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 0x00010007 |
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; |
… | |
… | |
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 | #else |
70 | #else |
60 | #include <inttypes.h> |
71 | #include <inttypes.h> |
61 | #if UINTMAX_MAX > 0xffffffffU |
72 | #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
62 | #define ECB_PTRSIZE 8 |
73 | #define ECB_PTRSIZE 8 |
63 | #else |
74 | #else |
64 | #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 | #ifndef ECB_OPTIMIZE_SIZE |
|
|
83 | #if __OPTIMIZE_SIZE__ |
|
|
84 | #define ECB_OPTIMIZE_SIZE 1 |
|
|
85 | #else |
|
|
86 | #define ECB_OPTIMIZE_SIZE 0 |
|
|
87 | #endif |
|
|
88 | #endif |
|
|
89 | |
|
|
90 | /* work around x32 idiocy by defining proper macros */ |
|
|
91 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
|
|
92 | #if _ILP32 |
|
|
93 | #define ECB_AMD64_X32 1 |
|
|
94 | #else |
|
|
95 | #define ECB_AMD64 1 |
65 | #endif |
96 | #endif |
66 | #endif |
97 | #endif |
67 | |
98 | |
68 | /* many compilers define _GNUC_ to some versions but then only implement |
99 | /* many compilers define _GNUC_ to some versions but then only implement |
69 | * what their idiot authors think are the "more important" extensions, |
100 | * what their idiot authors think are the "more important" extensions, |
70 | * causing enormous grief in return for some better fake benchmark numbers. |
101 | * causing enormous grief in return for some better fake benchmark numbers. |
71 | * or so. |
102 | * or so. |
72 | * we try to detect these and simply assume they are not gcc - if they have |
103 | * we try to detect these and simply assume they are not gcc - if they have |
73 | * an issue with that they should have done it right in the first place. |
104 | * an issue with that they should have done it right in the first place. |
74 | */ |
105 | */ |
75 | #ifndef ECB_GCC_VERSION |
|
|
76 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
106 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
77 | #define ECB_GCC_VERSION(major,minor) 0 |
107 | #define ECB_GCC_VERSION(major,minor) 0 |
78 | #else |
108 | #else |
79 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
109 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
80 | #endif |
110 | #endif |
81 | #endif |
|
|
82 | |
111 | |
83 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
112 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
84 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
113 | |
85 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
114 | #if __clang__ && defined __has_builtin |
|
|
115 | #define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
|
|
116 | #else |
|
|
117 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
118 | #endif |
|
|
119 | |
|
|
120 | #if __clang__ && defined __has_extension |
|
|
121 | #define ECB_CLANG_EXTENSION(x) __has_extension (x) |
|
|
122 | #else |
|
|
123 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
124 | #endif |
|
|
125 | |
86 | #define ECB_CPP (__cplusplus+0) |
126 | #define ECB_CPP (__cplusplus+0) |
87 | #define ECB_CPP11 (__cplusplus >= 201103L) |
127 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
128 | #define ECB_CPP14 (__cplusplus >= 201402L) |
|
|
129 | #define ECB_CPP17 (__cplusplus >= 201703L) |
|
|
130 | |
|
|
131 | #if ECB_CPP |
|
|
132 | #define ECB_C 0 |
|
|
133 | #define ECB_STDC_VERSION 0 |
|
|
134 | #else |
|
|
135 | #define ECB_C 1 |
|
|
136 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
137 | #endif |
|
|
138 | |
|
|
139 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
140 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
141 | #define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
88 | |
142 | |
89 | #if ECB_CPP |
143 | #if ECB_CPP |
90 | #define ECB_EXTERN_C extern "C" |
144 | #define ECB_EXTERN_C extern "C" |
91 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
145 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
92 | #define ECB_EXTERN_C_END } |
146 | #define ECB_EXTERN_C_END } |
… | |
… | |
107 | |
161 | |
108 | #if ECB_NO_SMP |
162 | #if ECB_NO_SMP |
109 | #define ECB_MEMORY_FENCE do { } while (0) |
163 | #define ECB_MEMORY_FENCE do { } while (0) |
110 | #endif |
164 | #endif |
111 | |
165 | |
|
|
166 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
|
|
167 | #if __xlC__ && ECB_CPP |
|
|
168 | #include <builtins.h> |
|
|
169 | #endif |
|
|
170 | |
|
|
171 | #if 1400 <= _MSC_VER |
|
|
172 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
|
|
173 | #endif |
|
|
174 | |
112 | #ifndef ECB_MEMORY_FENCE |
175 | #ifndef ECB_MEMORY_FENCE |
113 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
176 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
177 | #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") |
114 | #if __i386 || __i386__ |
178 | #if __i386 || __i386__ |
115 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
179 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
116 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
180 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
117 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
181 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
118 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
182 | #elif ECB_GCC_AMD64 |
119 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
183 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
120 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
184 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
121 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
185 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
122 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
186 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
123 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
187 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
188 | #elif defined __ARM_ARCH_2__ \ |
|
|
189 | || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
|
|
190 | || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
|
|
191 | || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
|
|
192 | || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
|
|
193 | || defined __ARM_ARCH_5TEJ__ |
|
|
194 | /* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
124 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
195 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
125 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
196 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
|
|
197 | || defined __ARM_ARCH_6T2__ |
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
198 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
127 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
199 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
128 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
200 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
201 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
130 | #elif __sparc || __sparc__ |
202 | #elif __aarch64__ |
|
|
203 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
|
|
204 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
131 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
205 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
132 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
206 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
207 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
134 | #elif defined __s390__ || defined __s390x__ |
208 | #elif defined __s390__ || defined __s390x__ |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
209 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
136 | #elif defined __mips__ |
210 | #elif defined __mips__ |
|
|
211 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
212 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
137 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
213 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
138 | #elif defined __alpha__ |
214 | #elif defined __alpha__ |
139 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
215 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
140 | #elif defined __hppa__ |
216 | #elif defined __hppa__ |
141 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
217 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
218 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
143 | #elif defined __ia64__ |
219 | #elif defined __ia64__ |
144 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
220 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
221 | #elif defined __m68k__ |
|
|
222 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
223 | #elif defined __m88k__ |
|
|
224 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
225 | #elif defined __sh__ |
|
|
226 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
145 | #endif |
227 | #endif |
146 | #endif |
228 | #endif |
147 | #endif |
229 | #endif |
148 | |
230 | |
149 | #ifndef ECB_MEMORY_FENCE |
231 | #ifndef ECB_MEMORY_FENCE |
150 | #if ECB_GCC_VERSION(4,7) |
232 | #if ECB_GCC_VERSION(4,7) |
151 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
233 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
152 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
234 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
153 | /*#elif defined __clang && __has_feature (cxx_atomic)*/ |
235 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
236 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
237 | #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
238 | |
|
|
239 | #elif ECB_CLANG_EXTENSION(c_atomic) |
154 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
240 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
155 | /*#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)*/ |
241 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
242 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
243 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
244 | #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
245 | |
156 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
246 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
157 | #define ECB_MEMORY_FENCE __sync_synchronize () |
247 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
248 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
|
|
249 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
250 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
251 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
252 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
253 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
158 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
254 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
159 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
255 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
160 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
256 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
161 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
257 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
162 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
258 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
163 | #elif defined _WIN32 |
259 | #elif defined _WIN32 |
164 | #include <WinNT.h> |
260 | #include <WinNT.h> |
165 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
261 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
166 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
262 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
167 | #include <mbarrier.h> |
263 | #include <mbarrier.h> |
168 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
264 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
169 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
265 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
170 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
266 | #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |
|
|
267 | #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () |
171 | #elif __xlC__ |
268 | #elif __xlC__ |
172 | #define ECB_MEMORY_FENCE __sync () |
269 | #define ECB_MEMORY_FENCE __sync () |
173 | #endif |
270 | #endif |
174 | #endif |
271 | #endif |
175 | |
272 | |
176 | #ifndef ECB_MEMORY_FENCE |
273 | #ifndef ECB_MEMORY_FENCE |
177 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
274 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
178 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
275 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
179 | /* not just C11 atomics and atomic accesses */ |
276 | /* not just C11 atomics and atomic accesses */ |
180 | #include <stdatomic.h> |
277 | #include <stdatomic.h> |
181 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
182 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
183 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
184 | /* for most usages, or gcc and clang have a bug */ |
|
|
185 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
186 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
187 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
278 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
279 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
280 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
188 | #endif |
281 | #endif |
189 | #endif |
282 | #endif |
190 | |
283 | |
191 | #ifndef ECB_MEMORY_FENCE |
284 | #ifndef ECB_MEMORY_FENCE |
192 | #if !ECB_AVOID_PTHREADS |
285 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
212 | |
305 | |
213 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
306 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
214 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
307 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
215 | #endif |
308 | #endif |
216 | |
309 | |
|
|
310 | #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE |
|
|
311 | #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ |
|
|
312 | #endif |
|
|
313 | |
217 | /*****************************************************************************/ |
314 | /*****************************************************************************/ |
218 | |
315 | |
219 | #if __cplusplus |
316 | #if ECB_CPP |
220 | #define ecb_inline static inline |
317 | #define ecb_inline static inline |
221 | #elif ECB_GCC_VERSION(2,5) |
318 | #elif ECB_GCC_VERSION(2,5) |
222 | #define ecb_inline static __inline__ |
319 | #define ecb_inline static __inline__ |
223 | #elif ECB_C99 |
320 | #elif ECB_C99 |
224 | #define ecb_inline static inline |
321 | #define ecb_inline static inline |
… | |
… | |
238 | |
335 | |
239 | #define ECB_CONCAT_(a, b) a ## b |
336 | #define ECB_CONCAT_(a, b) a ## b |
240 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
337 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
241 | #define ECB_STRINGIFY_(a) # a |
338 | #define ECB_STRINGIFY_(a) # a |
242 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
339 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
340 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
243 | |
341 | |
244 | #define ecb_function_ ecb_inline |
342 | #define ecb_function_ ecb_inline |
245 | |
343 | |
246 | #if ECB_GCC_VERSION(3,1) |
344 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
247 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
345 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
346 | #else |
|
|
347 | #define ecb_attribute(attrlist) |
|
|
348 | #endif |
|
|
349 | |
|
|
350 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
248 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
351 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
352 | #else |
|
|
353 | /* possible C11 impl for integral types |
|
|
354 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
355 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
356 | |
|
|
357 | #define ecb_is_constant(expr) 0 |
|
|
358 | #endif |
|
|
359 | |
|
|
360 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
249 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
361 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
362 | #else |
|
|
363 | #define ecb_expect(expr,value) (expr) |
|
|
364 | #endif |
|
|
365 | |
|
|
366 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
250 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
367 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
251 | #else |
368 | #else |
252 | #define ecb_attribute(attrlist) |
|
|
253 | #define ecb_is_constant(expr) 0 |
|
|
254 | #define ecb_expect(expr,value) (expr) |
|
|
255 | #define ecb_prefetch(addr,rw,locality) |
369 | #define ecb_prefetch(addr,rw,locality) |
256 | #endif |
370 | #endif |
257 | |
371 | |
258 | /* no emulation for ecb_decltype */ |
372 | /* no emulation for ecb_decltype */ |
259 | #if ECB_GCC_VERSION(4,5) |
373 | #if ECB_CPP11 |
|
|
374 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
375 | template<class T> struct ecb_decltype_t { typedef T type; }; |
260 | #define ecb_decltype(x) __decltype(x) |
376 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
261 | #elif ECB_GCC_VERSION(3,0) |
377 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
262 | #define ecb_decltype(x) __typeof(x) |
378 | #define ecb_decltype(x) __typeof__ (x) |
263 | #endif |
379 | #endif |
264 | |
380 | |
|
|
381 | #if _MSC_VER >= 1300 |
|
|
382 | #define ecb_deprecated __declspec (deprecated) |
|
|
383 | #else |
|
|
384 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
|
|
385 | #endif |
|
|
386 | |
|
|
387 | #if _MSC_VER >= 1500 |
|
|
388 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
|
|
389 | #elif ECB_GCC_VERSION(4,5) |
|
|
390 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
|
|
391 | #else |
|
|
392 | #define ecb_deprecated_message(msg) ecb_deprecated |
|
|
393 | #endif |
|
|
394 | |
|
|
395 | #if _MSC_VER >= 1400 |
|
|
396 | #define ecb_noinline __declspec (noinline) |
|
|
397 | #else |
265 | #define ecb_noinline ecb_attribute ((__noinline__)) |
398 | #define ecb_noinline ecb_attribute ((__noinline__)) |
|
|
399 | #endif |
|
|
400 | |
266 | #define ecb_unused ecb_attribute ((__unused__)) |
401 | #define ecb_unused ecb_attribute ((__unused__)) |
267 | #define ecb_const ecb_attribute ((__const__)) |
402 | #define ecb_const ecb_attribute ((__const__)) |
268 | #define ecb_pure ecb_attribute ((__pure__)) |
403 | #define ecb_pure ecb_attribute ((__pure__)) |
269 | |
404 | |
270 | #if ECB_C11 |
405 | #if ECB_C11 || __IBMC_NORETURN |
|
|
406 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
271 | #define ecb_noreturn _Noreturn |
407 | #define ecb_noreturn _Noreturn |
|
|
408 | #elif ECB_CPP11 |
|
|
409 | #define ecb_noreturn [[noreturn]] |
|
|
410 | #elif _MSC_VER >= 1200 |
|
|
411 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
|
|
412 | #define ecb_noreturn __declspec (noreturn) |
272 | #else |
413 | #else |
273 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
414 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
274 | #endif |
415 | #endif |
275 | |
416 | |
276 | #if ECB_GCC_VERSION(4,3) |
417 | #if ECB_GCC_VERSION(4,3) |
… | |
… | |
291 | /* for compatibility to the rest of the world */ |
432 | /* for compatibility to the rest of the world */ |
292 | #define ecb_likely(expr) ecb_expect_true (expr) |
433 | #define ecb_likely(expr) ecb_expect_true (expr) |
293 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
434 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
294 | |
435 | |
295 | /* count trailing zero bits and count # of one bits */ |
436 | /* count trailing zero bits and count # of one bits */ |
296 | #if ECB_GCC_VERSION(3,4) |
437 | #if ECB_GCC_VERSION(3,4) \ |
|
|
438 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
439 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
440 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
297 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
441 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
298 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
442 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
299 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
443 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
300 | #define ecb_ctz32(x) __builtin_ctz (x) |
444 | #define ecb_ctz32(x) __builtin_ctz (x) |
301 | #define ecb_ctz64(x) __builtin_ctzll (x) |
445 | #define ecb_ctz64(x) __builtin_ctzll (x) |
302 | #define ecb_popcount32(x) __builtin_popcount (x) |
446 | #define ecb_popcount32(x) __builtin_popcount (x) |
303 | /* no popcountll */ |
447 | /* no popcountll */ |
304 | #else |
448 | #else |
305 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
449 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
306 | ecb_function_ int |
450 | ecb_function_ ecb_const int |
307 | ecb_ctz32 (uint32_t x) |
451 | ecb_ctz32 (uint32_t x) |
308 | { |
452 | { |
|
|
453 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
454 | unsigned long r; |
|
|
455 | _BitScanForward (&r, x); |
|
|
456 | return (int)r; |
|
|
457 | #else |
309 | int r = 0; |
458 | int r = 0; |
310 | |
459 | |
311 | x &= ~x + 1; /* this isolates the lowest bit */ |
460 | x &= ~x + 1; /* this isolates the lowest bit */ |
312 | |
461 | |
313 | #if ECB_branchless_on_i386 |
462 | #if ECB_branchless_on_i386 |
… | |
… | |
323 | if (x & 0xff00ff00) r += 8; |
472 | if (x & 0xff00ff00) r += 8; |
324 | if (x & 0xffff0000) r += 16; |
473 | if (x & 0xffff0000) r += 16; |
325 | #endif |
474 | #endif |
326 | |
475 | |
327 | return r; |
476 | return r; |
|
|
477 | #endif |
328 | } |
478 | } |
329 | |
479 | |
330 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
480 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
331 | ecb_function_ int |
481 | ecb_function_ ecb_const int |
332 | ecb_ctz64 (uint64_t x) |
482 | ecb_ctz64 (uint64_t x) |
333 | { |
483 | { |
|
|
484 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
485 | unsigned long r; |
|
|
486 | _BitScanForward64 (&r, x); |
|
|
487 | return (int)r; |
|
|
488 | #else |
334 | int shift = x & 0xffffffffU ? 0 : 32; |
489 | int shift = x & 0xffffffff ? 0 : 32; |
335 | return ecb_ctz32 (x >> shift) + shift; |
490 | return ecb_ctz32 (x >> shift) + shift; |
|
|
491 | #endif |
336 | } |
492 | } |
337 | |
493 | |
338 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
494 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
339 | ecb_function_ int |
495 | ecb_function_ ecb_const int |
340 | ecb_popcount32 (uint32_t x) |
496 | ecb_popcount32 (uint32_t x) |
341 | { |
497 | { |
342 | x -= (x >> 1) & 0x55555555; |
498 | x -= (x >> 1) & 0x55555555; |
343 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
499 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
344 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
500 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
345 | x *= 0x01010101; |
501 | x *= 0x01010101; |
346 | |
502 | |
347 | return x >> 24; |
503 | return x >> 24; |
348 | } |
504 | } |
349 | |
505 | |
350 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
506 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
351 | ecb_function_ int ecb_ld32 (uint32_t x) |
507 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
352 | { |
508 | { |
|
|
509 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
510 | unsigned long r; |
|
|
511 | _BitScanReverse (&r, x); |
|
|
512 | return (int)r; |
|
|
513 | #else |
353 | int r = 0; |
514 | int r = 0; |
354 | |
515 | |
355 | if (x >> 16) { x >>= 16; r += 16; } |
516 | if (x >> 16) { x >>= 16; r += 16; } |
356 | if (x >> 8) { x >>= 8; r += 8; } |
517 | if (x >> 8) { x >>= 8; r += 8; } |
357 | if (x >> 4) { x >>= 4; r += 4; } |
518 | if (x >> 4) { x >>= 4; r += 4; } |
358 | if (x >> 2) { x >>= 2; r += 2; } |
519 | if (x >> 2) { x >>= 2; r += 2; } |
359 | if (x >> 1) { r += 1; } |
520 | if (x >> 1) { r += 1; } |
360 | |
521 | |
361 | return r; |
522 | return r; |
|
|
523 | #endif |
362 | } |
524 | } |
363 | |
525 | |
364 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
526 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
365 | ecb_function_ int ecb_ld64 (uint64_t x) |
527 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
366 | { |
528 | { |
|
|
529 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
530 | unsigned long r; |
|
|
531 | _BitScanReverse64 (&r, x); |
|
|
532 | return (int)r; |
|
|
533 | #else |
367 | int r = 0; |
534 | int r = 0; |
368 | |
535 | |
369 | if (x >> 32) { x >>= 32; r += 32; } |
536 | if (x >> 32) { x >>= 32; r += 32; } |
370 | |
537 | |
371 | return r + ecb_ld32 (x); |
538 | return r + ecb_ld32 (x); |
372 | } |
|
|
373 | #endif |
539 | #endif |
|
|
540 | } |
|
|
541 | #endif |
374 | |
542 | |
375 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
543 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
376 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
544 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
377 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
545 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
378 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
546 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
379 | |
547 | |
380 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
548 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
381 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
549 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
382 | { |
550 | { |
383 | return ( (x * 0x0802U & 0x22110U) |
551 | return ( (x * 0x0802U & 0x22110U) |
384 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
552 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
385 | } |
553 | } |
386 | |
554 | |
387 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
555 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
388 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
556 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
389 | { |
557 | { |
390 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
558 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
391 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
559 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
392 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
560 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
393 | x = ( x >> 8 ) | ( x << 8); |
561 | x = ( x >> 8 ) | ( x << 8); |
394 | |
562 | |
395 | return x; |
563 | return x; |
396 | } |
564 | } |
397 | |
565 | |
398 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
566 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
399 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
567 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
400 | { |
568 | { |
401 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
569 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
402 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
570 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
403 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
571 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
404 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
572 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
… | |
… | |
407 | return x; |
575 | return x; |
408 | } |
576 | } |
409 | |
577 | |
410 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
578 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
411 | /* so for this version we are lazy */ |
579 | /* so for this version we are lazy */ |
412 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
580 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
413 | ecb_function_ int |
581 | ecb_function_ ecb_const int |
414 | ecb_popcount64 (uint64_t x) |
582 | ecb_popcount64 (uint64_t x) |
415 | { |
583 | { |
416 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
584 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
417 | } |
585 | } |
418 | |
586 | |
419 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
587 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
420 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
588 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
421 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
589 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
422 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
590 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
423 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
591 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
424 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
592 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
425 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
593 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
426 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
594 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
427 | |
595 | |
428 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
596 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
429 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
597 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
430 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
598 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
431 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
599 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
432 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
600 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
433 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
601 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
434 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
602 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
435 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
603 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
436 | |
604 | |
437 | #if ECB_GCC_VERSION(4,3) |
605 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
|
|
606 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
|
|
607 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
|
|
608 | #else |
438 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
609 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
610 | #endif |
439 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
611 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
440 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
612 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
|
|
613 | #elif _MSC_VER |
|
|
614 | #include <stdlib.h> |
|
|
615 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
|
|
616 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
|
|
617 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
441 | #else |
618 | #else |
442 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
619 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
443 | ecb_function_ uint16_t |
620 | ecb_function_ ecb_const uint16_t |
444 | ecb_bswap16 (uint16_t x) |
621 | ecb_bswap16 (uint16_t x) |
445 | { |
622 | { |
446 | return ecb_rotl16 (x, 8); |
623 | return ecb_rotl16 (x, 8); |
447 | } |
624 | } |
448 | |
625 | |
449 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
626 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
450 | ecb_function_ uint32_t |
627 | ecb_function_ ecb_const uint32_t |
451 | ecb_bswap32 (uint32_t x) |
628 | ecb_bswap32 (uint32_t x) |
452 | { |
629 | { |
453 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
630 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
454 | } |
631 | } |
455 | |
632 | |
456 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
633 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
457 | ecb_function_ uint64_t |
634 | ecb_function_ ecb_const uint64_t |
458 | ecb_bswap64 (uint64_t x) |
635 | ecb_bswap64 (uint64_t x) |
459 | { |
636 | { |
460 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
637 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
461 | } |
638 | } |
462 | #endif |
639 | #endif |
463 | |
640 | |
464 | #if ECB_GCC_VERSION(4,5) |
641 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
465 | #define ecb_unreachable() __builtin_unreachable () |
642 | #define ecb_unreachable() __builtin_unreachable () |
466 | #else |
643 | #else |
467 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
644 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
468 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
645 | ecb_inline ecb_noreturn void ecb_unreachable (void); |
469 | ecb_inline void ecb_unreachable (void) { } |
646 | ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
470 | #endif |
647 | #endif |
471 | |
648 | |
472 | /* try to tell the compiler that some condition is definitely true */ |
649 | /* try to tell the compiler that some condition is definitely true */ |
473 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
650 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
474 | |
651 | |
475 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
652 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
476 | ecb_inline unsigned char |
653 | ecb_inline ecb_const uint32_t |
477 | ecb_byteorder_helper (void) |
654 | ecb_byteorder_helper (void) |
478 | { |
655 | { |
479 | /* the union code still generates code under pressure in gcc, */ |
656 | /* the union code still generates code under pressure in gcc, */ |
480 | /* but less than using pointers, and always seem to */ |
657 | /* but less than using pointers, and always seems to */ |
481 | /* successfully return a constant. */ |
658 | /* successfully return a constant. */ |
482 | /* the reason why we have this horrible preprocessor mess */ |
659 | /* the reason why we have this horrible preprocessor mess */ |
483 | /* is to avoid it in all cases, at least on common architectures */ |
660 | /* is to avoid it in all cases, at least on common architectures */ |
484 | /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ |
661 | /* or when using a recent enough gcc version (>= 4.6) */ |
485 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
486 | return 0x44; |
|
|
487 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
662 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
|
|
663 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
|
|
664 | #define ECB_LITTLE_ENDIAN 1 |
488 | return 0x44; |
665 | return 0x44332211; |
489 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
666 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
|
|
667 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
|
|
668 | #define ECB_BIG_ENDIAN 1 |
490 | retrurn 0x11; |
669 | return 0x11223344; |
491 | #else |
670 | #else |
492 | union |
671 | union |
493 | { |
672 | { |
|
|
673 | uint8_t c[4]; |
494 | uint32_t i; |
674 | uint32_t u; |
495 | uint8_t c; |
|
|
496 | } u = { 0x11223344 }; |
675 | } u = { 0x11, 0x22, 0x33, 0x44 }; |
497 | return u.c; |
676 | return u.u; |
498 | #endif |
677 | #endif |
499 | } |
678 | } |
500 | |
679 | |
501 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
680 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
502 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
681 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
503 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
682 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
504 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
683 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
505 | |
684 | |
506 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
685 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
507 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
686 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
508 | #else |
687 | #else |
509 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
688 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
510 | #endif |
689 | #endif |
511 | |
690 | |
512 | #if __cplusplus |
691 | #if ECB_CPP |
513 | template<typename T> |
692 | template<typename T> |
514 | static inline T ecb_div_rd (T val, T div) |
693 | static inline T ecb_div_rd (T val, T div) |
515 | { |
694 | { |
516 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
695 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
517 | } |
696 | } |
… | |
… | |
534 | } |
713 | } |
535 | #else |
714 | #else |
536 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
715 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
537 | #endif |
716 | #endif |
538 | |
717 | |
|
|
718 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
|
|
719 | ecb_function_ ecb_const uint32_t |
|
|
720 | ecb_binary16_to_binary32 (uint32_t x) |
|
|
721 | { |
|
|
722 | unsigned int s = (x & 0x8000) << (31 - 15); |
|
|
723 | int e = (x >> 10) & 0x001f; |
|
|
724 | unsigned int m = x & 0x03ff; |
|
|
725 | |
|
|
726 | if (ecb_expect_false (e == 31)) |
|
|
727 | /* infinity or NaN */ |
|
|
728 | e = 255 - (127 - 15); |
|
|
729 | else if (ecb_expect_false (!e)) |
|
|
730 | { |
|
|
731 | if (ecb_expect_true (!m)) |
|
|
732 | /* zero, handled by code below by forcing e to 0 */ |
|
|
733 | e = 0 - (127 - 15); |
|
|
734 | else |
|
|
735 | { |
|
|
736 | /* subnormal, renormalise */ |
|
|
737 | unsigned int s = 10 - ecb_ld32 (m); |
|
|
738 | |
|
|
739 | m = (m << s) & 0x3ff; /* mask implicit bit */ |
|
|
740 | e -= s - 1; |
|
|
741 | } |
|
|
742 | } |
|
|
743 | |
|
|
744 | /* e and m now are normalised, or zero, (or inf or nan) */ |
|
|
745 | e += 127 - 15; |
|
|
746 | |
|
|
747 | return s | (e << 23) | (m << (23 - 10)); |
|
|
748 | } |
|
|
749 | |
|
|
750 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
|
|
751 | ecb_function_ ecb_const uint16_t |
|
|
752 | ecb_binary32_to_binary16 (uint32_t x) |
|
|
753 | { |
|
|
754 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
|
|
755 | unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
|
|
756 | unsigned int m = x & 0x007fffff; |
|
|
757 | |
|
|
758 | x &= 0x7fffffff; |
|
|
759 | |
|
|
760 | /* if it's within range of binary16 normals, use fast path */ |
|
|
761 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
|
|
762 | { |
|
|
763 | /* mantissa round-to-even */ |
|
|
764 | m += 0x00000fff + ((m >> (23 - 10)) & 1); |
|
|
765 | |
|
|
766 | /* handle overflow */ |
|
|
767 | if (ecb_expect_false (m >= 0x00800000)) |
|
|
768 | { |
|
|
769 | m >>= 1; |
|
|
770 | e += 1; |
|
|
771 | } |
|
|
772 | |
|
|
773 | return s | (e << 10) | (m >> (23 - 10)); |
|
|
774 | } |
|
|
775 | |
|
|
776 | /* handle large numbers and infinity */ |
|
|
777 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
|
|
778 | return s | 0x7c00; |
|
|
779 | |
|
|
780 | /* handle zero, subnormals and small numbers */ |
|
|
781 | if (ecb_expect_true (x < 0x38800000)) |
|
|
782 | { |
|
|
783 | /* zero */ |
|
|
784 | if (ecb_expect_true (!x)) |
|
|
785 | return s; |
|
|
786 | |
|
|
787 | /* handle subnormals */ |
|
|
788 | |
|
|
789 | /* too small, will be zero */ |
|
|
790 | if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
|
|
791 | return s; |
|
|
792 | |
|
|
793 | m |= 0x00800000; /* make implicit bit explicit */ |
|
|
794 | |
|
|
795 | /* very tricky - we need to round to the nearest e (+10) bit value */ |
|
|
796 | { |
|
|
797 | unsigned int bits = 14 - e; |
|
|
798 | unsigned int half = (1 << (bits - 1)) - 1; |
|
|
799 | unsigned int even = (m >> bits) & 1; |
|
|
800 | |
|
|
801 | /* if this overflows, we will end up with a normalised number */ |
|
|
802 | m = (m + half + even) >> bits; |
|
|
803 | } |
|
|
804 | |
|
|
805 | return s | m; |
|
|
806 | } |
|
|
807 | |
|
|
808 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
|
|
809 | m >>= 13; |
|
|
810 | |
|
|
811 | return s | 0x7c00 | m | !m; |
|
|
812 | } |
|
|
813 | |
539 | /*******************************************************************************/ |
814 | /*******************************************************************************/ |
540 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
815 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
541 | |
816 | |
542 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
817 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
543 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
818 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
544 | #if 0 \ |
819 | #if 0 \ |
545 | || __i386 || __i386__ \ |
820 | || __i386 || __i386__ \ |
546 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
821 | || ECB_GCC_AMD64 \ |
547 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
822 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
548 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
549 | || defined __s390__ || defined __s390x__ \ |
823 | || defined __s390__ || defined __s390x__ \ |
550 | || defined __mips__ \ |
824 | || defined __mips__ \ |
551 | || defined __alpha__ \ |
825 | || defined __alpha__ \ |
552 | || defined __hppa__ \ |
826 | || defined __hppa__ \ |
553 | || defined __ia64__ \ |
827 | || defined __ia64__ \ |
|
|
828 | || defined __m68k__ \ |
|
|
829 | || defined __m88k__ \ |
|
|
830 | || defined __sh__ \ |
554 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
831 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
|
|
832 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
833 | || defined __aarch64__ |
555 | #define ECB_STDFP 1 |
834 | #define ECB_STDFP 1 |
556 | #include <string.h> /* for memcpy */ |
835 | #include <string.h> /* for memcpy */ |
557 | #else |
836 | #else |
558 | #define ECB_STDFP 0 |
837 | #define ECB_STDFP 0 |
559 | #include <math.h> /* for frexp*, ldexp* */ |
|
|
560 | #endif |
838 | #endif |
561 | |
839 | |
562 | #ifndef ECB_NO_LIBM |
840 | #ifndef ECB_NO_LIBM |
563 | |
841 | |
|
|
842 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
843 | |
|
|
844 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
845 | #ifdef INFINITY |
|
|
846 | #define ECB_INFINITY INFINITY |
|
|
847 | #else |
|
|
848 | #define ECB_INFINITY HUGE_VAL |
|
|
849 | #endif |
|
|
850 | |
|
|
851 | #ifdef NAN |
|
|
852 | #define ECB_NAN NAN |
|
|
853 | #else |
|
|
854 | #define ECB_NAN ECB_INFINITY |
|
|
855 | #endif |
|
|
856 | |
|
|
857 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
|
|
858 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
859 | #define ecb_frexpf(x,e) frexpf ((x), (e)) |
|
|
860 | #else |
|
|
861 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
|
|
862 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
|
|
863 | #endif |
|
|
864 | |
564 | /* convert a float to ieee single/binary32 */ |
865 | /* convert a float to ieee single/binary32 */ |
565 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
866 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
566 | ecb_function_ uint32_t |
867 | ecb_function_ ecb_const uint32_t |
567 | ecb_float_to_binary32 (float x) |
868 | ecb_float_to_binary32 (float x) |
568 | { |
869 | { |
569 | uint32_t r; |
870 | uint32_t r; |
570 | |
871 | |
571 | #if ECB_STDFP |
872 | #if ECB_STDFP |
… | |
… | |
578 | if (x == 0e0f ) return 0x00000000U; |
879 | if (x == 0e0f ) return 0x00000000U; |
579 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
880 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
580 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
881 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
581 | if (x != x ) return 0x7fbfffffU; |
882 | if (x != x ) return 0x7fbfffffU; |
582 | |
883 | |
583 | m = frexpf (x, &e) * 0x1000000U; |
884 | m = ecb_frexpf (x, &e) * 0x1000000U; |
584 | |
885 | |
585 | r = m & 0x80000000U; |
886 | r = m & 0x80000000U; |
586 | |
887 | |
587 | if (r) |
888 | if (r) |
588 | m = -m; |
889 | m = -m; |
… | |
… | |
600 | |
901 | |
601 | return r; |
902 | return r; |
602 | } |
903 | } |
603 | |
904 | |
604 | /* converts an ieee single/binary32 to a float */ |
905 | /* converts an ieee single/binary32 to a float */ |
605 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
906 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
606 | ecb_function_ float |
907 | ecb_function_ ecb_const float |
607 | ecb_binary32_to_float (uint32_t x) |
908 | ecb_binary32_to_float (uint32_t x) |
608 | { |
909 | { |
609 | float r; |
910 | float r; |
610 | |
911 | |
611 | #if ECB_STDFP |
912 | #if ECB_STDFP |
… | |
… | |
621 | x |= 0x800000U; |
922 | x |= 0x800000U; |
622 | else |
923 | else |
623 | e = 1; |
924 | e = 1; |
624 | |
925 | |
625 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
926 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
626 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
927 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
627 | |
928 | |
628 | r = neg ? -r : r; |
929 | r = neg ? -r : r; |
629 | #endif |
930 | #endif |
630 | |
931 | |
631 | return r; |
932 | return r; |
632 | } |
933 | } |
633 | |
934 | |
634 | /* convert a double to ieee double/binary64 */ |
935 | /* convert a double to ieee double/binary64 */ |
635 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
936 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
636 | ecb_function_ uint64_t |
937 | ecb_function_ ecb_const uint64_t |
637 | ecb_double_to_binary64 (double x) |
938 | ecb_double_to_binary64 (double x) |
638 | { |
939 | { |
639 | uint64_t r; |
940 | uint64_t r; |
640 | |
941 | |
641 | #if ECB_STDFP |
942 | #if ECB_STDFP |
… | |
… | |
670 | |
971 | |
671 | return r; |
972 | return r; |
672 | } |
973 | } |
673 | |
974 | |
674 | /* converts an ieee double/binary64 to a double */ |
975 | /* converts an ieee double/binary64 to a double */ |
675 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
976 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
676 | ecb_function_ double |
977 | ecb_function_ ecb_const double |
677 | ecb_binary64_to_double (uint64_t x) |
978 | ecb_binary64_to_double (uint64_t x) |
678 | { |
979 | { |
679 | double r; |
980 | double r; |
680 | |
981 | |
681 | #if ECB_STDFP |
982 | #if ECB_STDFP |
… | |
… | |
699 | #endif |
1000 | #endif |
700 | |
1001 | |
701 | return r; |
1002 | return r; |
702 | } |
1003 | } |
703 | |
1004 | |
704 | #endif |
1005 | /* convert a float to ieee half/binary16 */ |
|
|
1006 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
|
|
1007 | ecb_function_ ecb_const uint16_t |
|
|
1008 | ecb_float_to_binary16 (float x) |
|
|
1009 | { |
|
|
1010 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
|
|
1011 | } |
705 | |
1012 | |
706 | #endif |
1013 | /* convert an ieee half/binary16 to float */ |
|
|
1014 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
1015 | ecb_function_ ecb_const float |
|
|
1016 | ecb_binary16_to_float (uint16_t x) |
|
|
1017 | { |
|
|
1018 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
|
|
1019 | } |
707 | |
1020 | |
|
|
1021 | #endif |
|
|
1022 | |
|
|
1023 | #endif |
|
|
1024 | |