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 |
|
|
43 | |
|
|
44 | /* 16 bits major, 16 bits minor */ |
|
|
45 | #define ECB_VERSION 0x00010006 |
32 | |
46 | |
33 | #ifdef _WIN32 |
47 | #ifdef _WIN32 |
34 | typedef signed char int8_t; |
48 | typedef signed char int8_t; |
35 | typedef unsigned char uint8_t; |
49 | typedef unsigned char uint8_t; |
36 | typedef signed short int16_t; |
50 | typedef signed short int16_t; |
… | |
… | |
42 | typedef unsigned long long uint64_t; |
56 | typedef unsigned long long uint64_t; |
43 | #else /* _MSC_VER || __BORLANDC__ */ |
57 | #else /* _MSC_VER || __BORLANDC__ */ |
44 | typedef signed __int64 int64_t; |
58 | typedef signed __int64 int64_t; |
45 | typedef unsigned __int64 uint64_t; |
59 | typedef unsigned __int64 uint64_t; |
46 | #endif |
60 | #endif |
|
|
61 | #ifdef _WIN64 |
|
|
62 | #define ECB_PTRSIZE 8 |
|
|
63 | typedef uint64_t uintptr_t; |
|
|
64 | typedef int64_t intptr_t; |
|
|
65 | #else |
|
|
66 | #define ECB_PTRSIZE 4 |
|
|
67 | typedef uint32_t uintptr_t; |
|
|
68 | typedef int32_t intptr_t; |
|
|
69 | #endif |
47 | #else |
70 | #else |
48 | #include <inttypes.h> |
71 | #include <inttypes.h> |
|
|
72 | #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
|
|
73 | #define ECB_PTRSIZE 8 |
|
|
74 | #else |
|
|
75 | #define ECB_PTRSIZE 4 |
|
|
76 | #endif |
|
|
77 | #endif |
|
|
78 | |
|
|
79 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
|
|
80 | #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
|
|
81 | |
|
|
82 | /* work around x32 idiocy by defining proper macros */ |
|
|
83 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
|
|
84 | #if _ILP32 |
|
|
85 | #define ECB_AMD64_X32 1 |
|
|
86 | #else |
|
|
87 | #define ECB_AMD64 1 |
|
|
88 | #endif |
49 | #endif |
89 | #endif |
50 | |
90 | |
51 | /* many compilers define _GNUC_ to some versions but then only implement |
91 | /* many compilers define _GNUC_ to some versions but then only implement |
52 | * what their idiot authors think are the "more important" extensions, |
92 | * what their idiot authors think are the "more important" extensions, |
53 | * causing enormous grief in return for some better fake benchmark numbers. |
93 | * causing enormous grief in return for some better fake benchmark numbers. |
54 | * or so. |
94 | * or so. |
55 | * 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 |
56 | * 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. |
57 | */ |
97 | */ |
58 | #ifndef ECB_GCC_VERSION |
|
|
59 | #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__ |
60 | #define ECB_GCC_VERSION(major,minor) 0 |
99 | #define ECB_GCC_VERSION(major,minor) 0 |
61 | #else |
100 | #else |
62 | #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))) |
63 | #endif |
102 | #endif |
|
|
103 | |
|
|
104 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
|
|
105 | |
|
|
106 | #if __clang__ && defined __has_builtin |
|
|
107 | #define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
|
|
108 | #else |
|
|
109 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
110 | #endif |
|
|
111 | |
|
|
112 | #if __clang__ && defined __has_extension |
|
|
113 | #define ECB_CLANG_EXTENSION(x) __has_extension (x) |
|
|
114 | #else |
|
|
115 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
116 | #endif |
|
|
117 | |
|
|
118 | #define ECB_CPP (__cplusplus+0) |
|
|
119 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
120 | #define ECB_CPP14 (__cplusplus >= 201402L) |
|
|
121 | #define ECB_CPP17 (__cplusplus >= 201703L) |
|
|
122 | |
|
|
123 | #if ECB_CPP |
|
|
124 | #define ECB_C 0 |
|
|
125 | #define ECB_STDC_VERSION 0 |
|
|
126 | #else |
|
|
127 | #define ECB_C 1 |
|
|
128 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
129 | #endif |
|
|
130 | |
|
|
131 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
132 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
133 | #define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
|
|
134 | |
|
|
135 | #if ECB_CPP |
|
|
136 | #define ECB_EXTERN_C extern "C" |
|
|
137 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
|
|
138 | #define ECB_EXTERN_C_END } |
|
|
139 | #else |
|
|
140 | #define ECB_EXTERN_C extern |
|
|
141 | #define ECB_EXTERN_C_BEG |
|
|
142 | #define ECB_EXTERN_C_END |
64 | #endif |
143 | #endif |
65 | |
144 | |
66 | /*****************************************************************************/ |
145 | /*****************************************************************************/ |
67 | |
146 | |
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
147 | /* 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 */ |
148 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
70 | |
149 | |
71 | #if ECB_NO_THREADS |
150 | #if ECB_NO_THREADS |
72 | # define ECB_NO_SMP 1 |
151 | #define ECB_NO_SMP 1 |
73 | #endif |
152 | #endif |
74 | |
153 | |
75 | #if ECB_NO_THREADS || ECB_NO_SMP |
154 | #if ECB_NO_SMP |
76 | #define ECB_MEMORY_FENCE do { } while (0) |
155 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
156 | #endif |
|
|
157 | |
|
|
158 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
|
|
159 | #if __xlC__ && ECB_CPP |
|
|
160 | #include <builtins.h> |
|
|
161 | #endif |
|
|
162 | |
|
|
163 | #if 1400 <= _MSC_VER |
|
|
164 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
77 | #endif |
165 | #endif |
78 | |
166 | |
79 | #ifndef ECB_MEMORY_FENCE |
167 | #ifndef ECB_MEMORY_FENCE |
80 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
168 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
169 | #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") |
81 | #if __i386 || __i386__ |
170 | #if __i386 || __i386__ |
82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
83 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
172 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
84 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
173 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
85 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
174 | #elif ECB_GCC_AMD64 |
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
175 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
87 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
176 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
88 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
177 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
89 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
178 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
90 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
179 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
180 | #elif defined __ARM_ARCH_2__ \ |
|
|
181 | || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
|
|
182 | || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
|
|
183 | || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
|
|
184 | || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
|
|
185 | || defined __ARM_ARCH_5TEJ__ |
|
|
186 | /* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
91 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
187 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
92 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
188 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
|
|
189 | || defined __ARM_ARCH_6T2__ |
93 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
190 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
94 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
191 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
95 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
192 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
96 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
193 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
97 | #elif __sparc || __sparc__ |
194 | #elif __aarch64__ |
|
|
195 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
|
|
196 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
98 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
197 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
99 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
198 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
100 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
199 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
101 | #elif defined __s390__ || defined __s390x__ |
200 | #elif defined __s390__ || defined __s390x__ |
102 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
201 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
103 | #elif defined __mips__ |
202 | #elif defined __mips__ |
|
|
203 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
204 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
104 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
205 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
105 | #elif defined __alpha__ |
206 | #elif defined __alpha__ |
106 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
207 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
208 | #elif defined __hppa__ |
|
|
209 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
210 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
211 | #elif defined __ia64__ |
|
|
212 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
213 | #elif defined __m68k__ |
|
|
214 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
215 | #elif defined __m88k__ |
|
|
216 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
217 | #elif defined __sh__ |
|
|
218 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
107 | #endif |
219 | #endif |
108 | #endif |
220 | #endif |
109 | #endif |
221 | #endif |
110 | |
222 | |
111 | #ifndef ECB_MEMORY_FENCE |
223 | #ifndef ECB_MEMORY_FENCE |
|
|
224 | #if ECB_GCC_VERSION(4,7) |
|
|
225 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
226 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
227 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
228 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
229 | #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
230 | |
|
|
231 | #elif ECB_CLANG_EXTENSION(c_atomic) |
|
|
232 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
233 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
234 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
235 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
236 | #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
237 | |
112 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
238 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
113 | #define ECB_MEMORY_FENCE __sync_synchronize () |
239 | #define ECB_MEMORY_FENCE __sync_synchronize () |
114 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
240 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
115 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
241 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
242 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
243 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
244 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
245 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
116 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
246 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
117 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
247 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
118 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
248 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
119 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
249 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
120 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
250 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
121 | #elif defined _WIN32 |
251 | #elif defined _WIN32 |
122 | #include <WinNT.h> |
252 | #include <WinNT.h> |
123 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
253 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
124 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
254 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
125 | #include <mbarrier.h> |
255 | #include <mbarrier.h> |
126 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
256 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
127 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
257 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
128 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
258 | #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |
|
|
259 | #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () |
129 | #elif __xlC__ |
260 | #elif __xlC__ |
130 | #define ECB_MEMORY_FENCE __sync () |
261 | #define ECB_MEMORY_FENCE __sync () |
|
|
262 | #endif |
|
|
263 | #endif |
|
|
264 | |
|
|
265 | #ifndef ECB_MEMORY_FENCE |
|
|
266 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
267 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
268 | /* not just C11 atomics and atomic accesses */ |
|
|
269 | #include <stdatomic.h> |
|
|
270 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
271 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
272 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
131 | #endif |
273 | #endif |
132 | #endif |
274 | #endif |
133 | |
275 | |
134 | #ifndef ECB_MEMORY_FENCE |
276 | #ifndef ECB_MEMORY_FENCE |
135 | #if !ECB_AVOID_PTHREADS |
277 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
155 | |
297 | |
156 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
298 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
157 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
299 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
158 | #endif |
300 | #endif |
159 | |
301 | |
|
|
302 | #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE |
|
|
303 | #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ |
|
|
304 | #endif |
|
|
305 | |
160 | /*****************************************************************************/ |
306 | /*****************************************************************************/ |
161 | |
307 | |
162 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
308 | #if ECB_CPP |
163 | |
|
|
164 | #if __cplusplus |
|
|
165 | #define ecb_inline static inline |
309 | #define ecb_inline static inline |
166 | #elif ECB_GCC_VERSION(2,5) |
310 | #elif ECB_GCC_VERSION(2,5) |
167 | #define ecb_inline static __inline__ |
311 | #define ecb_inline static __inline__ |
168 | #elif ECB_C99 |
312 | #elif ECB_C99 |
169 | #define ecb_inline static inline |
313 | #define ecb_inline static inline |
… | |
… | |
183 | |
327 | |
184 | #define ECB_CONCAT_(a, b) a ## b |
328 | #define ECB_CONCAT_(a, b) a ## b |
185 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
329 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
186 | #define ECB_STRINGIFY_(a) # a |
330 | #define ECB_STRINGIFY_(a) # a |
187 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
331 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
332 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
188 | |
333 | |
189 | #define ecb_function_ ecb_inline |
334 | #define ecb_function_ ecb_inline |
190 | |
335 | |
191 | #if ECB_GCC_VERSION(3,1) |
336 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
192 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
337 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
338 | #else |
|
|
339 | #define ecb_attribute(attrlist) |
|
|
340 | #endif |
|
|
341 | |
|
|
342 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
193 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
343 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
344 | #else |
|
|
345 | /* possible C11 impl for integral types |
|
|
346 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
347 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
348 | |
|
|
349 | #define ecb_is_constant(expr) 0 |
|
|
350 | #endif |
|
|
351 | |
|
|
352 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
194 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
353 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
354 | #else |
|
|
355 | #define ecb_expect(expr,value) (expr) |
|
|
356 | #endif |
|
|
357 | |
|
|
358 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
195 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
359 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
196 | #else |
360 | #else |
197 | #define ecb_attribute(attrlist) |
|
|
198 | #define ecb_is_constant(expr) 0 |
|
|
199 | #define ecb_expect(expr,value) (expr) |
|
|
200 | #define ecb_prefetch(addr,rw,locality) |
361 | #define ecb_prefetch(addr,rw,locality) |
201 | #endif |
362 | #endif |
202 | |
363 | |
203 | /* no emulation for ecb_decltype */ |
364 | /* no emulation for ecb_decltype */ |
204 | #if ECB_GCC_VERSION(4,5) |
365 | #if ECB_CPP11 |
|
|
366 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
367 | template<class T> struct ecb_decltype_t { typedef T type; }; |
205 | #define ecb_decltype(x) __decltype(x) |
368 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
206 | #elif ECB_GCC_VERSION(3,0) |
369 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
207 | #define ecb_decltype(x) __typeof(x) |
370 | #define ecb_decltype(x) __typeof__ (x) |
208 | #endif |
371 | #endif |
209 | |
372 | |
|
|
373 | #if _MSC_VER >= 1300 |
|
|
374 | #define ecb_deprecated __declspec (deprecated) |
|
|
375 | #else |
|
|
376 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
|
|
377 | #endif |
|
|
378 | |
|
|
379 | #if _MSC_VER >= 1500 |
|
|
380 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
|
|
381 | #elif ECB_GCC_VERSION(4,5) |
|
|
382 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
|
|
383 | #else |
|
|
384 | #define ecb_deprecated_message(msg) ecb_deprecated |
|
|
385 | #endif |
|
|
386 | |
|
|
387 | #if _MSC_VER >= 1400 |
|
|
388 | #define ecb_noinline __declspec (noinline) |
|
|
389 | #else |
210 | #define ecb_noinline ecb_attribute ((__noinline__)) |
390 | #define ecb_noinline ecb_attribute ((__noinline__)) |
211 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
391 | #endif |
|
|
392 | |
212 | #define ecb_unused ecb_attribute ((__unused__)) |
393 | #define ecb_unused ecb_attribute ((__unused__)) |
213 | #define ecb_const ecb_attribute ((__const__)) |
394 | #define ecb_const ecb_attribute ((__const__)) |
214 | #define ecb_pure ecb_attribute ((__pure__)) |
395 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
396 | |
|
|
397 | #if ECB_C11 || __IBMC_NORETURN |
|
|
398 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
|
|
399 | #define ecb_noreturn _Noreturn |
|
|
400 | #elif ECB_CPP11 |
|
|
401 | #define ecb_noreturn [[noreturn]] |
|
|
402 | #elif _MSC_VER >= 1200 |
|
|
403 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
|
|
404 | #define ecb_noreturn __declspec (noreturn) |
|
|
405 | #else |
|
|
406 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
407 | #endif |
215 | |
408 | |
216 | #if ECB_GCC_VERSION(4,3) |
409 | #if ECB_GCC_VERSION(4,3) |
217 | #define ecb_artificial ecb_attribute ((__artificial__)) |
410 | #define ecb_artificial ecb_attribute ((__artificial__)) |
218 | #define ecb_hot ecb_attribute ((__hot__)) |
411 | #define ecb_hot ecb_attribute ((__hot__)) |
219 | #define ecb_cold ecb_attribute ((__cold__)) |
412 | #define ecb_cold ecb_attribute ((__cold__)) |
… | |
… | |
231 | /* for compatibility to the rest of the world */ |
424 | /* for compatibility to the rest of the world */ |
232 | #define ecb_likely(expr) ecb_expect_true (expr) |
425 | #define ecb_likely(expr) ecb_expect_true (expr) |
233 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
426 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
234 | |
427 | |
235 | /* count trailing zero bits and count # of one bits */ |
428 | /* count trailing zero bits and count # of one bits */ |
236 | #if ECB_GCC_VERSION(3,4) |
429 | #if ECB_GCC_VERSION(3,4) \ |
|
|
430 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
431 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
432 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
237 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
433 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
238 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
434 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
239 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
435 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
240 | #define ecb_ctz32(x) __builtin_ctz (x) |
436 | #define ecb_ctz32(x) __builtin_ctz (x) |
241 | #define ecb_ctz64(x) __builtin_ctzll (x) |
437 | #define ecb_ctz64(x) __builtin_ctzll (x) |
242 | #define ecb_popcount32(x) __builtin_popcount (x) |
438 | #define ecb_popcount32(x) __builtin_popcount (x) |
243 | /* no popcountll */ |
439 | /* no popcountll */ |
244 | #else |
440 | #else |
245 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
441 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
246 | ecb_function_ int |
442 | ecb_function_ ecb_const int |
247 | ecb_ctz32 (uint32_t x) |
443 | ecb_ctz32 (uint32_t x) |
248 | { |
444 | { |
|
|
445 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
446 | unsigned long r; |
|
|
447 | _BitScanForward (&r, x); |
|
|
448 | return (int)r; |
|
|
449 | #else |
249 | int r = 0; |
450 | int r = 0; |
250 | |
451 | |
251 | x &= ~x + 1; /* this isolates the lowest bit */ |
452 | x &= ~x + 1; /* this isolates the lowest bit */ |
252 | |
453 | |
253 | #if ECB_branchless_on_i386 |
454 | #if ECB_branchless_on_i386 |
… | |
… | |
263 | if (x & 0xff00ff00) r += 8; |
464 | if (x & 0xff00ff00) r += 8; |
264 | if (x & 0xffff0000) r += 16; |
465 | if (x & 0xffff0000) r += 16; |
265 | #endif |
466 | #endif |
266 | |
467 | |
267 | return r; |
468 | return r; |
|
|
469 | #endif |
268 | } |
470 | } |
269 | |
471 | |
270 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
472 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
271 | ecb_function_ int |
473 | ecb_function_ ecb_const int |
272 | ecb_ctz64 (uint64_t x) |
474 | ecb_ctz64 (uint64_t x) |
273 | { |
475 | { |
|
|
476 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
477 | unsigned long r; |
|
|
478 | _BitScanForward64 (&r, x); |
|
|
479 | return (int)r; |
|
|
480 | #else |
274 | int shift = x & 0xffffffffU ? 0 : 32; |
481 | int shift = x & 0xffffffff ? 0 : 32; |
275 | return ecb_ctz32 (x >> shift) + shift; |
482 | return ecb_ctz32 (x >> shift) + shift; |
|
|
483 | #endif |
276 | } |
484 | } |
277 | |
485 | |
278 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
486 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
279 | ecb_function_ int |
487 | ecb_function_ ecb_const int |
280 | ecb_popcount32 (uint32_t x) |
488 | ecb_popcount32 (uint32_t x) |
281 | { |
489 | { |
282 | x -= (x >> 1) & 0x55555555; |
490 | x -= (x >> 1) & 0x55555555; |
283 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
491 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
284 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
492 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
285 | x *= 0x01010101; |
493 | x *= 0x01010101; |
286 | |
494 | |
287 | return x >> 24; |
495 | return x >> 24; |
288 | } |
496 | } |
289 | |
497 | |
290 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
498 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
291 | ecb_function_ int ecb_ld32 (uint32_t x) |
499 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
292 | { |
500 | { |
|
|
501 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
502 | unsigned long r; |
|
|
503 | _BitScanReverse (&r, x); |
|
|
504 | return (int)r; |
|
|
505 | #else |
293 | int r = 0; |
506 | int r = 0; |
294 | |
507 | |
295 | if (x >> 16) { x >>= 16; r += 16; } |
508 | if (x >> 16) { x >>= 16; r += 16; } |
296 | if (x >> 8) { x >>= 8; r += 8; } |
509 | if (x >> 8) { x >>= 8; r += 8; } |
297 | if (x >> 4) { x >>= 4; r += 4; } |
510 | if (x >> 4) { x >>= 4; r += 4; } |
298 | if (x >> 2) { x >>= 2; r += 2; } |
511 | if (x >> 2) { x >>= 2; r += 2; } |
299 | if (x >> 1) { r += 1; } |
512 | if (x >> 1) { r += 1; } |
300 | |
513 | |
301 | return r; |
514 | return r; |
|
|
515 | #endif |
302 | } |
516 | } |
303 | |
517 | |
304 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
518 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
305 | ecb_function_ int ecb_ld64 (uint64_t x) |
519 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
306 | { |
520 | { |
|
|
521 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
522 | unsigned long r; |
|
|
523 | _BitScanReverse64 (&r, x); |
|
|
524 | return (int)r; |
|
|
525 | #else |
307 | int r = 0; |
526 | int r = 0; |
308 | |
527 | |
309 | if (x >> 32) { x >>= 32; r += 32; } |
528 | if (x >> 32) { x >>= 32; r += 32; } |
310 | |
529 | |
311 | return r + ecb_ld32 (x); |
530 | return r + ecb_ld32 (x); |
312 | } |
|
|
313 | #endif |
531 | #endif |
|
|
532 | } |
|
|
533 | #endif |
314 | |
534 | |
|
|
535 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
|
|
536 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
537 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
|
|
538 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
539 | |
315 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
540 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
316 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
541 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
317 | { |
542 | { |
318 | return ( (x * 0x0802U & 0x22110U) |
543 | return ( (x * 0x0802U & 0x22110U) |
319 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
544 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
320 | } |
545 | } |
321 | |
546 | |
322 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
547 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
323 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
548 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
324 | { |
549 | { |
325 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
550 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
326 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
551 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
327 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
552 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
328 | x = ( x >> 8 ) | ( x << 8); |
553 | x = ( x >> 8 ) | ( x << 8); |
329 | |
554 | |
330 | return x; |
555 | return x; |
331 | } |
556 | } |
332 | |
557 | |
333 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
558 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
334 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
559 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
335 | { |
560 | { |
336 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
561 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
337 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
562 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
338 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
563 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
339 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
564 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
… | |
… | |
342 | return x; |
567 | return x; |
343 | } |
568 | } |
344 | |
569 | |
345 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
570 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
346 | /* so for this version we are lazy */ |
571 | /* so for this version we are lazy */ |
347 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
572 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
348 | ecb_function_ int |
573 | ecb_function_ ecb_const int |
349 | ecb_popcount64 (uint64_t x) |
574 | ecb_popcount64 (uint64_t x) |
350 | { |
575 | { |
351 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
576 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
352 | } |
577 | } |
353 | |
578 | |
354 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
579 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
355 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
580 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
356 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
581 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
357 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
582 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
358 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
583 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
359 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
584 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
360 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
585 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
361 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
586 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
362 | |
587 | |
363 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
588 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
364 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
589 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
365 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
590 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
366 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
591 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
367 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
592 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
368 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
593 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
369 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
594 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
370 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
595 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
371 | |
596 | |
372 | #if ECB_GCC_VERSION(4,3) |
597 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
|
|
598 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
|
|
599 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
|
|
600 | #else |
373 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
601 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
602 | #endif |
374 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
603 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
375 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
604 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
|
|
605 | #elif _MSC_VER |
|
|
606 | #include <stdlib.h> |
|
|
607 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
|
|
608 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
|
|
609 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
376 | #else |
610 | #else |
377 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
611 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
378 | ecb_function_ uint16_t |
612 | ecb_function_ ecb_const uint16_t |
379 | ecb_bswap16 (uint16_t x) |
613 | ecb_bswap16 (uint16_t x) |
380 | { |
614 | { |
381 | return ecb_rotl16 (x, 8); |
615 | return ecb_rotl16 (x, 8); |
382 | } |
616 | } |
383 | |
617 | |
384 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
618 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
385 | ecb_function_ uint32_t |
619 | ecb_function_ ecb_const uint32_t |
386 | ecb_bswap32 (uint32_t x) |
620 | ecb_bswap32 (uint32_t x) |
387 | { |
621 | { |
388 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
622 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
389 | } |
623 | } |
390 | |
624 | |
391 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
625 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
392 | ecb_function_ uint64_t |
626 | ecb_function_ ecb_const uint64_t |
393 | ecb_bswap64 (uint64_t x) |
627 | ecb_bswap64 (uint64_t x) |
394 | { |
628 | { |
395 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
629 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
396 | } |
630 | } |
397 | #endif |
631 | #endif |
398 | |
632 | |
399 | #if ECB_GCC_VERSION(4,5) |
633 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
400 | #define ecb_unreachable() __builtin_unreachable () |
634 | #define ecb_unreachable() __builtin_unreachable () |
401 | #else |
635 | #else |
402 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
636 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
403 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
637 | ecb_inline ecb_noreturn void ecb_unreachable (void); |
404 | ecb_inline void ecb_unreachable (void) { } |
638 | ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
405 | #endif |
639 | #endif |
406 | |
640 | |
407 | /* try to tell the compiler that some condition is definitely true */ |
641 | /* try to tell the compiler that some condition is definitely true */ |
408 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
642 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
409 | |
643 | |
410 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
644 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
411 | ecb_inline unsigned char |
645 | ecb_inline ecb_const uint32_t |
412 | ecb_byteorder_helper (void) |
646 | ecb_byteorder_helper (void) |
413 | { |
647 | { |
414 | const uint32_t u = 0x11223344; |
648 | /* the union code still generates code under pressure in gcc, */ |
415 | return *(unsigned char *)&u; |
649 | /* but less than using pointers, and always seems to */ |
|
|
650 | /* successfully return a constant. */ |
|
|
651 | /* the reason why we have this horrible preprocessor mess */ |
|
|
652 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
653 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
654 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
|
|
655 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
|
|
656 | #define ECB_LITTLE_ENDIAN 1 |
|
|
657 | return 0x44332211; |
|
|
658 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
|
|
659 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
|
|
660 | #define ECB_BIG_ENDIAN 1 |
|
|
661 | return 0x11223344; |
|
|
662 | #else |
|
|
663 | union |
|
|
664 | { |
|
|
665 | uint8_t c[4]; |
|
|
666 | uint32_t u; |
|
|
667 | } u = { 0x11, 0x22, 0x33, 0x44 }; |
|
|
668 | return u.u; |
|
|
669 | #endif |
416 | } |
670 | } |
417 | |
671 | |
418 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
672 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
419 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
673 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
420 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
674 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
421 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
675 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
422 | |
676 | |
423 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
677 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
424 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
678 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
425 | #else |
679 | #else |
426 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
680 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
427 | #endif |
681 | #endif |
428 | |
682 | |
429 | #if __cplusplus |
683 | #if ECB_CPP |
430 | template<typename T> |
684 | template<typename T> |
431 | static inline T ecb_div_rd (T val, T div) |
685 | static inline T ecb_div_rd (T val, T div) |
432 | { |
686 | { |
433 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
687 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
434 | } |
688 | } |
… | |
… | |
451 | } |
705 | } |
452 | #else |
706 | #else |
453 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
707 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
454 | #endif |
708 | #endif |
455 | |
709 | |
|
|
710 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
|
|
711 | ecb_function_ ecb_const uint32_t |
|
|
712 | ecb_binary16_to_binary32 (uint32_t x) |
|
|
713 | { |
|
|
714 | unsigned int s = (x & 0x8000) << (31 - 15); |
|
|
715 | int e = (x >> 10) & 0x001f; |
|
|
716 | unsigned int m = x & 0x03ff; |
|
|
717 | |
|
|
718 | if (ecb_expect_false (e == 31)) |
|
|
719 | /* infinity or NaN */ |
|
|
720 | e = 255 - (127 - 15); |
|
|
721 | else if (ecb_expect_false (!e)) |
|
|
722 | { |
|
|
723 | if (ecb_expect_true (!m)) |
|
|
724 | /* zero, handled by code below by forcing e to 0 */ |
|
|
725 | e = 0 - (127 - 15); |
|
|
726 | else |
|
|
727 | { |
|
|
728 | /* subnormal, renormalise */ |
|
|
729 | unsigned int s = 10 - ecb_ld32 (m); |
|
|
730 | |
|
|
731 | m = (m << s) & 0x3ff; /* mask implicit bit */ |
|
|
732 | e -= s - 1; |
|
|
733 | } |
|
|
734 | } |
|
|
735 | |
|
|
736 | /* e and m now are normalised, or zero, (or inf or nan) */ |
|
|
737 | e += 127 - 15; |
|
|
738 | |
|
|
739 | return s | (e << 23) | (m << (23 - 10)); |
|
|
740 | } |
|
|
741 | |
|
|
742 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
|
|
743 | ecb_function_ ecb_const uint16_t |
|
|
744 | ecb_binary32_to_binary16 (uint32_t x) |
|
|
745 | { |
|
|
746 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
|
|
747 | unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
|
|
748 | unsigned int m = x & 0x007fffff; |
|
|
749 | |
|
|
750 | x &= 0x7fffffff; |
|
|
751 | |
|
|
752 | /* if it's within range of binary16 normals, use fast path */ |
|
|
753 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
|
|
754 | { |
|
|
755 | /* mantissa round-to-even */ |
|
|
756 | m += 0x00000fff + ((m >> (23 - 10)) & 1); |
|
|
757 | |
|
|
758 | /* handle overflow */ |
|
|
759 | if (ecb_expect_false (m >= 0x00800000)) |
|
|
760 | { |
|
|
761 | m >>= 1; |
|
|
762 | e += 1; |
|
|
763 | } |
|
|
764 | |
|
|
765 | return s | (e << 10) | (m >> (23 - 10)); |
|
|
766 | } |
|
|
767 | |
|
|
768 | /* handle large numbers and infinity */ |
|
|
769 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
|
|
770 | return s | 0x7c00; |
|
|
771 | |
|
|
772 | /* handle zero, subnormals and small numbers */ |
|
|
773 | if (ecb_expect_true (x < 0x38800000)) |
|
|
774 | { |
|
|
775 | /* zero */ |
|
|
776 | if (ecb_expect_true (!x)) |
|
|
777 | return s; |
|
|
778 | |
|
|
779 | /* handle subnormals */ |
|
|
780 | |
|
|
781 | /* too small, will be zero */ |
|
|
782 | if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
|
|
783 | return s; |
|
|
784 | |
|
|
785 | m |= 0x00800000; /* make implicit bit explicit */ |
|
|
786 | |
|
|
787 | /* very tricky - we need to round to the nearest e (+10) bit value */ |
|
|
788 | { |
|
|
789 | unsigned int bits = 14 - e; |
|
|
790 | unsigned int half = (1 << (bits - 1)) - 1; |
|
|
791 | unsigned int even = (m >> bits) & 1; |
|
|
792 | |
|
|
793 | /* if this overflows, we will end up with a normalised number */ |
|
|
794 | m = (m + half + even) >> bits; |
|
|
795 | } |
|
|
796 | |
|
|
797 | return s | m; |
|
|
798 | } |
|
|
799 | |
|
|
800 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
|
|
801 | m >>= 13; |
|
|
802 | |
|
|
803 | return s | 0x7c00 | m | !m; |
|
|
804 | } |
|
|
805 | |
|
|
806 | /*******************************************************************************/ |
|
|
807 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
808 | |
|
|
809 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
810 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
811 | #if 0 \ |
|
|
812 | || __i386 || __i386__ \ |
|
|
813 | || ECB_GCC_AMD64 \ |
|
|
814 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
815 | || defined __s390__ || defined __s390x__ \ |
|
|
816 | || defined __mips__ \ |
|
|
817 | || defined __alpha__ \ |
|
|
818 | || defined __hppa__ \ |
|
|
819 | || defined __ia64__ \ |
|
|
820 | || defined __m68k__ \ |
|
|
821 | || defined __m88k__ \ |
|
|
822 | || defined __sh__ \ |
|
|
823 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
|
|
824 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
825 | || defined __aarch64__ |
|
|
826 | #define ECB_STDFP 1 |
|
|
827 | #include <string.h> /* for memcpy */ |
|
|
828 | #else |
|
|
829 | #define ECB_STDFP 0 |
|
|
830 | #endif |
|
|
831 | |
|
|
832 | #ifndef ECB_NO_LIBM |
|
|
833 | |
|
|
834 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
835 | |
|
|
836 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
837 | #ifdef INFINITY |
|
|
838 | #define ECB_INFINITY INFINITY |
|
|
839 | #else |
|
|
840 | #define ECB_INFINITY HUGE_VAL |
456 | #endif |
841 | #endif |
457 | |
842 | |
|
|
843 | #ifdef NAN |
|
|
844 | #define ECB_NAN NAN |
|
|
845 | #else |
|
|
846 | #define ECB_NAN ECB_INFINITY |
|
|
847 | #endif |
|
|
848 | |
|
|
849 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
|
|
850 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
851 | #define ecb_frexpf(x,e) frexpf ((x), (e)) |
|
|
852 | #else |
|
|
853 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
|
|
854 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
|
|
855 | #endif |
|
|
856 | |
|
|
857 | /* convert a float to ieee single/binary32 */ |
|
|
858 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
|
|
859 | ecb_function_ ecb_const uint32_t |
|
|
860 | ecb_float_to_binary32 (float x) |
|
|
861 | { |
|
|
862 | uint32_t r; |
|
|
863 | |
|
|
864 | #if ECB_STDFP |
|
|
865 | memcpy (&r, &x, 4); |
|
|
866 | #else |
|
|
867 | /* slow emulation, works for anything but -0 */ |
|
|
868 | uint32_t m; |
|
|
869 | int e; |
|
|
870 | |
|
|
871 | if (x == 0e0f ) return 0x00000000U; |
|
|
872 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
873 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
874 | if (x != x ) return 0x7fbfffffU; |
|
|
875 | |
|
|
876 | m = ecb_frexpf (x, &e) * 0x1000000U; |
|
|
877 | |
|
|
878 | r = m & 0x80000000U; |
|
|
879 | |
|
|
880 | if (r) |
|
|
881 | m = -m; |
|
|
882 | |
|
|
883 | if (e <= -126) |
|
|
884 | { |
|
|
885 | m &= 0xffffffU; |
|
|
886 | m >>= (-125 - e); |
|
|
887 | e = -126; |
|
|
888 | } |
|
|
889 | |
|
|
890 | r |= (e + 126) << 23; |
|
|
891 | r |= m & 0x7fffffU; |
|
|
892 | #endif |
|
|
893 | |
|
|
894 | return r; |
|
|
895 | } |
|
|
896 | |
|
|
897 | /* converts an ieee single/binary32 to a float */ |
|
|
898 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
|
|
899 | ecb_function_ ecb_const float |
|
|
900 | ecb_binary32_to_float (uint32_t x) |
|
|
901 | { |
|
|
902 | float r; |
|
|
903 | |
|
|
904 | #if ECB_STDFP |
|
|
905 | memcpy (&r, &x, 4); |
|
|
906 | #else |
|
|
907 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
908 | int neg = x >> 31; |
|
|
909 | int e = (x >> 23) & 0xffU; |
|
|
910 | |
|
|
911 | x &= 0x7fffffU; |
|
|
912 | |
|
|
913 | if (e) |
|
|
914 | x |= 0x800000U; |
|
|
915 | else |
|
|
916 | e = 1; |
|
|
917 | |
|
|
918 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
919 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
920 | |
|
|
921 | r = neg ? -r : r; |
|
|
922 | #endif |
|
|
923 | |
|
|
924 | return r; |
|
|
925 | } |
|
|
926 | |
|
|
927 | /* convert a double to ieee double/binary64 */ |
|
|
928 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
|
|
929 | ecb_function_ ecb_const uint64_t |
|
|
930 | ecb_double_to_binary64 (double x) |
|
|
931 | { |
|
|
932 | uint64_t r; |
|
|
933 | |
|
|
934 | #if ECB_STDFP |
|
|
935 | memcpy (&r, &x, 8); |
|
|
936 | #else |
|
|
937 | /* slow emulation, works for anything but -0 */ |
|
|
938 | uint64_t m; |
|
|
939 | int e; |
|
|
940 | |
|
|
941 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
942 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
943 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
944 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
945 | |
|
|
946 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
947 | |
|
|
948 | r = m & 0x8000000000000000;; |
|
|
949 | |
|
|
950 | if (r) |
|
|
951 | m = -m; |
|
|
952 | |
|
|
953 | if (e <= -1022) |
|
|
954 | { |
|
|
955 | m &= 0x1fffffffffffffU; |
|
|
956 | m >>= (-1021 - e); |
|
|
957 | e = -1022; |
|
|
958 | } |
|
|
959 | |
|
|
960 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
961 | r |= m & 0xfffffffffffffU; |
|
|
962 | #endif |
|
|
963 | |
|
|
964 | return r; |
|
|
965 | } |
|
|
966 | |
|
|
967 | /* converts an ieee double/binary64 to a double */ |
|
|
968 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
|
|
969 | ecb_function_ ecb_const double |
|
|
970 | ecb_binary64_to_double (uint64_t x) |
|
|
971 | { |
|
|
972 | double r; |
|
|
973 | |
|
|
974 | #if ECB_STDFP |
|
|
975 | memcpy (&r, &x, 8); |
|
|
976 | #else |
|
|
977 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
978 | int neg = x >> 63; |
|
|
979 | int e = (x >> 52) & 0x7ffU; |
|
|
980 | |
|
|
981 | x &= 0xfffffffffffffU; |
|
|
982 | |
|
|
983 | if (e) |
|
|
984 | x |= 0x10000000000000U; |
|
|
985 | else |
|
|
986 | e = 1; |
|
|
987 | |
|
|
988 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
989 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
990 | |
|
|
991 | r = neg ? -r : r; |
|
|
992 | #endif |
|
|
993 | |
|
|
994 | return r; |
|
|
995 | } |
|
|
996 | |
|
|
997 | /* convert a float to ieee half/binary16 */ |
|
|
998 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
|
|
999 | ecb_function_ ecb_const uint16_t |
|
|
1000 | ecb_float_to_binary16 (float x) |
|
|
1001 | { |
|
|
1002 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
|
|
1003 | } |
|
|
1004 | |
|
|
1005 | /* convert an ieee half/binary16 to float */ |
|
|
1006 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
1007 | ecb_function_ ecb_const float |
|
|
1008 | ecb_binary16_to_float (uint16_t x) |
|
|
1009 | { |
|
|
1010 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
|
|
1011 | } |
|
|
1012 | |
|
|
1013 | #endif |
|
|
1014 | |
|
|
1015 | #endif |
|
|
1016 | |