NO_SMP |
151 | #if ECB_NO_SMP |
101 | #define ECB_MEMORY_FENCE do { } while (0) |
152 | #define ECB_MEMORY_FENCE do { } while (0) |
102 | #endif |
153 | #endif |
103 | |
154 | |
104 | #ifndef ECB_MEMORY_FENCE |
155 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
105 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
156 | #if __xlC__ && ECB_CPP |
106 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
157 | #include <builtins.h> |
107 | /* not just C11 atomics and atomic accesses */ |
|
|
108 | #include <stdatomic.h> |
|
|
109 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_acq_rel) |
|
|
110 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
111 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
|
|
112 | #endif |
|
|
113 | #endif |
158 | #endif |
114 | |
159 | |
115 | #ifndef ECB_MEMORY_FENCE |
160 | #ifndef ECB_MEMORY_FENCE |
116 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
161 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
117 | #if __i386 || __i386__ |
162 | #if __i386 || __i386__ |
118 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
163 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
119 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
164 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
120 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
165 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
121 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
166 | #elif ECB_GCC_AMD64 |
122 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
167 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
123 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
168 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
124 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
169 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
125 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
170 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
127 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
172 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
128 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
173 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
174 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
130 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
175 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
131 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
176 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
132 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
177 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
133 | #elif __sparc || __sparc__ |
178 | #elif __aarch64__ |
|
|
179 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
|
|
180 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
134 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
181 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
135 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
182 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
136 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
183 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
137 | #elif defined __s390__ || defined __s390x__ |
184 | #elif defined __s390__ || defined __s390x__ |
138 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
185 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
139 | #elif defined __mips__ |
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. */ |
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
189 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
141 | #elif defined __alpha__ |
190 | #elif defined __alpha__ |
142 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
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") |
143 | #endif |
203 | #endif |
144 | #endif |
204 | #endif |
145 | #endif |
205 | #endif |
146 | |
206 | |
147 | #ifndef ECB_MEMORY_FENCE |
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 | |
148 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
220 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
149 | #define ECB_MEMORY_FENCE __sync_synchronize () |
221 | #define ECB_MEMORY_FENCE __sync_synchronize () |
150 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
222 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
151 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
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() |
152 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
228 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
153 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
229 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
154 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
230 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
155 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
231 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
156 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
232 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
166 | #define ECB_MEMORY_FENCE __sync () |
242 | #define ECB_MEMORY_FENCE __sync () |
167 | #endif |
243 | #endif |
168 | #endif |
244 | #endif |
169 | |
245 | |
170 | #ifndef ECB_MEMORY_FENCE |
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 |
171 | #if !ECB_AVOID_PTHREADS |
264 | #if !ECB_AVOID_PTHREADS |
172 | /* |
265 | /* |
173 | * if you get undefined symbol references to pthread_mutex_lock, |
266 | * if you get undefined symbol references to pthread_mutex_lock, |
174 | * or failure to find pthread.h, then you should implement |
267 | * or failure to find pthread.h, then you should implement |
175 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
268 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
… | |
… | |
193 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
286 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
194 | #endif |
287 | #endif |
195 | |
288 | |
196 | /*****************************************************************************/ |
289 | /*****************************************************************************/ |
197 | |
290 | |
198 | #if __cplusplus |
291 | #if ECB_CPP |
199 | #define ecb_inline static inline |
292 | #define ecb_inline static inline |
200 | #elif ECB_GCC_VERSION(2,5) |
293 | #elif ECB_GCC_VERSION(2,5) |
201 | #define ecb_inline static __inline__ |
294 | #define ecb_inline static __inline__ |
202 | #elif ECB_C99 |
295 | #elif ECB_C99 |
203 | #define ecb_inline static inline |
296 | #define ecb_inline static inline |
… | |
… | |
217 | |
310 | |
218 | #define ECB_CONCAT_(a, b) a ## b |
311 | #define ECB_CONCAT_(a, b) a ## b |
219 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
312 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
220 | #define ECB_STRINGIFY_(a) # a |
313 | #define ECB_STRINGIFY_(a) # a |
221 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
314 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
315 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
222 | |
316 | |
223 | #define ecb_function_ ecb_inline |
317 | #define ecb_function_ ecb_inline |
224 | |
318 | |
225 | #if ECB_GCC_VERSION(3,1) |
319 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
226 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
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) |
227 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
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) |
228 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
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) |
229 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
342 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
230 | #else |
343 | #else |
231 | #define ecb_attribute(attrlist) |
|
|
232 | #define ecb_is_constant(expr) 0 |
|
|
233 | #define ecb_expect(expr,value) (expr) |
|
|
234 | #define ecb_prefetch(addr,rw,locality) |
344 | #define ecb_prefetch(addr,rw,locality) |
235 | #endif |
345 | #endif |
236 | |
346 | |
237 | /* no emulation for ecb_decltype */ |
347 | /* no emulation for ecb_decltype */ |
238 | #if ECB_GCC_VERSION(4,5) |
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; }; |
239 | #define ecb_decltype(x) __decltype(x) |
351 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
240 | #elif ECB_GCC_VERSION(3,0) |
352 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
241 | #define ecb_decltype(x) __typeof(x) |
353 | #define ecb_decltype(x) __typeof__ (x) |
242 | #endif |
354 | #endif |
243 | |
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 |
244 | #define ecb_noinline ecb_attribute ((__noinline__)) |
373 | #define ecb_noinline ecb_attribute ((__noinline__)) |
|
|
374 | #endif |
|
|
375 | |
245 | #define ecb_unused ecb_attribute ((__unused__)) |
376 | #define ecb_unused ecb_attribute ((__unused__)) |
246 | #define ecb_const ecb_attribute ((__const__)) |
377 | #define ecb_const ecb_attribute ((__const__)) |
247 | #define ecb_pure ecb_attribute ((__pure__)) |
378 | #define ecb_pure ecb_attribute ((__pure__)) |
248 | |
379 | |
249 | #if ECB_C11 |
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 */ |
250 | #define ecb_noreturn _Noreturn |
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) |
251 | #else |
388 | #else |
252 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
389 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
253 | #endif |
390 | #endif |
254 | |
391 | |
255 | #if ECB_GCC_VERSION(4,3) |
392 | #if ECB_GCC_VERSION(4,3) |
… | |
… | |
270 | /* for compatibility to the rest of the world */ |
407 | /* for compatibility to the rest of the world */ |
271 | #define ecb_likely(expr) ecb_expect_true (expr) |
408 | #define ecb_likely(expr) ecb_expect_true (expr) |
272 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
409 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
273 | |
410 | |
274 | /* count trailing zero bits and count # of one bits */ |
411 | /* count trailing zero bits and count # of one bits */ |
275 | #if ECB_GCC_VERSION(3,4) |
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)) |
276 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
416 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
277 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
417 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
278 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
418 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
279 | #define ecb_ctz32(x) __builtin_ctz (x) |
419 | #define ecb_ctz32(x) __builtin_ctz (x) |
280 | #define ecb_ctz64(x) __builtin_ctzll (x) |
420 | #define ecb_ctz64(x) __builtin_ctzll (x) |
281 | #define ecb_popcount32(x) __builtin_popcount (x) |
421 | #define ecb_popcount32(x) __builtin_popcount (x) |
282 | /* no popcountll */ |
422 | /* no popcountll */ |
283 | #else |
423 | #else |
284 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
424 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
285 | ecb_function_ int |
425 | ecb_function_ ecb_const int |
286 | ecb_ctz32 (uint32_t x) |
426 | ecb_ctz32 (uint32_t x) |
287 | { |
427 | { |
288 | int r = 0; |
428 | int r = 0; |
289 | |
429 | |
290 | x &= ~x + 1; /* this isolates the lowest bit */ |
430 | x &= ~x + 1; /* this isolates the lowest bit */ |
… | |
… | |
304 | #endif |
444 | #endif |
305 | |
445 | |
306 | return r; |
446 | return r; |
307 | } |
447 | } |
308 | |
448 | |
309 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
449 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
310 | ecb_function_ int |
450 | ecb_function_ ecb_const int |
311 | ecb_ctz64 (uint64_t x) |
451 | ecb_ctz64 (uint64_t x) |
312 | { |
452 | { |
313 | int shift = x & 0xffffffffU ? 0 : 32; |
453 | int shift = x & 0xffffffffU ? 0 : 32; |
314 | return ecb_ctz32 (x >> shift) + shift; |
454 | return ecb_ctz32 (x >> shift) + shift; |
315 | } |
455 | } |
316 | |
456 | |
317 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
457 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
318 | ecb_function_ int |
458 | ecb_function_ ecb_const int |
319 | ecb_popcount32 (uint32_t x) |
459 | ecb_popcount32 (uint32_t x) |
320 | { |
460 | { |
321 | x -= (x >> 1) & 0x55555555; |
461 | x -= (x >> 1) & 0x55555555; |
322 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
462 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
323 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
463 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
324 | x *= 0x01010101; |
464 | x *= 0x01010101; |
325 | |
465 | |
326 | return x >> 24; |
466 | return x >> 24; |
327 | } |
467 | } |
328 | |
468 | |
329 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
469 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
330 | ecb_function_ int ecb_ld32 (uint32_t x) |
470 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
331 | { |
471 | { |
332 | int r = 0; |
472 | int r = 0; |
333 | |
473 | |
334 | if (x >> 16) { x >>= 16; r += 16; } |
474 | if (x >> 16) { x >>= 16; r += 16; } |
335 | if (x >> 8) { x >>= 8; r += 8; } |
475 | if (x >> 8) { x >>= 8; r += 8; } |
… | |
… | |
338 | if (x >> 1) { r += 1; } |
478 | if (x >> 1) { r += 1; } |
339 | |
479 | |
340 | return r; |
480 | return r; |
341 | } |
481 | } |
342 | |
482 | |
343 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
483 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
344 | ecb_function_ int ecb_ld64 (uint64_t x) |
484 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
345 | { |
485 | { |
346 | int r = 0; |
486 | int r = 0; |
347 | |
487 | |
348 | if (x >> 32) { x >>= 32; r += 32; } |
488 | if (x >> 32) { x >>= 32; r += 32; } |
349 | |
489 | |
350 | return r + ecb_ld32 (x); |
490 | return r + ecb_ld32 (x); |
351 | } |
491 | } |
352 | #endif |
492 | #endif |
353 | |
493 | |
354 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
494 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
355 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
495 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
356 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
496 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
357 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
497 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
358 | |
498 | |
359 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
499 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
360 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
500 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
361 | { |
501 | { |
362 | return ( (x * 0x0802U & 0x22110U) |
502 | return ( (x * 0x0802U & 0x22110U) |
363 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
503 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
364 | } |
504 | } |
365 | |
505 | |
366 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
506 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
367 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
507 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
368 | { |
508 | { |
369 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
509 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
370 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
510 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
371 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
511 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
372 | x = ( x >> 8 ) | ( x << 8); |
512 | x = ( x >> 8 ) | ( x << 8); |
373 | |
513 | |
374 | return x; |
514 | return x; |
375 | } |
515 | } |
376 | |
516 | |
377 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
517 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
378 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
518 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
379 | { |
519 | { |
380 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
520 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
381 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
521 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
382 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
522 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
383 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
523 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
… | |
… | |
386 | return x; |
526 | return x; |
387 | } |
527 | } |
388 | |
528 | |
389 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
529 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
390 | /* so for this version we are lazy */ |
530 | /* so for this version we are lazy */ |
391 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
531 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
392 | ecb_function_ int |
532 | ecb_function_ ecb_const int |
393 | ecb_popcount64 (uint64_t x) |
533 | ecb_popcount64 (uint64_t x) |
394 | { |
534 | { |
395 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
535 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
396 | } |
536 | } |
397 | |
537 | |
398 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
538 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
399 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
539 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
400 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
540 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
401 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
541 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
402 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
542 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
403 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
543 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
404 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
544 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
405 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
545 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
406 | |
546 | |
407 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
547 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
408 | ecb_inline uint8_t ecb_rotr8 (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); } |
409 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
549 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
410 | ecb_inline uint16_t ecb_rotr16 (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); } |
411 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
551 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
412 | ecb_inline uint32_t ecb_rotr32 (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); } |
413 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
553 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
414 | ecb_inline uint64_t ecb_rotr64 (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); } |
415 | |
555 | |
416 | #if ECB_GCC_VERSION(4,3) |
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 |
417 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
560 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
561 | #endif |
418 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
562 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
419 | #define ecb_bswap64(x) __builtin_bswap64 (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))) |
420 | #else |
569 | #else |
421 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
570 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
422 | ecb_function_ uint16_t |
571 | ecb_function_ ecb_const uint16_t |
423 | ecb_bswap16 (uint16_t x) |
572 | ecb_bswap16 (uint16_t x) |
424 | { |
573 | { |
425 | return ecb_rotl16 (x, 8); |
574 | return ecb_rotl16 (x, 8); |
426 | } |
575 | } |
427 | |
576 | |
428 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
577 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
429 | ecb_function_ uint32_t |
578 | ecb_function_ ecb_const uint32_t |
430 | ecb_bswap32 (uint32_t x) |
579 | ecb_bswap32 (uint32_t x) |
431 | { |
580 | { |
432 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
581 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
433 | } |
582 | } |
434 | |
583 | |
435 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
584 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
436 | ecb_function_ uint64_t |
585 | ecb_function_ ecb_const uint64_t |
437 | ecb_bswap64 (uint64_t x) |
586 | ecb_bswap64 (uint64_t x) |
438 | { |
587 | { |
439 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
588 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
440 | } |
589 | } |
441 | #endif |
590 | #endif |
442 | |
591 | |
443 | #if ECB_GCC_VERSION(4,5) |
592 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
444 | #define ecb_unreachable() __builtin_unreachable () |
593 | #define ecb_unreachable() __builtin_unreachable () |
445 | #else |
594 | #else |
446 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
595 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
447 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
596 | ecb_inline ecb_noreturn void ecb_unreachable (void); |
448 | ecb_inline void ecb_unreachable (void) { } |
597 | ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
449 | #endif |
598 | #endif |
450 | |
599 | |
451 | /* try to tell the compiler that some condition is definitely true */ |
600 | /* try to tell the compiler that some condition is definitely true */ |
452 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
601 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
453 | |
602 | |
454 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
603 | ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); |
455 | ecb_inline unsigned char |
604 | ecb_inline ecb_const unsigned char |
456 | ecb_byteorder_helper (void) |
605 | ecb_byteorder_helper (void) |
457 | { |
606 | { |
458 | const uint32_t u = 0x11223344; |
607 | /* the union code still generates code under pressure in gcc, */ |
459 | return *(unsigned char *)&u; |
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 |
460 | } |
627 | } |
461 | |
628 | |
462 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
629 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
463 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
630 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
464 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
631 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
465 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
632 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
466 | |
633 | |
467 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
634 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
468 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
635 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
469 | #else |
636 | #else |
470 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
637 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
471 | #endif |
638 | #endif |
472 | |
639 | |
473 | #if __cplusplus |
640 | #if ECB_CPP |
474 | template<typename T> |
641 | template<typename T> |
475 | static inline T ecb_div_rd (T val, T div) |
642 | static inline T ecb_div_rd (T val, T div) |
476 | { |
643 | { |
477 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
644 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
478 | } |
645 | } |
… | |
… | |
495 | } |
662 | } |
496 | #else |
663 | #else |
497 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
664 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
498 | #endif |
665 | #endif |
499 | |
666 | |
|
|
667 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint16_t x); |
|
|
668 | ecb_function_ ecb_const uint32_t |
|
|
669 | ecb_binary16_to_binary32 (uint16_t x) |
|
|
670 | { |
|
|
671 | unsigned int s = (x & 0x8000) << (31 - 15); |
|
|
672 | int e = (x >> 10) & 0x001f; |
|
|
673 | unsigned int m = x & 0x03ff; |
|
|
674 | |
|
|
675 | if (ecb_expect_false (e == 31)) |
|
|
676 | /* infinity or NaN */ |
|
|
677 | e = 255 - (127 - 15); |
|
|
678 | else if (ecb_expect_false (!e)) |
|
|
679 | { |
|
|
680 | if (ecb_expect_true (!m)) |
|
|
681 | /* zero, handled by code below by forcing e to 0 */ |
|
|
682 | e = 0 - (127 - 15); |
|
|
683 | else |
|
|
684 | { |
|
|
685 | /* subnormal, renormalise */ |
|
|
686 | unsigned int s = 10 - ecb_ld32 (m); |
|
|
687 | |
|
|
688 | m = (m << s) & 0x3ff; /* mask implicit bit */ |
|
|
689 | e -= s - 1; |
|
|
690 | } |
|
|
691 | } |
|
|
692 | |
|
|
693 | /* e and m now are normalised, or zero, (or inf or nan) */ |
|
|
694 | e += 127 - 15; |
|
|
695 | |
|
|
696 | return s | (e << 23) | (m << (23 - 10)); |
|
|
697 | } |
|
|
698 | |
|
|
699 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
|
|
700 | ecb_function_ ecb_const uint16_t |
|
|
701 | ecb_binary32_to_binary16 (uint32_t x) |
|
|
702 | { |
|
|
703 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
|
|
704 | unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
|
|
705 | unsigned int m = x & 0x007fffff; |
|
|
706 | |
|
|
707 | x &= 0x7fffffff; |
|
|
708 | |
|
|
709 | /* if it's within range of binary16 normals, use fast path */ |
|
|
710 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
|
|
711 | { |
|
|
712 | /* mantissa round-to-even */ |
|
|
713 | m += 0x00000fff + ((m >> (23 - 10)) & 1); |
|
|
714 | |
|
|
715 | /* handle overflow */ |
|
|
716 | if (ecb_expect_false (m >= 0x00800000)) |
|
|
717 | { |
|
|
718 | m >>= 1; |
|
|
719 | e += 1; |
|
|
720 | } |
|
|
721 | |
|
|
722 | return s | (e << 10) | (m >> (23 - 10)); |
|
|
723 | } |
|
|
724 | |
|
|
725 | /* handle large numbers and infinity */ |
|
|
726 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
|
|
727 | return s | 0x7c00; |
|
|
728 | |
|
|
729 | /* handle zero and subnormals */ |
|
|
730 | if (ecb_expect_true (x < 0x38800000)) |
|
|
731 | { |
|
|
732 | /* zero */ |
|
|
733 | if (ecb_expect_true (!x)) |
|
|
734 | return s; |
|
|
735 | |
|
|
736 | /* handle subnormals */ |
|
|
737 | |
|
|
738 | m |= 0x00800000; /* make implicit bit explicit */ |
|
|
739 | |
|
|
740 | /* very tricky - we need to round to the nearest e (+10) bit value */ |
|
|
741 | { |
|
|
742 | unsigned int bits = 14 - e; |
|
|
743 | unsigned int half = (1 << (bits - 1)) - 1; |
|
|
744 | unsigned int even = (m >> bits) & 1; |
|
|
745 | |
|
|
746 | /* if this overflows, we will end up with a normalised number */ |
|
|
747 | m = (m + half + even) >> bits; |
|
|
748 | } |
|
|
749 | |
|
|
750 | return s | m; |
|
|
751 | } |
|
|
752 | |
|
|
753 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
|
|
754 | m >>= 13; |
|
|
755 | |
|
|
756 | return s | 0x7c00 | m | !m; |
|
|
757 | } |
|
|
758 | |
|
|
759 | /*******************************************************************************/ |
|
|
760 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
761 | |
|
|
762 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
763 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
764 | #if 0 \ |
|
|
765 | || __i386 || __i386__ \ |
|
|
766 | || ECB_GCC_AMD64 \ |
|
|
767 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
768 | || defined __s390__ || defined __s390x__ \ |
|
|
769 | || defined __mips__ \ |
|
|
770 | || defined __alpha__ \ |
|
|
771 | || defined __hppa__ \ |
|
|
772 | || defined __ia64__ \ |
|
|
773 | || defined __m68k__ \ |
|
|
774 | || defined __m88k__ \ |
|
|
775 | || defined __sh__ \ |
|
|
776 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
|
|
777 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
778 | || defined __aarch64__ |
|
|
779 | #define ECB_STDFP 1 |
|
|
780 | #include <string.h> /* for memcpy */ |
|
|
781 | #else |
|
|
782 | #define ECB_STDFP 0 |
|
|
783 | #endif |
|
|
784 | |
|
|
785 | #ifndef ECB_NO_LIBM |
|
|
786 | |
|
|
787 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
788 | |
|
|
789 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
790 | #ifdef INFINITY |
|
|
791 | #define ECB_INFINITY INFINITY |
|
|
792 | #else |
|
|
793 | #define ECB_INFINITY HUGE_VAL |
500 | #endif |
794 | #endif |
501 | |
795 | |
|
|
796 | #ifdef NAN |
|
|
797 | #define ECB_NAN NAN |
|
|
798 | #else |
|
|
799 | #define ECB_NAN ECB_INFINITY |
|
|
800 | #endif |
|
|
801 | |
|
|
802 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
|
|
803 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
804 | #define ecb_frexpf(x,e) frexpf ((x), (e)) |
|
|
805 | #else |
|
|
806 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
|
|
807 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
|
|
808 | #endif |
|
|
809 | |
|
|
810 | /* convert a float to ieee single/binary32 */ |
|
|
811 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
|
|
812 | ecb_function_ ecb_const uint32_t |
|
|
813 | ecb_float_to_binary32 (float x) |
|
|
814 | { |
|
|
815 | uint32_t r; |
|
|
816 | |
|
|
817 | #if ECB_STDFP |
|
|
818 | memcpy (&r, &x, 4); |
|
|
819 | #else |
|
|
820 | /* slow emulation, works for anything but -0 */ |
|
|
821 | uint32_t m; |
|
|
822 | int e; |
|
|
823 | |
|
|
824 | if (x == 0e0f ) return 0x00000000U; |
|
|
825 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
826 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
827 | if (x != x ) return 0x7fbfffffU; |
|
|
828 | |
|
|
829 | m = ecb_frexpf (x, &e) * 0x1000000U; |
|
|
830 | |
|
|
831 | r = m & 0x80000000U; |
|
|
832 | |
|
|
833 | if (r) |
|
|
834 | m = -m; |
|
|
835 | |
|
|
836 | if (e <= -126) |
|
|
837 | { |
|
|
838 | m &= 0xffffffU; |
|
|
839 | m >>= (-125 - e); |
|
|
840 | e = -126; |
|
|
841 | } |
|
|
842 | |
|
|
843 | r |= (e + 126) << 23; |
|
|
844 | r |= m & 0x7fffffU; |
|
|
845 | #endif |
|
|
846 | |
|
|
847 | return r; |
|
|
848 | } |
|
|
849 | |
|
|
850 | /* converts an ieee single/binary32 to a float */ |
|
|
851 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
|
|
852 | ecb_function_ ecb_const float |
|
|
853 | ecb_binary32_to_float (uint32_t x) |
|
|
854 | { |
|
|
855 | float r; |
|
|
856 | |
|
|
857 | #if ECB_STDFP |
|
|
858 | memcpy (&r, &x, 4); |
|
|
859 | #else |
|
|
860 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
861 | int neg = x >> 31; |
|
|
862 | int e = (x >> 23) & 0xffU; |
|
|
863 | |
|
|
864 | x &= 0x7fffffU; |
|
|
865 | |
|
|
866 | if (e) |
|
|
867 | x |= 0x800000U; |
|
|
868 | else |
|
|
869 | e = 1; |
|
|
870 | |
|
|
871 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
872 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
873 | |
|
|
874 | r = neg ? -r : r; |
|
|
875 | #endif |
|
|
876 | |
|
|
877 | return r; |
|
|
878 | } |
|
|
879 | |
|
|
880 | /* convert a double to ieee double/binary64 */ |
|
|
881 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
|
|
882 | ecb_function_ ecb_const uint64_t |
|
|
883 | ecb_double_to_binary64 (double x) |
|
|
884 | { |
|
|
885 | uint64_t r; |
|
|
886 | |
|
|
887 | #if ECB_STDFP |
|
|
888 | memcpy (&r, &x, 8); |
|
|
889 | #else |
|
|
890 | /* slow emulation, works for anything but -0 */ |
|
|
891 | uint64_t m; |
|
|
892 | int e; |
|
|
893 | |
|
|
894 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
895 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
896 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
897 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
898 | |
|
|
899 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
900 | |
|
|
901 | r = m & 0x8000000000000000;; |
|
|
902 | |
|
|
903 | if (r) |
|
|
904 | m = -m; |
|
|
905 | |
|
|
906 | if (e <= -1022) |
|
|
907 | { |
|
|
908 | m &= 0x1fffffffffffffU; |
|
|
909 | m >>= (-1021 - e); |
|
|
910 | e = -1022; |
|
|
911 | } |
|
|
912 | |
|
|
913 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
914 | r |= m & 0xfffffffffffffU; |
|
|
915 | #endif |
|
|
916 | |
|
|
917 | return r; |
|
|
918 | } |
|
|
919 | |
|
|
920 | /* converts an ieee double/binary64 to a double */ |
|
|
921 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
|
|
922 | ecb_function_ ecb_const double |
|
|
923 | ecb_binary64_to_double (uint64_t x) |
|
|
924 | { |
|
|
925 | double r; |
|
|
926 | |
|
|
927 | #if ECB_STDFP |
|
|
928 | memcpy (&r, &x, 8); |
|
|
929 | #else |
|
|
930 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
931 | int neg = x >> 63; |
|
|
932 | int e = (x >> 52) & 0x7ffU; |
|
|
933 | |
|
|
934 | x &= 0xfffffffffffffU; |
|
|
935 | |
|
|
936 | if (e) |
|
|
937 | x |= 0x10000000000000U; |
|
|
938 | else |
|
|
939 | e = 1; |
|
|
940 | |
|
|
941 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
942 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
943 | |
|
|
944 | r = neg ? -r : r; |
|
|
945 | #endif |
|
|
946 | |
|
|
947 | return r; |
|
|
948 | } |
|
|
949 | |
|
|
950 | /* convert a float to ieee half/binary16 */ |
|
|
951 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
|
|
952 | ecb_function_ ecb_const uint16_t |
|
|
953 | ecb_float_to_binary16 (float x) |
|
|
954 | { |
|
|
955 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
|
|
956 | } |
|
|
957 | |
|
|
958 | /* convert an ieee half/binary16 to float */ |
|
|
959 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
960 | ecb_function_ ecb_const float |
|
|
961 | ecb_binary16_to_float (uint16_t x) |
|
|
962 | { |
|
|
963 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
|
|
964 | } |
|
|
965 | |
|
|
966 | #endif |
|
|
967 | |
|
|
968 | #endif |
|
|
969 | |