| 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 (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de> |
| 5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
| 6 | * All rights reserved. |
6 | * All rights reserved. |
| 7 | * |
7 | * |
| 8 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
| 9 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
| … | |
… | |
| 23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
| 24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
| 25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
| 26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
| 27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
28 | * |
|
|
29 | * Alternatively, the contents of this file may be used under the terms of |
|
|
30 | * the GNU General Public License ("GPL") version 2 or any later version, |
|
|
31 | * in which case the provisions of the GPL are applicable instead of |
|
|
32 | * the above. If you wish to allow the use of your version of this file |
|
|
33 | * only under the terms of the GPL and not to allow others to use your |
|
|
34 | * version of this file under the BSD license, indicate your decision |
|
|
35 | * by deleting the provisions above and replace them with the notice |
|
|
36 | * and other provisions required by the GPL. If you do not delete the |
|
|
37 | * provisions above, a recipient may use your version of this file under |
|
|
38 | * either the BSD or the GPL. |
| 28 | */ |
39 | */ |
| 29 | |
40 | |
| 30 | #ifndef ECB_H |
41 | #ifndef ECB_H |
| 31 | #define ECB_H |
42 | #define ECB_H |
| 32 | |
43 | |
| 33 | /* 16 bits major, 16 bits minor */ |
44 | /* 16 bits major, 16 bits minor */ |
| 34 | #define ECB_VERSION 0x00010001 |
45 | #define ECB_VERSION 0x00010009 |
| 35 | |
46 | |
| 36 | #ifdef _WIN32 |
47 | #include <string.h> /* for memcpy */ |
|
|
48 | |
|
|
49 | #if defined (_WIN32) && !defined (__MINGW32__) |
| 37 | typedef signed char int8_t; |
50 | typedef signed char int8_t; |
| 38 | typedef unsigned char uint8_t; |
51 | typedef unsigned char uint8_t; |
|
|
52 | typedef signed char int_fast8_t; |
|
|
53 | typedef unsigned char uint_fast8_t; |
| 39 | typedef signed short int16_t; |
54 | typedef signed short int16_t; |
| 40 | typedef unsigned short uint16_t; |
55 | typedef unsigned short uint16_t; |
|
|
56 | typedef signed int int_fast16_t; |
|
|
57 | typedef unsigned int uint_fast16_t; |
| 41 | typedef signed int int32_t; |
58 | typedef signed int int32_t; |
| 42 | typedef unsigned int uint32_t; |
59 | typedef unsigned int uint32_t; |
|
|
60 | typedef signed int int_fast32_t; |
|
|
61 | typedef unsigned int uint_fast32_t; |
| 43 | #if __GNUC__ |
62 | #if __GNUC__ |
| 44 | typedef signed long long int64_t; |
63 | typedef signed long long int64_t; |
| 45 | typedef unsigned long long uint64_t; |
64 | typedef unsigned long long uint64_t; |
| 46 | #else /* _MSC_VER || __BORLANDC__ */ |
65 | #else /* _MSC_VER || __BORLANDC__ */ |
| 47 | typedef signed __int64 int64_t; |
66 | typedef signed __int64 int64_t; |
| 48 | typedef unsigned __int64 uint64_t; |
67 | typedef unsigned __int64 uint64_t; |
| 49 | #endif |
68 | #endif |
|
|
69 | typedef int64_t int_fast64_t; |
|
|
70 | typedef uint64_t uint_fast64_t; |
| 50 | #ifdef _WIN64 |
71 | #ifdef _WIN64 |
| 51 | #define ECB_PTRSIZE 8 |
72 | #define ECB_PTRSIZE 8 |
| 52 | typedef uint64_t uintptr_t; |
73 | typedef uint64_t uintptr_t; |
| 53 | typedef int64_t intptr_t; |
74 | typedef int64_t intptr_t; |
| 54 | #else |
75 | #else |
| 55 | #define ECB_PTRSIZE 4 |
76 | #define ECB_PTRSIZE 4 |
| 56 | typedef uint32_t uintptr_t; |
77 | typedef uint32_t uintptr_t; |
| 57 | typedef int32_t intptr_t; |
78 | typedef int32_t intptr_t; |
| 58 | #endif |
79 | #endif |
| 59 | typedef intptr_t ptrdiff_t; |
|
|
| 60 | #else |
80 | #else |
| 61 | #include <inttypes.h> |
81 | #include <inttypes.h> |
| 62 | #if UINTMAX_MAX > 0xffffffffU |
82 | #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
| 63 | #define ECB_PTRSIZE 8 |
83 | #define ECB_PTRSIZE 8 |
| 64 | #else |
84 | #else |
| 65 | #define ECB_PTRSIZE 4 |
85 | #define ECB_PTRSIZE 4 |
| 66 | #endif |
86 | #endif |
|
|
87 | #endif |
|
|
88 | |
|
|
89 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
|
|
90 | #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
|
|
91 | |
|
|
92 | #ifndef ECB_OPTIMIZE_SIZE |
|
|
93 | #if __OPTIMIZE_SIZE__ |
|
|
94 | #define ECB_OPTIMIZE_SIZE 1 |
|
|
95 | #else |
|
|
96 | #define ECB_OPTIMIZE_SIZE 0 |
|
|
97 | #endif |
|
|
98 | #endif |
|
|
99 | |
|
|
100 | /* work around x32 idiocy by defining proper macros */ |
|
|
101 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
|
|
102 | #if _ILP32 |
|
|
103 | #define ECB_AMD64_X32 1 |
|
|
104 | #else |
|
|
105 | #define ECB_AMD64 1 |
|
|
106 | #endif |
|
|
107 | #endif |
|
|
108 | |
|
|
109 | #if ECB_PTRSIZE >= 8 || ECB_AMD64_X32 |
|
|
110 | #define ECB_64BIT_NATIVE 1 |
|
|
111 | #else |
|
|
112 | #define ECB_64BIT_NATIVE 0 |
| 67 | #endif |
113 | #endif |
| 68 | |
114 | |
| 69 | /* many compilers define _GNUC_ to some versions but then only implement |
115 | /* many compilers define _GNUC_ to some versions but then only implement |
| 70 | * what their idiot authors think are the "more important" extensions, |
116 | * what their idiot authors think are the "more important" extensions, |
| 71 | * causing enormous grief in return for some better fake benchmark numbers. |
117 | * causing enormous grief in return for some better fake benchmark numbers. |
| 72 | * or so. |
118 | * or so. |
| 73 | * we try to detect these and simply assume they are not gcc - if they have |
119 | * we try to detect these and simply assume they are not gcc - if they have |
| 74 | * an issue with that they should have done it right in the first place. |
120 | * an issue with that they should have done it right in the first place. |
| 75 | */ |
121 | */ |
| 76 | #ifndef ECB_GCC_VERSION |
|
|
| 77 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
122 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
| 78 | #define ECB_GCC_VERSION(major,minor) 0 |
123 | #define ECB_GCC_VERSION(major,minor) 0 |
| 79 | #else |
124 | #else |
| 80 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
125 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 81 | #endif |
126 | #endif |
| 82 | #endif |
|
|
| 83 | |
127 | |
| 84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
128 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
| 85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
129 | |
| 86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
130 | #if __clang__ && defined __has_builtin |
|
|
131 | #define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
|
|
132 | #else |
|
|
133 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
134 | #endif |
|
|
135 | |
|
|
136 | #if __clang__ && defined __has_extension |
|
|
137 | #define ECB_CLANG_EXTENSION(x) __has_extension (x) |
|
|
138 | #else |
|
|
139 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
140 | #endif |
|
|
141 | |
| 87 | #define ECB_CPP (__cplusplus+0) |
142 | #define ECB_CPP (__cplusplus+0) |
| 88 | #define ECB_CPP98 (__cplusplus >= 199711L) |
|
|
| 89 | #define ECB_CPP11 (__cplusplus >= 201103L) |
143 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
144 | #define ECB_CPP14 (__cplusplus >= 201402L) |
|
|
145 | #define ECB_CPP17 (__cplusplus >= 201703L) |
|
|
146 | |
|
|
147 | #if ECB_CPP |
|
|
148 | #define ECB_C 0 |
|
|
149 | #define ECB_STDC_VERSION 0 |
|
|
150 | #else |
|
|
151 | #define ECB_C 1 |
|
|
152 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
153 | #endif |
|
|
154 | |
|
|
155 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
156 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
157 | #define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
|
|
158 | |
|
|
159 | #if ECB_CPP |
|
|
160 | #define ECB_EXTERN_C extern "C" |
|
|
161 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
|
|
162 | #define ECB_EXTERN_C_END } |
|
|
163 | #else |
|
|
164 | #define ECB_EXTERN_C extern |
|
|
165 | #define ECB_EXTERN_C_BEG |
|
|
166 | #define ECB_EXTERN_C_END |
|
|
167 | #endif |
| 90 | |
168 | |
| 91 | /*****************************************************************************/ |
169 | /*****************************************************************************/ |
| 92 | |
170 | |
| 93 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
171 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
| 94 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
172 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
| 95 | |
173 | |
| 96 | #if ECB_NO_THREADS |
174 | #if ECB_NO_THREADS |
| 97 | # define ECB_NO_SMP 1 |
175 | #define ECB_NO_SMP 1 |
| 98 | #endif |
176 | #endif |
| 99 | |
177 | |
| 100 | #if ECB_NO_THREADS || ECB_NO_SMP |
178 | #if ECB_NO_SMP |
| 101 | #define ECB_MEMORY_FENCE do { } while (0) |
179 | #define ECB_MEMORY_FENCE do { } while (0) |
| 102 | #endif |
180 | #endif |
| 103 | |
181 | |
| 104 | #ifndef ECB_MEMORY_FENCE |
182 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
| 105 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
183 | #if __xlC__ && ECB_CPP |
| 106 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
184 | #include <builtins.h> |
| 107 | /* not just C11 atomics and atomic accesses */ |
|
|
| 108 | #include <stdatomic.h> |
|
|
| 109 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_acq_rel) |
|
|
| 110 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
| 111 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
|
|
| 112 | #endif |
185 | #endif |
| 113 | #endif |
|
|
| 114 | |
186 | |
| 115 | #ifndef ECB_MEMORY_FENCE_RELEASE |
187 | #if 1400 <= _MSC_VER |
| 116 | #if ECB_GCC_VERSION(4,7) |
188 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
| 117 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_ACQ_REL) |
|
|
| 118 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
| 119 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
| 120 | #endif |
|
|
| 121 | #endif |
189 | #endif |
| 122 | |
190 | |
| 123 | #ifndef ECB_MEMORY_FENCE |
191 | #ifndef ECB_MEMORY_FENCE |
| 124 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
192 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
193 | #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") |
| 125 | #if __i386 || __i386__ |
194 | #if __i386 || __i386__ |
| 126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
195 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
| 127 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
196 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 128 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
197 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 129 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
198 | #elif ECB_GCC_AMD64 |
| 130 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
199 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 131 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
200 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 132 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
201 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 133 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
202 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
| 134 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
203 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
204 | #elif defined __ARM_ARCH_2__ \ |
|
|
205 | || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
|
|
206 | || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
|
|
207 | || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
|
|
208 | || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
|
|
209 | || defined __ARM_ARCH_5TEJ__ |
|
|
210 | /* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
| 135 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
211 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
| 136 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
212 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
|
|
213 | || defined __ARM_ARCH_6T2__ |
| 137 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
214 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
| 138 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
215 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
| 139 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
216 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
| 140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
217 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
| 141 | #elif __sparc || __sparc__ |
218 | #elif __aarch64__ |
|
|
219 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
|
|
220 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
| 142 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
221 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
| 143 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
222 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
| 144 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
223 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
| 145 | #elif defined __s390__ || defined __s390x__ |
224 | #elif defined __s390__ || defined __s390x__ |
| 146 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
225 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
| 147 | #elif defined __mips__ |
226 | #elif defined __mips__ |
|
|
227 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
228 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
| 148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
229 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
| 149 | #elif defined __alpha__ |
230 | #elif defined __alpha__ |
| 150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
231 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
232 | #elif defined __hppa__ |
|
|
233 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
234 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
235 | #elif defined __ia64__ |
|
|
236 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
237 | #elif defined __m68k__ |
|
|
238 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
239 | #elif defined __m88k__ |
|
|
240 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
241 | #elif defined __sh__ |
|
|
242 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 151 | #endif |
243 | #endif |
| 152 | #endif |
244 | #endif |
| 153 | #endif |
245 | #endif |
| 154 | |
246 | |
| 155 | #ifndef ECB_MEMORY_FENCE |
247 | #ifndef ECB_MEMORY_FENCE |
|
|
248 | #if ECB_GCC_VERSION(4,7) |
|
|
249 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
250 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
251 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
252 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
253 | #undef ECB_MEMORY_FENCE_RELAXED |
|
|
254 | #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
255 | |
|
|
256 | #elif ECB_CLANG_EXTENSION(c_atomic) |
|
|
257 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
258 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
259 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
260 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
261 | #undef ECB_MEMORY_FENCE_RELAXED |
|
|
262 | #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
263 | |
| 156 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
264 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 157 | #define ECB_MEMORY_FENCE __sync_synchronize () |
265 | #define ECB_MEMORY_FENCE __sync_synchronize () |
| 158 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
266 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 159 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
267 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
268 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
269 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
270 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
271 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 160 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
272 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 161 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
273 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 162 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
274 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 163 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
275 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 164 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
276 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 165 | #elif defined _WIN32 |
277 | #elif defined _WIN32 |
| 166 | #include <WinNT.h> |
278 | #include <WinNT.h> |
| 167 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
279 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
| 168 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
280 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 169 | #include <mbarrier.h> |
281 | #include <mbarrier.h> |
| 170 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
282 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
| 171 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
283 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
| 172 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
284 | #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |
|
|
285 | #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () |
| 173 | #elif __xlC__ |
286 | #elif __xlC__ |
| 174 | #define ECB_MEMORY_FENCE __sync () |
287 | #define ECB_MEMORY_FENCE __sync () |
|
|
288 | #endif |
|
|
289 | #endif |
|
|
290 | |
|
|
291 | #ifndef ECB_MEMORY_FENCE |
|
|
292 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
293 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
294 | /* not just C11 atomics and atomic accesses */ |
|
|
295 | #include <stdatomic.h> |
|
|
296 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
297 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
298 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
| 175 | #endif |
299 | #endif |
| 176 | #endif |
300 | #endif |
| 177 | |
301 | |
| 178 | #ifndef ECB_MEMORY_FENCE |
302 | #ifndef ECB_MEMORY_FENCE |
| 179 | #if !ECB_AVOID_PTHREADS |
303 | #if !ECB_AVOID_PTHREADS |
| … | |
… | |
| 199 | |
323 | |
| 200 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
324 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 201 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
325 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 202 | #endif |
326 | #endif |
| 203 | |
327 | |
|
|
328 | #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE |
|
|
329 | #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ |
|
|
330 | #endif |
|
|
331 | |
| 204 | /*****************************************************************************/ |
332 | /*****************************************************************************/ |
| 205 | |
333 | |
| 206 | #if __cplusplus |
334 | #if ECB_CPP |
| 207 | #define ecb_inline static inline |
335 | #define ecb_inline static inline |
| 208 | #elif ECB_GCC_VERSION(2,5) |
336 | #elif ECB_GCC_VERSION(2,5) |
| 209 | #define ecb_inline static __inline__ |
337 | #define ecb_inline static __inline__ |
| 210 | #elif ECB_C99 |
338 | #elif ECB_C99 |
| 211 | #define ecb_inline static inline |
339 | #define ecb_inline static inline |
| … | |
… | |
| 225 | |
353 | |
| 226 | #define ECB_CONCAT_(a, b) a ## b |
354 | #define ECB_CONCAT_(a, b) a ## b |
| 227 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
355 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
| 228 | #define ECB_STRINGIFY_(a) # a |
356 | #define ECB_STRINGIFY_(a) # a |
| 229 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
357 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
358 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
| 230 | |
359 | |
| 231 | #define ecb_function_ ecb_inline |
360 | #define ecb_function_ ecb_inline |
| 232 | |
361 | |
| 233 | #if ECB_GCC_VERSION(3,1) |
362 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
| 234 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
363 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
364 | #else |
|
|
365 | #define ecb_attribute(attrlist) |
|
|
366 | #endif |
|
|
367 | |
|
|
368 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
| 235 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
369 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
370 | #else |
|
|
371 | /* possible C11 impl for integral types |
|
|
372 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
373 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
374 | |
|
|
375 | #define ecb_is_constant(expr) 0 |
|
|
376 | #endif |
|
|
377 | |
|
|
378 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
| 236 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
379 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
380 | #else |
|
|
381 | #define ecb_expect(expr,value) (expr) |
|
|
382 | #endif |
|
|
383 | |
|
|
384 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
| 237 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
385 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 238 | #else |
386 | #else |
| 239 | #define ecb_attribute(attrlist) |
|
|
| 240 | #define ecb_is_constant(expr) 0 |
|
|
| 241 | #define ecb_expect(expr,value) (expr) |
|
|
| 242 | #define ecb_prefetch(addr,rw,locality) |
387 | #define ecb_prefetch(addr,rw,locality) |
| 243 | #endif |
388 | #endif |
| 244 | |
389 | |
| 245 | /* no emulation for ecb_decltype */ |
390 | /* no emulation for ecb_decltype */ |
| 246 | #if ECB_GCC_VERSION(4,5) |
391 | #if ECB_CPP11 |
|
|
392 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
393 | template<class T> struct ecb_decltype_t { typedef T type; }; |
| 247 | #define ecb_decltype(x) __decltype(x) |
394 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
| 248 | #elif ECB_GCC_VERSION(3,0) |
395 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
| 249 | #define ecb_decltype(x) __typeof(x) |
396 | #define ecb_decltype(x) __typeof__ (x) |
| 250 | #endif |
397 | #endif |
| 251 | |
398 | |
|
|
399 | #if _MSC_VER >= 1300 |
|
|
400 | #define ecb_deprecated __declspec (deprecated) |
|
|
401 | #else |
|
|
402 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
|
|
403 | #endif |
|
|
404 | |
|
|
405 | #if _MSC_VER >= 1500 |
|
|
406 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
|
|
407 | #elif ECB_GCC_VERSION(4,5) |
|
|
408 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
|
|
409 | #else |
|
|
410 | #define ecb_deprecated_message(msg) ecb_deprecated |
|
|
411 | #endif |
|
|
412 | |
|
|
413 | #if _MSC_VER >= 1400 |
|
|
414 | #define ecb_noinline __declspec (noinline) |
|
|
415 | #else |
| 252 | #define ecb_noinline ecb_attribute ((__noinline__)) |
416 | #define ecb_noinline ecb_attribute ((__noinline__)) |
|
|
417 | #endif |
|
|
418 | |
| 253 | #define ecb_unused ecb_attribute ((__unused__)) |
419 | #define ecb_unused ecb_attribute ((__unused__)) |
| 254 | #define ecb_const ecb_attribute ((__const__)) |
420 | #define ecb_const ecb_attribute ((__const__)) |
| 255 | #define ecb_pure ecb_attribute ((__pure__)) |
421 | #define ecb_pure ecb_attribute ((__pure__)) |
| 256 | |
422 | |
| 257 | #if ECB_C11 |
423 | #if ECB_C11 || __IBMC_NORETURN |
|
|
424 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
| 258 | #define ecb_noreturn _Noreturn |
425 | #define ecb_noreturn _Noreturn |
|
|
426 | #elif ECB_CPP11 |
|
|
427 | #define ecb_noreturn [[noreturn]] |
|
|
428 | #elif _MSC_VER >= 1200 |
|
|
429 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
|
|
430 | #define ecb_noreturn __declspec (noreturn) |
| 259 | #else |
431 | #else |
| 260 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
432 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
| 261 | #endif |
433 | #endif |
| 262 | |
434 | |
| 263 | #if ECB_GCC_VERSION(4,3) |
435 | #if ECB_GCC_VERSION(4,3) |
| … | |
… | |
| 278 | /* for compatibility to the rest of the world */ |
450 | /* for compatibility to the rest of the world */ |
| 279 | #define ecb_likely(expr) ecb_expect_true (expr) |
451 | #define ecb_likely(expr) ecb_expect_true (expr) |
| 280 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
452 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
| 281 | |
453 | |
| 282 | /* count trailing zero bits and count # of one bits */ |
454 | /* count trailing zero bits and count # of one bits */ |
| 283 | #if ECB_GCC_VERSION(3,4) |
455 | #if ECB_GCC_VERSION(3,4) \ |
|
|
456 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
457 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
458 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
| 284 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
459 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
| 285 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
460 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
| 286 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
461 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
| 287 | #define ecb_ctz32(x) __builtin_ctz (x) |
462 | #define ecb_ctz32(x) __builtin_ctz (x) |
| 288 | #define ecb_ctz64(x) __builtin_ctzll (x) |
463 | #define ecb_ctz64(x) __builtin_ctzll (x) |
| 289 | #define ecb_popcount32(x) __builtin_popcount (x) |
464 | #define ecb_popcount32(x) __builtin_popcount (x) |
| 290 | /* no popcountll */ |
465 | /* no popcountll */ |
| 291 | #else |
466 | #else |
| 292 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
467 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
| 293 | ecb_function_ int |
468 | ecb_function_ ecb_const int |
| 294 | ecb_ctz32 (uint32_t x) |
469 | ecb_ctz32 (uint32_t x) |
| 295 | { |
470 | { |
|
|
471 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
472 | unsigned long r; |
|
|
473 | _BitScanForward (&r, x); |
|
|
474 | return (int)r; |
|
|
475 | #else |
| 296 | int r = 0; |
476 | int r = 0; |
| 297 | |
477 | |
| 298 | x &= ~x + 1; /* this isolates the lowest bit */ |
478 | x &= ~x + 1; /* this isolates the lowest bit */ |
| 299 | |
479 | |
| 300 | #if ECB_branchless_on_i386 |
480 | #if ECB_branchless_on_i386 |
| … | |
… | |
| 310 | if (x & 0xff00ff00) r += 8; |
490 | if (x & 0xff00ff00) r += 8; |
| 311 | if (x & 0xffff0000) r += 16; |
491 | if (x & 0xffff0000) r += 16; |
| 312 | #endif |
492 | #endif |
| 313 | |
493 | |
| 314 | return r; |
494 | return r; |
|
|
495 | #endif |
| 315 | } |
496 | } |
| 316 | |
497 | |
| 317 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
498 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
| 318 | ecb_function_ int |
499 | ecb_function_ ecb_const int |
| 319 | ecb_ctz64 (uint64_t x) |
500 | ecb_ctz64 (uint64_t x) |
| 320 | { |
501 | { |
|
|
502 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
503 | unsigned long r; |
|
|
504 | _BitScanForward64 (&r, x); |
|
|
505 | return (int)r; |
|
|
506 | #else |
| 321 | int shift = x & 0xffffffffU ? 0 : 32; |
507 | int shift = x & 0xffffffff ? 0 : 32; |
| 322 | return ecb_ctz32 (x >> shift) + shift; |
508 | return ecb_ctz32 (x >> shift) + shift; |
|
|
509 | #endif |
| 323 | } |
510 | } |
| 324 | |
511 | |
| 325 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
512 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
| 326 | ecb_function_ int |
513 | ecb_function_ ecb_const int |
| 327 | ecb_popcount32 (uint32_t x) |
514 | ecb_popcount32 (uint32_t x) |
| 328 | { |
515 | { |
| 329 | x -= (x >> 1) & 0x55555555; |
516 | x -= (x >> 1) & 0x55555555; |
| 330 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
517 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
| 331 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
518 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
| 332 | x *= 0x01010101; |
519 | x *= 0x01010101; |
| 333 | |
520 | |
| 334 | return x >> 24; |
521 | return x >> 24; |
| 335 | } |
522 | } |
| 336 | |
523 | |
| 337 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
524 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
| 338 | ecb_function_ int ecb_ld32 (uint32_t x) |
525 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
| 339 | { |
526 | { |
|
|
527 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
528 | unsigned long r; |
|
|
529 | _BitScanReverse (&r, x); |
|
|
530 | return (int)r; |
|
|
531 | #else |
| 340 | int r = 0; |
532 | int r = 0; |
| 341 | |
533 | |
| 342 | if (x >> 16) { x >>= 16; r += 16; } |
534 | if (x >> 16) { x >>= 16; r += 16; } |
| 343 | if (x >> 8) { x >>= 8; r += 8; } |
535 | if (x >> 8) { x >>= 8; r += 8; } |
| 344 | if (x >> 4) { x >>= 4; r += 4; } |
536 | if (x >> 4) { x >>= 4; r += 4; } |
| 345 | if (x >> 2) { x >>= 2; r += 2; } |
537 | if (x >> 2) { x >>= 2; r += 2; } |
| 346 | if (x >> 1) { r += 1; } |
538 | if (x >> 1) { r += 1; } |
| 347 | |
539 | |
| 348 | return r; |
540 | return r; |
|
|
541 | #endif |
| 349 | } |
542 | } |
| 350 | |
543 | |
| 351 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
544 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
| 352 | ecb_function_ int ecb_ld64 (uint64_t x) |
545 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
| 353 | { |
546 | { |
|
|
547 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
548 | unsigned long r; |
|
|
549 | _BitScanReverse64 (&r, x); |
|
|
550 | return (int)r; |
|
|
551 | #else |
| 354 | int r = 0; |
552 | int r = 0; |
| 355 | |
553 | |
| 356 | if (x >> 32) { x >>= 32; r += 32; } |
554 | if (x >> 32) { x >>= 32; r += 32; } |
| 357 | |
555 | |
| 358 | return r + ecb_ld32 (x); |
556 | return r + ecb_ld32 (x); |
| 359 | } |
|
|
| 360 | #endif |
557 | #endif |
|
|
558 | } |
|
|
559 | #endif |
| 361 | |
560 | |
| 362 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
561 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
| 363 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
562 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
| 364 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
563 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
| 365 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
564 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
| 366 | |
565 | |
| 367 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
566 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
| 368 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
567 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
| 369 | { |
568 | { |
| 370 | return ( (x * 0x0802U & 0x22110U) |
569 | return ( (x * 0x0802U & 0x22110U) |
| 371 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
570 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
| 372 | } |
571 | } |
| 373 | |
572 | |
| 374 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
573 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
| 375 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
574 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
| 376 | { |
575 | { |
| 377 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
576 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
| 378 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
577 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
| 379 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
578 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
| 380 | x = ( x >> 8 ) | ( x << 8); |
579 | x = ( x >> 8 ) | ( x << 8); |
| 381 | |
580 | |
| 382 | return x; |
581 | return x; |
| 383 | } |
582 | } |
| 384 | |
583 | |
| 385 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
584 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
| 386 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
585 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
| 387 | { |
586 | { |
| 388 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
587 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
| 389 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
588 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
| 390 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
589 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
| 391 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
590 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
| … | |
… | |
| 394 | return x; |
593 | return x; |
| 395 | } |
594 | } |
| 396 | |
595 | |
| 397 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
596 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 398 | /* so for this version we are lazy */ |
597 | /* so for this version we are lazy */ |
| 399 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
598 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
| 400 | ecb_function_ int |
599 | ecb_function_ ecb_const int |
| 401 | ecb_popcount64 (uint64_t x) |
600 | ecb_popcount64 (uint64_t x) |
| 402 | { |
601 | { |
| 403 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
602 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
| 404 | } |
603 | } |
| 405 | |
604 | |
| 406 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
605 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
| 407 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
606 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
| 408 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
607 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
| 409 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
608 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
| 410 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
609 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
| 411 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
610 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
| 412 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
611 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
| 413 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
612 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
| 414 | |
613 | |
| 415 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
614 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
| 416 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
615 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
| 417 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
616 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
| 418 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
617 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
| 419 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
618 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
| 420 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
619 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
| 421 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
620 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
| 422 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
621 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
| 423 | |
622 | |
| 424 | #if ECB_GCC_VERSION(4,3) |
623 | #if ECB_CPP |
|
|
624 | |
|
|
625 | inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } |
|
|
626 | inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } |
|
|
627 | inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); } |
|
|
628 | inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); } |
|
|
629 | |
|
|
630 | inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); } |
|
|
631 | inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); } |
|
|
632 | inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); } |
|
|
633 | inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); } |
|
|
634 | |
|
|
635 | inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); } |
|
|
636 | inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); } |
|
|
637 | inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); } |
|
|
638 | inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); } |
|
|
639 | |
|
|
640 | inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); } |
|
|
641 | inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); } |
|
|
642 | inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); } |
|
|
643 | inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); } |
|
|
644 | |
|
|
645 | inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); } |
|
|
646 | inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); } |
|
|
647 | inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); } |
|
|
648 | |
|
|
649 | inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); } |
|
|
650 | inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); } |
|
|
651 | inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); } |
|
|
652 | inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); } |
|
|
653 | |
|
|
654 | inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); } |
|
|
655 | inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); } |
|
|
656 | inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); } |
|
|
657 | inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); } |
|
|
658 | |
|
|
659 | #endif |
|
|
660 | |
|
|
661 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
|
|
662 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
|
|
663 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
|
|
664 | #else |
| 425 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
665 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
666 | #endif |
| 426 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
667 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
| 427 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
668 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
|
|
669 | #elif _MSC_VER |
|
|
670 | #include <stdlib.h> |
|
|
671 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
|
|
672 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
|
|
673 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
| 428 | #else |
674 | #else |
| 429 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
675 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
| 430 | ecb_function_ uint16_t |
676 | ecb_function_ ecb_const uint16_t |
| 431 | ecb_bswap16 (uint16_t x) |
677 | ecb_bswap16 (uint16_t x) |
| 432 | { |
678 | { |
| 433 | return ecb_rotl16 (x, 8); |
679 | return ecb_rotl16 (x, 8); |
| 434 | } |
680 | } |
| 435 | |
681 | |
| 436 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
682 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
| 437 | ecb_function_ uint32_t |
683 | ecb_function_ ecb_const uint32_t |
| 438 | ecb_bswap32 (uint32_t x) |
684 | ecb_bswap32 (uint32_t x) |
| 439 | { |
685 | { |
| 440 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
686 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
| 441 | } |
687 | } |
| 442 | |
688 | |
| 443 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
689 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
| 444 | ecb_function_ uint64_t |
690 | ecb_function_ ecb_const uint64_t |
| 445 | ecb_bswap64 (uint64_t x) |
691 | ecb_bswap64 (uint64_t x) |
| 446 | { |
692 | { |
| 447 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
693 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 448 | } |
694 | } |
| 449 | #endif |
695 | #endif |
| 450 | |
696 | |
| 451 | #if ECB_GCC_VERSION(4,5) |
697 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
| 452 | #define ecb_unreachable() __builtin_unreachable () |
698 | #define ecb_unreachable() __builtin_unreachable () |
| 453 | #else |
699 | #else |
| 454 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
700 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 455 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
701 | ecb_inline ecb_noreturn void ecb_unreachable (void); |
| 456 | ecb_inline void ecb_unreachable (void) { } |
702 | ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
| 457 | #endif |
703 | #endif |
| 458 | |
704 | |
| 459 | /* try to tell the compiler that some condition is definitely true */ |
705 | /* try to tell the compiler that some condition is definitely true */ |
| 460 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
706 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 461 | |
707 | |
| 462 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
708 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
| 463 | ecb_inline unsigned char |
709 | ecb_inline ecb_const uint32_t |
| 464 | ecb_byteorder_helper (void) |
710 | ecb_byteorder_helper (void) |
| 465 | { |
711 | { |
| 466 | const uint32_t u = 0x11223344; |
712 | /* the union code still generates code under pressure in gcc, */ |
| 467 | return *(unsigned char *)&u; |
713 | /* but less than using pointers, and always seems to */ |
|
|
714 | /* successfully return a constant. */ |
|
|
715 | /* the reason why we have this horrible preprocessor mess */ |
|
|
716 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
717 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
718 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
|
|
719 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
|
|
720 | #define ECB_LITTLE_ENDIAN 1 |
|
|
721 | return 0x44332211; |
|
|
722 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
|
|
723 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
|
|
724 | #define ECB_BIG_ENDIAN 1 |
|
|
725 | return 0x11223344; |
|
|
726 | #else |
|
|
727 | union |
|
|
728 | { |
|
|
729 | uint8_t c[4]; |
|
|
730 | uint32_t u; |
|
|
731 | } u = { 0x11, 0x22, 0x33, 0x44 }; |
|
|
732 | return u.u; |
|
|
733 | #endif |
| 468 | } |
734 | } |
| 469 | |
735 | |
| 470 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
736 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
| 471 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
737 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
| 472 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
738 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
| 473 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
739 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
|
|
740 | |
|
|
741 | /*****************************************************************************/ |
|
|
742 | /* unaligned load/store */ |
|
|
743 | |
|
|
744 | ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
|
|
745 | ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
|
|
746 | ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
|
|
747 | |
|
|
748 | ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
|
|
749 | ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
|
|
750 | ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
|
|
751 | |
|
|
752 | ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
753 | ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
754 | ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
755 | |
|
|
756 | ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); } |
|
|
757 | ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); } |
|
|
758 | ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); } |
|
|
759 | |
|
|
760 | ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); } |
|
|
761 | ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); } |
|
|
762 | ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); } |
|
|
763 | |
|
|
764 | ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
|
|
765 | ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
|
|
766 | ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
|
|
767 | |
|
|
768 | ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
|
|
769 | ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
|
|
770 | ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
|
|
771 | |
|
|
772 | ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
773 | ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
774 | ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
775 | |
|
|
776 | ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); } |
|
|
777 | ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); } |
|
|
778 | ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); } |
|
|
779 | |
|
|
780 | ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); } |
|
|
781 | ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); } |
|
|
782 | ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); } |
|
|
783 | |
|
|
784 | #if ECB_CPP |
|
|
785 | |
|
|
786 | inline uint8_t ecb_bswap (uint8_t v) { return v; } |
|
|
787 | inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); } |
|
|
788 | inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); } |
|
|
789 | inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); } |
|
|
790 | |
|
|
791 | template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
|
|
792 | template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
|
|
793 | template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; } |
|
|
794 | template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); } |
|
|
795 | template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); } |
|
|
796 | template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
797 | template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); } |
|
|
798 | template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); } |
|
|
799 | |
|
|
800 | template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
|
|
801 | template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
|
|
802 | template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; } |
|
|
803 | template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); } |
|
|
804 | template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); } |
|
|
805 | template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
806 | template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } |
|
|
807 | template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } |
|
|
808 | |
|
|
809 | #endif |
|
|
810 | |
|
|
811 | /*****************************************************************************/ |
|
|
812 | /* division */ |
| 474 | |
813 | |
| 475 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
814 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
|
|
815 | /* C99 tightened the definition of %, so we can use a more efficient version */ |
| 476 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
816 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 477 | #else |
817 | #else |
| 478 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
818 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
| 479 | #endif |
819 | #endif |
| 480 | |
820 | |
| 481 | #if __cplusplus |
821 | #if ECB_CPP |
| 482 | template<typename T> |
822 | template<typename T> |
| 483 | static inline T ecb_div_rd (T val, T div) |
823 | static inline T ecb_div_rd (T val, T div) |
| 484 | { |
824 | { |
| 485 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
825 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
| 486 | } |
826 | } |
| … | |
… | |
| 491 | } |
831 | } |
| 492 | #else |
832 | #else |
| 493 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
833 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
| 494 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
834 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 495 | #endif |
835 | #endif |
|
|
836 | |
|
|
837 | /*****************************************************************************/ |
|
|
838 | /* array length */ |
| 496 | |
839 | |
| 497 | #if ecb_cplusplus_does_not_suck |
840 | #if ecb_cplusplus_does_not_suck |
| 498 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
841 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
| 499 | template<typename T, int N> |
842 | template<typename T, int N> |
| 500 | static inline int ecb_array_length (const T (&arr)[N]) |
843 | static inline int ecb_array_length (const T (&arr)[N]) |
| … | |
… | |
| 503 | } |
846 | } |
| 504 | #else |
847 | #else |
| 505 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
848 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 506 | #endif |
849 | #endif |
| 507 | |
850 | |
|
|
851 | /*****************************************************************************/ |
|
|
852 | /* IEEE 754-2008 half float conversions */ |
|
|
853 | |
|
|
854 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
|
|
855 | ecb_function_ ecb_const uint32_t |
|
|
856 | ecb_binary16_to_binary32 (uint32_t x) |
|
|
857 | { |
|
|
858 | unsigned int s = (x & 0x8000) << (31 - 15); |
|
|
859 | int e = (x >> 10) & 0x001f; |
|
|
860 | unsigned int m = x & 0x03ff; |
|
|
861 | |
|
|
862 | if (ecb_expect_false (e == 31)) |
|
|
863 | /* infinity or NaN */ |
|
|
864 | e = 255 - (127 - 15); |
|
|
865 | else if (ecb_expect_false (!e)) |
|
|
866 | { |
|
|
867 | if (ecb_expect_true (!m)) |
|
|
868 | /* zero, handled by code below by forcing e to 0 */ |
|
|
869 | e = 0 - (127 - 15); |
|
|
870 | else |
|
|
871 | { |
|
|
872 | /* subnormal, renormalise */ |
|
|
873 | unsigned int s = 10 - ecb_ld32 (m); |
|
|
874 | |
|
|
875 | m = (m << s) & 0x3ff; /* mask implicit bit */ |
|
|
876 | e -= s - 1; |
|
|
877 | } |
|
|
878 | } |
|
|
879 | |
|
|
880 | /* e and m now are normalised, or zero, (or inf or nan) */ |
|
|
881 | e += 127 - 15; |
|
|
882 | |
|
|
883 | return s | (e << 23) | (m << (23 - 10)); |
|
|
884 | } |
|
|
885 | |
|
|
886 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
|
|
887 | ecb_function_ ecb_const uint16_t |
|
|
888 | ecb_binary32_to_binary16 (uint32_t x) |
|
|
889 | { |
|
|
890 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
|
|
891 | int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
|
|
892 | unsigned int m = x & 0x007fffff; |
|
|
893 | |
|
|
894 | x &= 0x7fffffff; |
|
|
895 | |
|
|
896 | /* if it's within range of binary16 normals, use fast path */ |
|
|
897 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
|
|
898 | { |
|
|
899 | /* mantissa round-to-even */ |
|
|
900 | m += 0x00000fff + ((m >> (23 - 10)) & 1); |
|
|
901 | |
|
|
902 | /* handle overflow */ |
|
|
903 | if (ecb_expect_false (m >= 0x00800000)) |
|
|
904 | { |
|
|
905 | m >>= 1; |
|
|
906 | e += 1; |
|
|
907 | } |
|
|
908 | |
|
|
909 | return s | (e << 10) | (m >> (23 - 10)); |
|
|
910 | } |
|
|
911 | |
|
|
912 | /* handle large numbers and infinity */ |
|
|
913 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
|
|
914 | return s | 0x7c00; |
|
|
915 | |
|
|
916 | /* handle zero, subnormals and small numbers */ |
|
|
917 | if (ecb_expect_true (x < 0x38800000)) |
|
|
918 | { |
|
|
919 | /* zero */ |
|
|
920 | if (ecb_expect_true (!x)) |
|
|
921 | return s; |
|
|
922 | |
|
|
923 | /* handle subnormals */ |
|
|
924 | |
|
|
925 | /* too small, will be zero */ |
|
|
926 | if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
|
|
927 | return s; |
|
|
928 | |
|
|
929 | m |= 0x00800000; /* make implicit bit explicit */ |
|
|
930 | |
|
|
931 | /* very tricky - we need to round to the nearest e (+10) bit value */ |
|
|
932 | { |
|
|
933 | unsigned int bits = 14 - e; |
|
|
934 | unsigned int half = (1 << (bits - 1)) - 1; |
|
|
935 | unsigned int even = (m >> bits) & 1; |
|
|
936 | |
|
|
937 | /* if this overflows, we will end up with a normalised number */ |
|
|
938 | m = (m + half + even) >> bits; |
|
|
939 | } |
|
|
940 | |
|
|
941 | return s | m; |
|
|
942 | } |
|
|
943 | |
|
|
944 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
|
|
945 | m >>= 13; |
|
|
946 | |
|
|
947 | return s | 0x7c00 | m | !m; |
|
|
948 | } |
|
|
949 | |
|
|
950 | /*******************************************************************************/ |
|
|
951 | /* fast integer to ascii */ |
|
|
952 | |
|
|
953 | // simply return a mask with "bits" bits set |
|
|
954 | #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) |
|
|
955 | |
|
|
956 | // oputput a single digit. maskvalue is 10**digitidx |
|
|
957 | #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ |
|
|
958 | if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ |
|
|
959 | { \ |
|
|
960 | char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ |
|
|
961 | *ptr = digit + '0'; /* output it */ \ |
|
|
962 | nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ |
|
|
963 | ptr += nz; /* output digit only if non-zero digit seen */ \ |
|
|
964 | x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \ |
|
|
965 | } |
|
|
966 | |
|
|
967 | // convert integer to fixed point format and multiply out digits, highest first |
|
|
968 | // requires magic constants: max. digits and number of bits after the decimal point |
|
|
969 | #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \ |
|
|
970 | ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ |
|
|
971 | { \ |
|
|
972 | char nz = lz; /* non-zero digit seen? */ \ |
|
|
973 | /* convert to x.bits fixed-point */ \ |
|
|
974 | type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \ |
|
|
975 | /* output up to 10 digits */ \ |
|
|
976 | ecb_i2a_digit (type,bits,digitmask, 1, 0); \ |
|
|
977 | ecb_i2a_digit (type,bits,digitmask, 10, 1); \ |
|
|
978 | ecb_i2a_digit (type,bits,digitmask, 100, 2); \ |
|
|
979 | ecb_i2a_digit (type,bits,digitmask, 1000, 3); \ |
|
|
980 | ecb_i2a_digit (type,bits,digitmask, 10000, 4); \ |
|
|
981 | ecb_i2a_digit (type,bits,digitmask, 100000, 5); \ |
|
|
982 | ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \ |
|
|
983 | ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \ |
|
|
984 | ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ |
|
|
985 | ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ |
|
|
986 | return ptr; \ |
|
|
987 | } |
|
|
988 | |
|
|
989 | // predefined versions of the above, for various digits |
|
|
990 | // ecb_i2a_xN = almost N digits, limit defined by macro |
|
|
991 | // ecb_i2a_N = up to N digits, leading zeroes suppressed |
|
|
992 | // ecb_i2a_0N = exactly N digits, including leading zeroes |
|
|
993 | |
|
|
994 | // non-leading-zero versions, limited range |
|
|
995 | #define ECB_I2A_MAX_X5 59074 // limit for ecb_i2a_x5 |
|
|
996 | #define ECB_I2A_MAX_X10 2932500665 // limit for ecb_i2a_x10 |
|
|
997 | ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) |
|
|
998 | ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) |
|
|
999 | |
|
|
1000 | // non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit |
|
|
1001 | ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) |
|
|
1002 | ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) |
|
|
1003 | ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) |
|
|
1004 | ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) |
|
|
1005 | ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) |
|
|
1006 | ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) |
|
|
1007 | ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) |
|
|
1008 | ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) |
|
|
1009 | |
|
|
1010 | // leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit |
|
|
1011 | ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) |
|
|
1012 | ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) |
|
|
1013 | ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) |
|
|
1014 | ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) |
|
|
1015 | ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) |
|
|
1016 | ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1) |
|
|
1017 | ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1) |
|
|
1018 | ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1) |
|
|
1019 | |
|
|
1020 | ecb_inline char * |
|
|
1021 | ecb_i2a_u32 (char *ptr, uint32_t u) |
|
|
1022 | { |
|
|
1023 | #if ECB_64BIT_NATIVE |
|
|
1024 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
|
|
1025 | ptr = ecb_i2a_x10 (ptr, u); |
|
|
1026 | else // x10 almost, but not fully, covers 32 bit |
|
|
1027 | { |
|
|
1028 | uint32_t u1 = u % 1000000000; |
|
|
1029 | uint32_t u2 = u / 1000000000; |
|
|
1030 | |
|
|
1031 | *ptr++ = u2 + '0'; |
|
|
1032 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1033 | } |
|
|
1034 | #else |
|
|
1035 | if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) |
|
|
1036 | ecb_i2a_x5 (ptr, u); |
|
|
1037 | else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000)) |
|
|
1038 | { |
|
|
1039 | uint32_t u1 = u % 10000; |
|
|
1040 | uint32_t u2 = u / 10000; |
|
|
1041 | |
|
|
1042 | ptr = ecb_i2a_x5 (ptr, u2); |
|
|
1043 | ptr = ecb_i2a_04 (ptr, u1); |
|
|
1044 | } |
|
|
1045 | else |
|
|
1046 | { |
|
|
1047 | uint32_t u1 = u % 10000; |
|
|
1048 | uint32_t ua = u / 10000; |
|
|
1049 | uint32_t u2 = ua % 10000; |
|
|
1050 | uint32_t u3 = ua / 10000; |
|
|
1051 | |
|
|
1052 | ptr = ecb_i2a_2 (ptr, u3); |
|
|
1053 | ptr = ecb_i2a_04 (ptr, u2); |
|
|
1054 | ptr = ecb_i2a_04 (ptr, u1); |
|
|
1055 | } |
| 508 | #endif |
1056 | #endif |
| 509 | |
1057 | |
|
|
1058 | return ptr; |
|
|
1059 | } |
|
|
1060 | |
|
|
1061 | ecb_inline char * |
|
|
1062 | ecb_i2a_i32 (char *ptr, int32_t v) |
|
|
1063 | { |
|
|
1064 | *ptr = '-'; ptr += v < 0; |
|
|
1065 | uint32_t u = v < 0 ? -(uint32_t)v : v; |
|
|
1066 | |
|
|
1067 | #if ECB_64BIT_NATIVE |
|
|
1068 | ptr = ecb_i2a_x10 (ptr, u); // x10 fully covers 31 bit |
|
|
1069 | #else |
|
|
1070 | ptr = ecb_i2a_u32 (ptr, u); |
|
|
1071 | #endif |
|
|
1072 | |
|
|
1073 | return ptr; |
|
|
1074 | } |
|
|
1075 | |
|
|
1076 | ecb_inline char * |
|
|
1077 | ecb_i2a_u64 (char *ptr, uint64_t u) |
|
|
1078 | { |
|
|
1079 | #if ECB_64BIT_NATIVE |
|
|
1080 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
|
|
1081 | ptr = ecb_i2a_x10 (ptr, u); |
|
|
1082 | else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) |
|
|
1083 | { |
|
|
1084 | uint64_t u1 = u % 1000000000; |
|
|
1085 | uint64_t u2 = u / 1000000000; |
|
|
1086 | |
|
|
1087 | ptr = ecb_i2a_x10 (ptr, u2); |
|
|
1088 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1089 | } |
|
|
1090 | else |
|
|
1091 | { |
|
|
1092 | uint64_t u1 = u % 1000000000; |
|
|
1093 | uint64_t ua = u / 1000000000; |
|
|
1094 | uint64_t u2 = ua % 1000000000; |
|
|
1095 | uint64_t u3 = ua / 1000000000; |
|
|
1096 | |
|
|
1097 | ptr = ecb_i2a_2 (ptr, u3); |
|
|
1098 | ptr = ecb_i2a_09 (ptr, u2); |
|
|
1099 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1100 | } |
|
|
1101 | #else |
|
|
1102 | if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) |
|
|
1103 | ptr = ecb_i2a_x5 (ptr, u); |
|
|
1104 | else |
|
|
1105 | { |
|
|
1106 | uint64_t u1 = u % 10000; |
|
|
1107 | uint64_t u2 = u / 10000; |
|
|
1108 | |
|
|
1109 | ptr = ecb_i2a_u64 (ptr, u2); |
|
|
1110 | ptr = ecb_i2a_04 (ptr, u1); |
|
|
1111 | } |
|
|
1112 | #endif |
|
|
1113 | |
|
|
1114 | return ptr; |
|
|
1115 | } |
|
|
1116 | |
|
|
1117 | ecb_inline char * |
|
|
1118 | ecb_i2a_i64 (char *ptr, int64_t v) |
|
|
1119 | { |
|
|
1120 | *ptr = '-'; ptr += v < 0; |
|
|
1121 | uint64_t u = v < 0 ? -(uint64_t)v : v; |
|
|
1122 | |
|
|
1123 | #if ECB_64BIT_NATIVE |
|
|
1124 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
|
|
1125 | ptr = ecb_i2a_x10 (ptr, u); |
|
|
1126 | else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) |
|
|
1127 | { |
|
|
1128 | uint64_t u1 = u % 1000000000; |
|
|
1129 | uint64_t u2 = u / 1000000000; |
|
|
1130 | |
|
|
1131 | ptr = ecb_i2a_x10 (ptr, u2); |
|
|
1132 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1133 | } |
|
|
1134 | else |
|
|
1135 | { |
|
|
1136 | uint64_t u1 = u % 1000000000; |
|
|
1137 | uint64_t ua = u / 1000000000; |
|
|
1138 | uint64_t u2 = ua % 1000000000; |
|
|
1139 | uint64_t u3 = ua / 1000000000; |
|
|
1140 | |
|
|
1141 | // 2**31 is 19 digits, so the top is exactly one digit |
|
|
1142 | *ptr++ = u3 + '0'; |
|
|
1143 | ptr = ecb_i2a_09 (ptr, u2); |
|
|
1144 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1145 | } |
|
|
1146 | #else |
|
|
1147 | ptr = ecb_i2a_u64 (ptr, u); |
|
|
1148 | #endif |
|
|
1149 | |
|
|
1150 | return ptr; |
|
|
1151 | } |
|
|
1152 | |
|
|
1153 | /*******************************************************************************/ |
|
|
1154 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
1155 | |
|
|
1156 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
1157 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
1158 | #if 0 \ |
|
|
1159 | || __i386 || __i386__ \ |
|
|
1160 | || ECB_GCC_AMD64 \ |
|
|
1161 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
1162 | || defined __s390__ || defined __s390x__ \ |
|
|
1163 | || defined __mips__ \ |
|
|
1164 | || defined __alpha__ \ |
|
|
1165 | || defined __hppa__ \ |
|
|
1166 | || defined __ia64__ \ |
|
|
1167 | || defined __m68k__ \ |
|
|
1168 | || defined __m88k__ \ |
|
|
1169 | || defined __sh__ \ |
|
|
1170 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
|
|
1171 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
1172 | || defined __aarch64__ |
|
|
1173 | #define ECB_STDFP 1 |
|
|
1174 | #else |
|
|
1175 | #define ECB_STDFP 0 |
|
|
1176 | #endif |
|
|
1177 | |
|
|
1178 | #ifndef ECB_NO_LIBM |
|
|
1179 | |
|
|
1180 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
1181 | |
|
|
1182 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
1183 | #ifdef INFINITY |
|
|
1184 | #define ECB_INFINITY INFINITY |
|
|
1185 | #else |
|
|
1186 | #define ECB_INFINITY HUGE_VAL |
|
|
1187 | #endif |
|
|
1188 | |
|
|
1189 | #ifdef NAN |
|
|
1190 | #define ECB_NAN NAN |
|
|
1191 | #else |
|
|
1192 | #define ECB_NAN ECB_INFINITY |
|
|
1193 | #endif |
|
|
1194 | |
|
|
1195 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
|
|
1196 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
1197 | #define ecb_frexpf(x,e) frexpf ((x), (e)) |
|
|
1198 | #else |
|
|
1199 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
|
|
1200 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
|
|
1201 | #endif |
|
|
1202 | |
|
|
1203 | /* convert a float to ieee single/binary32 */ |
|
|
1204 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
|
|
1205 | ecb_function_ ecb_const uint32_t |
|
|
1206 | ecb_float_to_binary32 (float x) |
|
|
1207 | { |
|
|
1208 | uint32_t r; |
|
|
1209 | |
|
|
1210 | #if ECB_STDFP |
|
|
1211 | memcpy (&r, &x, 4); |
|
|
1212 | #else |
|
|
1213 | /* slow emulation, works for anything but -0 */ |
|
|
1214 | uint32_t m; |
|
|
1215 | int e; |
|
|
1216 | |
|
|
1217 | if (x == 0e0f ) return 0x00000000U; |
|
|
1218 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
1219 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
1220 | if (x != x ) return 0x7fbfffffU; |
|
|
1221 | |
|
|
1222 | m = ecb_frexpf (x, &e) * 0x1000000U; |
|
|
1223 | |
|
|
1224 | r = m & 0x80000000U; |
|
|
1225 | |
|
|
1226 | if (r) |
|
|
1227 | m = -m; |
|
|
1228 | |
|
|
1229 | if (e <= -126) |
|
|
1230 | { |
|
|
1231 | m &= 0xffffffU; |
|
|
1232 | m >>= (-125 - e); |
|
|
1233 | e = -126; |
|
|
1234 | } |
|
|
1235 | |
|
|
1236 | r |= (e + 126) << 23; |
|
|
1237 | r |= m & 0x7fffffU; |
|
|
1238 | #endif |
|
|
1239 | |
|
|
1240 | return r; |
|
|
1241 | } |
|
|
1242 | |
|
|
1243 | /* converts an ieee single/binary32 to a float */ |
|
|
1244 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
|
|
1245 | ecb_function_ ecb_const float |
|
|
1246 | ecb_binary32_to_float (uint32_t x) |
|
|
1247 | { |
|
|
1248 | float r; |
|
|
1249 | |
|
|
1250 | #if ECB_STDFP |
|
|
1251 | memcpy (&r, &x, 4); |
|
|
1252 | #else |
|
|
1253 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
1254 | int neg = x >> 31; |
|
|
1255 | int e = (x >> 23) & 0xffU; |
|
|
1256 | |
|
|
1257 | x &= 0x7fffffU; |
|
|
1258 | |
|
|
1259 | if (e) |
|
|
1260 | x |= 0x800000U; |
|
|
1261 | else |
|
|
1262 | e = 1; |
|
|
1263 | |
|
|
1264 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
1265 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
1266 | |
|
|
1267 | r = neg ? -r : r; |
|
|
1268 | #endif |
|
|
1269 | |
|
|
1270 | return r; |
|
|
1271 | } |
|
|
1272 | |
|
|
1273 | /* convert a double to ieee double/binary64 */ |
|
|
1274 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
|
|
1275 | ecb_function_ ecb_const uint64_t |
|
|
1276 | ecb_double_to_binary64 (double x) |
|
|
1277 | { |
|
|
1278 | uint64_t r; |
|
|
1279 | |
|
|
1280 | #if ECB_STDFP |
|
|
1281 | memcpy (&r, &x, 8); |
|
|
1282 | #else |
|
|
1283 | /* slow emulation, works for anything but -0 */ |
|
|
1284 | uint64_t m; |
|
|
1285 | int e; |
|
|
1286 | |
|
|
1287 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
1288 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
1289 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
1290 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
1291 | |
|
|
1292 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
1293 | |
|
|
1294 | r = m & 0x8000000000000000;; |
|
|
1295 | |
|
|
1296 | if (r) |
|
|
1297 | m = -m; |
|
|
1298 | |
|
|
1299 | if (e <= -1022) |
|
|
1300 | { |
|
|
1301 | m &= 0x1fffffffffffffU; |
|
|
1302 | m >>= (-1021 - e); |
|
|
1303 | e = -1022; |
|
|
1304 | } |
|
|
1305 | |
|
|
1306 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
1307 | r |= m & 0xfffffffffffffU; |
|
|
1308 | #endif |
|
|
1309 | |
|
|
1310 | return r; |
|
|
1311 | } |
|
|
1312 | |
|
|
1313 | /* converts an ieee double/binary64 to a double */ |
|
|
1314 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
|
|
1315 | ecb_function_ ecb_const double |
|
|
1316 | ecb_binary64_to_double (uint64_t x) |
|
|
1317 | { |
|
|
1318 | double r; |
|
|
1319 | |
|
|
1320 | #if ECB_STDFP |
|
|
1321 | memcpy (&r, &x, 8); |
|
|
1322 | #else |
|
|
1323 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
1324 | int neg = x >> 63; |
|
|
1325 | int e = (x >> 52) & 0x7ffU; |
|
|
1326 | |
|
|
1327 | x &= 0xfffffffffffffU; |
|
|
1328 | |
|
|
1329 | if (e) |
|
|
1330 | x |= 0x10000000000000U; |
|
|
1331 | else |
|
|
1332 | e = 1; |
|
|
1333 | |
|
|
1334 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
1335 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
1336 | |
|
|
1337 | r = neg ? -r : r; |
|
|
1338 | #endif |
|
|
1339 | |
|
|
1340 | return r; |
|
|
1341 | } |
|
|
1342 | |
|
|
1343 | /* convert a float to ieee half/binary16 */ |
|
|
1344 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
|
|
1345 | ecb_function_ ecb_const uint16_t |
|
|
1346 | ecb_float_to_binary16 (float x) |
|
|
1347 | { |
|
|
1348 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
|
|
1349 | } |
|
|
1350 | |
|
|
1351 | /* convert an ieee half/binary16 to float */ |
|
|
1352 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
1353 | ecb_function_ ecb_const float |
|
|
1354 | ecb_binary16_to_float (uint16_t x) |
|
|
1355 | { |
|
|
1356 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
|
|
1357 | } |
|
|
1358 | |
|
|
1359 | #endif |
|
|
1360 | |
|
|
1361 | #endif |
|
|
1362 | |
