| 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 (©) 2009-2015,2018-2020 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 0x00010008 |
| 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 |
|
|
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 |
| 66 | #endif |
106 | #endif |
| 67 | #endif |
107 | #endif |
| 68 | |
108 | |
| 69 | /* many compilers define _GNUC_ to some versions but then only implement |
109 | /* many compilers define _GNUC_ to some versions but then only implement |
| 70 | * what their idiot authors think are the "more important" extensions, |
110 | * what their idiot authors think are the "more important" extensions, |
| 71 | * causing enormous grief in return for some better fake benchmark numbers. |
111 | * causing enormous grief in return for some better fake benchmark numbers. |
| 72 | * or so. |
112 | * or so. |
| 73 | * we try to detect these and simply assume they are not gcc - if they have |
113 | * 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. |
114 | * an issue with that they should have done it right in the first place. |
| 75 | */ |
115 | */ |
| 76 | #ifndef ECB_GCC_VERSION |
|
|
| 77 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
116 | #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 |
117 | #define ECB_GCC_VERSION(major,minor) 0 |
| 79 | #else |
118 | #else |
| 80 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
119 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 81 | #endif |
120 | #endif |
| 82 | #endif |
|
|
| 83 | |
121 | |
| 84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
122 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
| 85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
123 | |
| 86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
124 | #if __clang__ && defined __has_builtin |
|
|
125 | #define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
|
|
126 | #else |
|
|
127 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
128 | #endif |
|
|
129 | |
|
|
130 | #if __clang__ && defined __has_extension |
|
|
131 | #define ECB_CLANG_EXTENSION(x) __has_extension (x) |
|
|
132 | #else |
|
|
133 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
134 | #endif |
|
|
135 | |
| 87 | #define ECB_CPP (__cplusplus+0) |
136 | #define ECB_CPP (__cplusplus+0) |
| 88 | #define ECB_CPP98 (__cplusplus >= 199711L) |
|
|
| 89 | #define ECB_CPP11 (__cplusplus >= 201103L) |
137 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
138 | #define ECB_CPP14 (__cplusplus >= 201402L) |
|
|
139 | #define ECB_CPP17 (__cplusplus >= 201703L) |
|
|
140 | |
|
|
141 | #if ECB_CPP |
|
|
142 | #define ECB_C 0 |
|
|
143 | #define ECB_STDC_VERSION 0 |
|
|
144 | #else |
|
|
145 | #define ECB_C 1 |
|
|
146 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
147 | #endif |
|
|
148 | |
|
|
149 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
150 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
151 | #define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
|
|
152 | |
|
|
153 | #if ECB_CPP |
|
|
154 | #define ECB_EXTERN_C extern "C" |
|
|
155 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
|
|
156 | #define ECB_EXTERN_C_END } |
|
|
157 | #else |
|
|
158 | #define ECB_EXTERN_C extern |
|
|
159 | #define ECB_EXTERN_C_BEG |
|
|
160 | #define ECB_EXTERN_C_END |
|
|
161 | #endif |
| 90 | |
162 | |
| 91 | /*****************************************************************************/ |
163 | /*****************************************************************************/ |
| 92 | |
164 | |
| 93 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
165 | /* 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 */ |
166 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
| 95 | |
167 | |
| 96 | #if ECB_NO_THREADS |
168 | #if ECB_NO_THREADS |
| 97 | # define ECB_NO_SMP 1 |
169 | #define ECB_NO_SMP 1 |
| 98 | #endif |
170 | #endif |
| 99 | |
171 | |
| 100 | #if ECB_NO_SMP |
172 | #if ECB_NO_SMP |
| 101 | #define ECB_MEMORY_FENCE do { } while (0) |
173 | #define ECB_MEMORY_FENCE do { } while (0) |
| 102 | #endif |
174 | #endif |
| 103 | |
175 | |
|
|
176 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
|
|
177 | #if __xlC__ && ECB_CPP |
|
|
178 | #include <builtins.h> |
|
|
179 | #endif |
|
|
180 | |
|
|
181 | #if 1400 <= _MSC_VER |
|
|
182 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
|
|
183 | #endif |
|
|
184 | |
| 104 | #ifndef ECB_MEMORY_FENCE |
185 | #ifndef ECB_MEMORY_FENCE |
| 105 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
186 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
187 | #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") |
| 106 | #if __i386 || __i386__ |
188 | #if __i386 || __i386__ |
| 107 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
189 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
| 108 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
190 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 109 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
191 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 110 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
192 | #elif ECB_GCC_AMD64 |
| 111 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
193 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
| 112 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
194 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
| 113 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
195 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
| 114 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
196 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
| 115 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
197 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
198 | #elif defined __ARM_ARCH_2__ \ |
|
|
199 | || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
|
|
200 | || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
|
|
201 | || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
|
|
202 | || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
|
|
203 | || defined __ARM_ARCH_5TEJ__ |
|
|
204 | /* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
| 116 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
205 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
| 117 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
206 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
|
|
207 | || defined __ARM_ARCH_6T2__ |
| 118 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
208 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
| 119 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
209 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
| 120 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
210 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
| 121 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
211 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
| 122 | #elif __sparc || __sparc__ |
212 | #elif __aarch64__ |
|
|
213 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
|
|
214 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
| 123 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
215 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
| 124 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
216 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
| 125 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
217 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
| 126 | #elif defined __s390__ || defined __s390x__ |
218 | #elif defined __s390__ || defined __s390x__ |
| 127 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
219 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
| 128 | #elif defined __mips__ |
220 | #elif defined __mips__ |
|
|
221 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
222 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
| 129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
223 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
| 130 | #elif defined __alpha__ |
224 | #elif defined __alpha__ |
| 131 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
225 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
| 132 | #elif defined __hppa__ |
226 | #elif defined __hppa__ |
| 133 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
227 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 134 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
228 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
| 135 | #elif defined __ia64__ |
229 | #elif defined __ia64__ |
| 136 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
230 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
231 | #elif defined __m68k__ |
|
|
232 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
233 | #elif defined __m88k__ |
|
|
234 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
235 | #elif defined __sh__ |
|
|
236 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
| 137 | #endif |
237 | #endif |
| 138 | #endif |
238 | #endif |
| 139 | #endif |
239 | #endif |
| 140 | |
240 | |
| 141 | #ifndef ECB_MEMORY_FENCE |
241 | #ifndef ECB_MEMORY_FENCE |
| 142 | #if ECB_GCC_VERSION(4,7) |
242 | #if ECB_GCC_VERSION(4,7) |
| 143 | /* see comment below about the C11 memory model. in short - avoid */ |
243 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
| 144 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
244 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
| 145 | #elif defined __clang && __has_feature (cxx_atomic) |
245 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
| 146 | /* see above */ |
246 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
247 | #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
248 | |
|
|
249 | #elif ECB_CLANG_EXTENSION(c_atomic) |
|
|
250 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
| 147 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
251 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
252 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
253 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
254 | #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) |
|
|
255 | |
| 148 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
256 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
| 149 | #define ECB_MEMORY_FENCE __sync_synchronize () |
257 | #define ECB_MEMORY_FENCE __sync_synchronize () |
| 150 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
258 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
| 151 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
259 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
260 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
261 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
262 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
263 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
| 152 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
264 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
| 153 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
265 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
| 154 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
266 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
| 155 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
267 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
| 156 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
268 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
| 157 | #elif defined _WIN32 |
269 | #elif defined _WIN32 |
| 158 | #include <WinNT.h> |
270 | #include <WinNT.h> |
| 159 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
271 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
| 160 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
272 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
| 161 | #include <mbarrier.h> |
273 | #include <mbarrier.h> |
| 162 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
274 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
| 163 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
275 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
| 164 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
276 | #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |
|
|
277 | #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () |
| 165 | #elif __xlC__ |
278 | #elif __xlC__ |
| 166 | #define ECB_MEMORY_FENCE __sync () |
279 | #define ECB_MEMORY_FENCE __sync () |
| 167 | #endif |
280 | #endif |
| 168 | #endif |
281 | #endif |
| 169 | |
282 | |
| 170 | #ifndef ECB_MEMORY_FENCE |
283 | #ifndef ECB_MEMORY_FENCE |
| 171 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
284 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
| 172 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
285 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
| 173 | /* not just C11 atomics and atomic accesses */ |
286 | /* not just C11 atomics and atomic accesses */ |
| 174 | #include <stdatomic.h> |
287 | #include <stdatomic.h> |
| 175 | /* unfortunately, the C11 memory model seems to be very limited, and unable to express */ |
|
|
| 176 | /* simple barrier semantics. That means we need to take out thor's hammer. */ |
|
|
| 177 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
288 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
| 178 | #endif |
289 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
290 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
| 179 | #endif |
291 | #endif |
| 180 | #endif |
292 | #endif |
| 181 | |
293 | |
| 182 | #ifndef ECB_MEMORY_FENCE |
294 | #ifndef ECB_MEMORY_FENCE |
| 183 | #if !ECB_AVOID_PTHREADS |
295 | #if !ECB_AVOID_PTHREADS |
| … | |
… | |
| 203 | |
315 | |
| 204 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
316 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
| 205 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
317 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
| 206 | #endif |
318 | #endif |
| 207 | |
319 | |
|
|
320 | #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE |
|
|
321 | #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ |
|
|
322 | #endif |
|
|
323 | |
| 208 | /*****************************************************************************/ |
324 | /*****************************************************************************/ |
| 209 | |
325 | |
| 210 | #if __cplusplus |
326 | #if ECB_CPP |
| 211 | #define ecb_inline static inline |
327 | #define ecb_inline static inline |
| 212 | #elif ECB_GCC_VERSION(2,5) |
328 | #elif ECB_GCC_VERSION(2,5) |
| 213 | #define ecb_inline static __inline__ |
329 | #define ecb_inline static __inline__ |
| 214 | #elif ECB_C99 |
330 | #elif ECB_C99 |
| 215 | #define ecb_inline static inline |
331 | #define ecb_inline static inline |
| … | |
… | |
| 229 | |
345 | |
| 230 | #define ECB_CONCAT_(a, b) a ## b |
346 | #define ECB_CONCAT_(a, b) a ## b |
| 231 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
347 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
| 232 | #define ECB_STRINGIFY_(a) # a |
348 | #define ECB_STRINGIFY_(a) # a |
| 233 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
349 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
350 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
| 234 | |
351 | |
| 235 | #define ecb_function_ ecb_inline |
352 | #define ecb_function_ ecb_inline |
| 236 | |
353 | |
| 237 | #if ECB_GCC_VERSION(3,1) |
354 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
| 238 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
355 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
356 | #else |
|
|
357 | #define ecb_attribute(attrlist) |
|
|
358 | #endif |
|
|
359 | |
|
|
360 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
| 239 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
361 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
362 | #else |
|
|
363 | /* possible C11 impl for integral types |
|
|
364 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
365 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
366 | |
|
|
367 | #define ecb_is_constant(expr) 0 |
|
|
368 | #endif |
|
|
369 | |
|
|
370 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
| 240 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
371 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
372 | #else |
|
|
373 | #define ecb_expect(expr,value) (expr) |
|
|
374 | #endif |
|
|
375 | |
|
|
376 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
| 241 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
377 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
| 242 | #else |
378 | #else |
| 243 | #define ecb_attribute(attrlist) |
|
|
| 244 | #define ecb_is_constant(expr) 0 |
|
|
| 245 | #define ecb_expect(expr,value) (expr) |
|
|
| 246 | #define ecb_prefetch(addr,rw,locality) |
379 | #define ecb_prefetch(addr,rw,locality) |
| 247 | #endif |
380 | #endif |
| 248 | |
381 | |
| 249 | /* no emulation for ecb_decltype */ |
382 | /* no emulation for ecb_decltype */ |
| 250 | #if ECB_GCC_VERSION(4,5) |
383 | #if ECB_CPP11 |
|
|
384 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
385 | template<class T> struct ecb_decltype_t { typedef T type; }; |
| 251 | #define ecb_decltype(x) __decltype(x) |
386 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
| 252 | #elif ECB_GCC_VERSION(3,0) |
387 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
| 253 | #define ecb_decltype(x) __typeof(x) |
388 | #define ecb_decltype(x) __typeof__ (x) |
| 254 | #endif |
389 | #endif |
| 255 | |
390 | |
|
|
391 | #if _MSC_VER >= 1300 |
|
|
392 | #define ecb_deprecated __declspec (deprecated) |
|
|
393 | #else |
|
|
394 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
|
|
395 | #endif |
|
|
396 | |
|
|
397 | #if _MSC_VER >= 1500 |
|
|
398 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
|
|
399 | #elif ECB_GCC_VERSION(4,5) |
|
|
400 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
|
|
401 | #else |
|
|
402 | #define ecb_deprecated_message(msg) ecb_deprecated |
|
|
403 | #endif |
|
|
404 | |
|
|
405 | #if _MSC_VER >= 1400 |
|
|
406 | #define ecb_noinline __declspec (noinline) |
|
|
407 | #else |
| 256 | #define ecb_noinline ecb_attribute ((__noinline__)) |
408 | #define ecb_noinline ecb_attribute ((__noinline__)) |
|
|
409 | #endif |
|
|
410 | |
| 257 | #define ecb_unused ecb_attribute ((__unused__)) |
411 | #define ecb_unused ecb_attribute ((__unused__)) |
| 258 | #define ecb_const ecb_attribute ((__const__)) |
412 | #define ecb_const ecb_attribute ((__const__)) |
| 259 | #define ecb_pure ecb_attribute ((__pure__)) |
413 | #define ecb_pure ecb_attribute ((__pure__)) |
| 260 | |
414 | |
| 261 | #if ECB_C11 |
415 | #if ECB_C11 || __IBMC_NORETURN |
|
|
416 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
| 262 | #define ecb_noreturn _Noreturn |
417 | #define ecb_noreturn _Noreturn |
|
|
418 | #elif ECB_CPP11 |
|
|
419 | #define ecb_noreturn [[noreturn]] |
|
|
420 | #elif _MSC_VER >= 1200 |
|
|
421 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
|
|
422 | #define ecb_noreturn __declspec (noreturn) |
| 263 | #else |
423 | #else |
| 264 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
424 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
| 265 | #endif |
425 | #endif |
| 266 | |
426 | |
| 267 | #if ECB_GCC_VERSION(4,3) |
427 | #if ECB_GCC_VERSION(4,3) |
| … | |
… | |
| 282 | /* for compatibility to the rest of the world */ |
442 | /* for compatibility to the rest of the world */ |
| 283 | #define ecb_likely(expr) ecb_expect_true (expr) |
443 | #define ecb_likely(expr) ecb_expect_true (expr) |
| 284 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
444 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
| 285 | |
445 | |
| 286 | /* count trailing zero bits and count # of one bits */ |
446 | /* count trailing zero bits and count # of one bits */ |
| 287 | #if ECB_GCC_VERSION(3,4) |
447 | #if ECB_GCC_VERSION(3,4) \ |
|
|
448 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
449 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
450 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
| 288 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
451 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
| 289 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
452 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
| 290 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
453 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
| 291 | #define ecb_ctz32(x) __builtin_ctz (x) |
454 | #define ecb_ctz32(x) __builtin_ctz (x) |
| 292 | #define ecb_ctz64(x) __builtin_ctzll (x) |
455 | #define ecb_ctz64(x) __builtin_ctzll (x) |
| 293 | #define ecb_popcount32(x) __builtin_popcount (x) |
456 | #define ecb_popcount32(x) __builtin_popcount (x) |
| 294 | /* no popcountll */ |
457 | /* no popcountll */ |
| 295 | #else |
458 | #else |
| 296 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
459 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
| 297 | ecb_function_ int |
460 | ecb_function_ ecb_const int |
| 298 | ecb_ctz32 (uint32_t x) |
461 | ecb_ctz32 (uint32_t x) |
| 299 | { |
462 | { |
|
|
463 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
464 | unsigned long r; |
|
|
465 | _BitScanForward (&r, x); |
|
|
466 | return (int)r; |
|
|
467 | #else |
| 300 | int r = 0; |
468 | int r = 0; |
| 301 | |
469 | |
| 302 | x &= ~x + 1; /* this isolates the lowest bit */ |
470 | x &= ~x + 1; /* this isolates the lowest bit */ |
| 303 | |
471 | |
| 304 | #if ECB_branchless_on_i386 |
472 | #if ECB_branchless_on_i386 |
| … | |
… | |
| 314 | if (x & 0xff00ff00) r += 8; |
482 | if (x & 0xff00ff00) r += 8; |
| 315 | if (x & 0xffff0000) r += 16; |
483 | if (x & 0xffff0000) r += 16; |
| 316 | #endif |
484 | #endif |
| 317 | |
485 | |
| 318 | return r; |
486 | return r; |
|
|
487 | #endif |
| 319 | } |
488 | } |
| 320 | |
489 | |
| 321 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
490 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
| 322 | ecb_function_ int |
491 | ecb_function_ ecb_const int |
| 323 | ecb_ctz64 (uint64_t x) |
492 | ecb_ctz64 (uint64_t x) |
| 324 | { |
493 | { |
|
|
494 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
495 | unsigned long r; |
|
|
496 | _BitScanForward64 (&r, x); |
|
|
497 | return (int)r; |
|
|
498 | #else |
| 325 | int shift = x & 0xffffffffU ? 0 : 32; |
499 | int shift = x & 0xffffffff ? 0 : 32; |
| 326 | return ecb_ctz32 (x >> shift) + shift; |
500 | return ecb_ctz32 (x >> shift) + shift; |
|
|
501 | #endif |
| 327 | } |
502 | } |
| 328 | |
503 | |
| 329 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
504 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
| 330 | ecb_function_ int |
505 | ecb_function_ ecb_const int |
| 331 | ecb_popcount32 (uint32_t x) |
506 | ecb_popcount32 (uint32_t x) |
| 332 | { |
507 | { |
| 333 | x -= (x >> 1) & 0x55555555; |
508 | x -= (x >> 1) & 0x55555555; |
| 334 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
509 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
| 335 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
510 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
| 336 | x *= 0x01010101; |
511 | x *= 0x01010101; |
| 337 | |
512 | |
| 338 | return x >> 24; |
513 | return x >> 24; |
| 339 | } |
514 | } |
| 340 | |
515 | |
| 341 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
516 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
| 342 | ecb_function_ int ecb_ld32 (uint32_t x) |
517 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
| 343 | { |
518 | { |
|
|
519 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
520 | unsigned long r; |
|
|
521 | _BitScanReverse (&r, x); |
|
|
522 | return (int)r; |
|
|
523 | #else |
| 344 | int r = 0; |
524 | int r = 0; |
| 345 | |
525 | |
| 346 | if (x >> 16) { x >>= 16; r += 16; } |
526 | if (x >> 16) { x >>= 16; r += 16; } |
| 347 | if (x >> 8) { x >>= 8; r += 8; } |
527 | if (x >> 8) { x >>= 8; r += 8; } |
| 348 | if (x >> 4) { x >>= 4; r += 4; } |
528 | if (x >> 4) { x >>= 4; r += 4; } |
| 349 | if (x >> 2) { x >>= 2; r += 2; } |
529 | if (x >> 2) { x >>= 2; r += 2; } |
| 350 | if (x >> 1) { r += 1; } |
530 | if (x >> 1) { r += 1; } |
| 351 | |
531 | |
| 352 | return r; |
532 | return r; |
|
|
533 | #endif |
| 353 | } |
534 | } |
| 354 | |
535 | |
| 355 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
536 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
| 356 | ecb_function_ int ecb_ld64 (uint64_t x) |
537 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
| 357 | { |
538 | { |
|
|
539 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
540 | unsigned long r; |
|
|
541 | _BitScanReverse64 (&r, x); |
|
|
542 | return (int)r; |
|
|
543 | #else |
| 358 | int r = 0; |
544 | int r = 0; |
| 359 | |
545 | |
| 360 | if (x >> 32) { x >>= 32; r += 32; } |
546 | if (x >> 32) { x >>= 32; r += 32; } |
| 361 | |
547 | |
| 362 | return r + ecb_ld32 (x); |
548 | return r + ecb_ld32 (x); |
| 363 | } |
|
|
| 364 | #endif |
549 | #endif |
|
|
550 | } |
|
|
551 | #endif |
| 365 | |
552 | |
| 366 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
553 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
| 367 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
554 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
| 368 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
555 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
| 369 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
556 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
| 370 | |
557 | |
| 371 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
558 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
| 372 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
559 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
| 373 | { |
560 | { |
| 374 | return ( (x * 0x0802U & 0x22110U) |
561 | return ( (x * 0x0802U & 0x22110U) |
| 375 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
562 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
| 376 | } |
563 | } |
| 377 | |
564 | |
| 378 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
565 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
| 379 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
566 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
| 380 | { |
567 | { |
| 381 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
568 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
| 382 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
569 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
| 383 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
570 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
| 384 | x = ( x >> 8 ) | ( x << 8); |
571 | x = ( x >> 8 ) | ( x << 8); |
| 385 | |
572 | |
| 386 | return x; |
573 | return x; |
| 387 | } |
574 | } |
| 388 | |
575 | |
| 389 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
576 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
| 390 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
577 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
| 391 | { |
578 | { |
| 392 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
579 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
| 393 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
580 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
| 394 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
581 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
| 395 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
582 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
| … | |
… | |
| 398 | return x; |
585 | return x; |
| 399 | } |
586 | } |
| 400 | |
587 | |
| 401 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
588 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 402 | /* so for this version we are lazy */ |
589 | /* so for this version we are lazy */ |
| 403 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
590 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
| 404 | ecb_function_ int |
591 | ecb_function_ ecb_const int |
| 405 | ecb_popcount64 (uint64_t x) |
592 | ecb_popcount64 (uint64_t x) |
| 406 | { |
593 | { |
| 407 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
594 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
| 408 | } |
595 | } |
| 409 | |
596 | |
| 410 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
597 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
| 411 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
598 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
| 412 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
599 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
| 413 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
600 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
| 414 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
601 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
| 415 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
602 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
| 416 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
603 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
| 417 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
604 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
| 418 | |
605 | |
| 419 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
606 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
| 420 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
607 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
| 421 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
608 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
| 422 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
609 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
| 423 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
610 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
| 424 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
611 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
| 425 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
612 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
| 426 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
613 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
| 427 | |
614 | |
| 428 | #if ECB_GCC_VERSION(4,3) |
615 | #if ECB_CPP |
|
|
616 | |
|
|
617 | inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } |
|
|
618 | inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } |
|
|
619 | inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); } |
|
|
620 | inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); } |
|
|
621 | |
|
|
622 | inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); } |
|
|
623 | inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); } |
|
|
624 | inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); } |
|
|
625 | inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); } |
|
|
626 | |
|
|
627 | inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); } |
|
|
628 | inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); } |
|
|
629 | inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); } |
|
|
630 | inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); } |
|
|
631 | |
|
|
632 | inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); } |
|
|
633 | inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); } |
|
|
634 | inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); } |
|
|
635 | inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); } |
|
|
636 | |
|
|
637 | inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); } |
|
|
638 | inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); } |
|
|
639 | inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); } |
|
|
640 | |
|
|
641 | inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); } |
|
|
642 | inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); } |
|
|
643 | inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); } |
|
|
644 | inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); } |
|
|
645 | |
|
|
646 | inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); } |
|
|
647 | inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); } |
|
|
648 | inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); } |
|
|
649 | inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); } |
|
|
650 | |
|
|
651 | #endif |
|
|
652 | |
|
|
653 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
|
|
654 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
|
|
655 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
|
|
656 | #else |
| 429 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
657 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
658 | #endif |
| 430 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
659 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
| 431 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
660 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
|
|
661 | #elif _MSC_VER |
|
|
662 | #include <stdlib.h> |
|
|
663 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
|
|
664 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
|
|
665 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
| 432 | #else |
666 | #else |
| 433 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
667 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
| 434 | ecb_function_ uint16_t |
668 | ecb_function_ ecb_const uint16_t |
| 435 | ecb_bswap16 (uint16_t x) |
669 | ecb_bswap16 (uint16_t x) |
| 436 | { |
670 | { |
| 437 | return ecb_rotl16 (x, 8); |
671 | return ecb_rotl16 (x, 8); |
| 438 | } |
672 | } |
| 439 | |
673 | |
| 440 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
674 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
| 441 | ecb_function_ uint32_t |
675 | ecb_function_ ecb_const uint32_t |
| 442 | ecb_bswap32 (uint32_t x) |
676 | ecb_bswap32 (uint32_t x) |
| 443 | { |
677 | { |
| 444 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
678 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
| 445 | } |
679 | } |
| 446 | |
680 | |
| 447 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
681 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
| 448 | ecb_function_ uint64_t |
682 | ecb_function_ ecb_const uint64_t |
| 449 | ecb_bswap64 (uint64_t x) |
683 | ecb_bswap64 (uint64_t x) |
| 450 | { |
684 | { |
| 451 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
685 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 452 | } |
686 | } |
| 453 | #endif |
687 | #endif |
| 454 | |
688 | |
| 455 | #if ECB_GCC_VERSION(4,5) |
689 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
| 456 | #define ecb_unreachable() __builtin_unreachable () |
690 | #define ecb_unreachable() __builtin_unreachable () |
| 457 | #else |
691 | #else |
| 458 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
692 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 459 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
693 | ecb_inline ecb_noreturn void ecb_unreachable (void); |
| 460 | ecb_inline void ecb_unreachable (void) { } |
694 | ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
| 461 | #endif |
695 | #endif |
| 462 | |
696 | |
| 463 | /* try to tell the compiler that some condition is definitely true */ |
697 | /* try to tell the compiler that some condition is definitely true */ |
| 464 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
698 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 465 | |
699 | |
| 466 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
700 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
| 467 | ecb_inline unsigned char |
701 | ecb_inline ecb_const uint32_t |
| 468 | ecb_byteorder_helper (void) |
702 | ecb_byteorder_helper (void) |
| 469 | { |
703 | { |
| 470 | const uint32_t u = 0x11223344; |
704 | /* the union code still generates code under pressure in gcc, */ |
| 471 | return *(unsigned char *)&u; |
705 | /* but less than using pointers, and always seems to */ |
|
|
706 | /* successfully return a constant. */ |
|
|
707 | /* the reason why we have this horrible preprocessor mess */ |
|
|
708 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
709 | /* or when using a recent enough gcc version (>= 4.6) */ |
|
|
710 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
|
|
711 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
|
|
712 | #define ECB_LITTLE_ENDIAN 1 |
|
|
713 | return 0x44332211; |
|
|
714 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
|
|
715 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
|
|
716 | #define ECB_BIG_ENDIAN 1 |
|
|
717 | return 0x11223344; |
|
|
718 | #else |
|
|
719 | union |
|
|
720 | { |
|
|
721 | uint8_t c[4]; |
|
|
722 | uint32_t u; |
|
|
723 | } u = { 0x11, 0x22, 0x33, 0x44 }; |
|
|
724 | return u.u; |
|
|
725 | #endif |
| 472 | } |
726 | } |
| 473 | |
727 | |
| 474 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
728 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
| 475 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
729 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
| 476 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
730 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
| 477 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
731 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
|
|
732 | |
|
|
733 | /*****************************************************************************/ |
|
|
734 | /* unaligned load/store */ |
|
|
735 | |
|
|
736 | ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
|
|
737 | ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
|
|
738 | ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
|
|
739 | |
|
|
740 | ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
|
|
741 | ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
|
|
742 | ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
|
|
743 | |
|
|
744 | ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
745 | ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
746 | ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
747 | |
|
|
748 | ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); } |
|
|
749 | ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); } |
|
|
750 | ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); } |
|
|
751 | |
|
|
752 | ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); } |
|
|
753 | ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); } |
|
|
754 | ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); } |
|
|
755 | |
|
|
756 | ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
|
|
757 | ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
|
|
758 | ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
|
|
759 | |
|
|
760 | ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
|
|
761 | ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
|
|
762 | ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
|
|
763 | |
|
|
764 | ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
765 | ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
766 | ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
767 | |
|
|
768 | 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)); } |
|
|
769 | 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)); } |
|
|
770 | 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)); } |
|
|
771 | |
|
|
772 | 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)); } |
|
|
773 | 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)); } |
|
|
774 | 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)); } |
|
|
775 | |
|
|
776 | #if ECB_CPP |
|
|
777 | |
|
|
778 | inline uint8_t ecb_bswap (uint8_t v) { return v; } |
|
|
779 | inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); } |
|
|
780 | inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); } |
|
|
781 | inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); } |
|
|
782 | |
|
|
783 | template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
|
|
784 | template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
|
|
785 | template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; } |
|
|
786 | template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); } |
|
|
787 | template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); } |
|
|
788 | template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
789 | template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); } |
|
|
790 | template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); } |
|
|
791 | |
|
|
792 | template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
|
|
793 | template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
|
|
794 | template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; } |
|
|
795 | template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); } |
|
|
796 | template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); } |
|
|
797 | template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
798 | template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } |
|
|
799 | template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } |
|
|
800 | |
|
|
801 | #endif |
|
|
802 | |
|
|
803 | /*****************************************************************************/ |
|
|
804 | /* division */ |
| 478 | |
805 | |
| 479 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
806 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
|
|
807 | /* C99 tightened the definition of %, so we can use a more efficient version */ |
| 480 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
808 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 481 | #else |
809 | #else |
| 482 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
810 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
| 483 | #endif |
811 | #endif |
| 484 | |
812 | |
| 485 | #if __cplusplus |
813 | #if ECB_CPP |
| 486 | template<typename T> |
814 | template<typename T> |
| 487 | static inline T ecb_div_rd (T val, T div) |
815 | static inline T ecb_div_rd (T val, T div) |
| 488 | { |
816 | { |
| 489 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
817 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
| 490 | } |
818 | } |
| … | |
… | |
| 495 | } |
823 | } |
| 496 | #else |
824 | #else |
| 497 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
825 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
| 498 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
826 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 499 | #endif |
827 | #endif |
|
|
828 | |
|
|
829 | /*****************************************************************************/ |
|
|
830 | /* array length */ |
| 500 | |
831 | |
| 501 | #if ecb_cplusplus_does_not_suck |
832 | #if ecb_cplusplus_does_not_suck |
| 502 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
833 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
| 503 | template<typename T, int N> |
834 | template<typename T, int N> |
| 504 | static inline int ecb_array_length (const T (&arr)[N]) |
835 | static inline int ecb_array_length (const T (&arr)[N]) |
| … | |
… | |
| 507 | } |
838 | } |
| 508 | #else |
839 | #else |
| 509 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
840 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 510 | #endif |
841 | #endif |
| 511 | |
842 | |
|
|
843 | /*****************************************************************************/ |
|
|
844 | /* IEEE 754-2008 half float conversions */ |
|
|
845 | |
|
|
846 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
|
|
847 | ecb_function_ ecb_const uint32_t |
|
|
848 | ecb_binary16_to_binary32 (uint32_t x) |
|
|
849 | { |
|
|
850 | unsigned int s = (x & 0x8000) << (31 - 15); |
|
|
851 | int e = (x >> 10) & 0x001f; |
|
|
852 | unsigned int m = x & 0x03ff; |
|
|
853 | |
|
|
854 | if (ecb_expect_false (e == 31)) |
|
|
855 | /* infinity or NaN */ |
|
|
856 | e = 255 - (127 - 15); |
|
|
857 | else if (ecb_expect_false (!e)) |
|
|
858 | { |
|
|
859 | if (ecb_expect_true (!m)) |
|
|
860 | /* zero, handled by code below by forcing e to 0 */ |
|
|
861 | e = 0 - (127 - 15); |
|
|
862 | else |
|
|
863 | { |
|
|
864 | /* subnormal, renormalise */ |
|
|
865 | unsigned int s = 10 - ecb_ld32 (m); |
|
|
866 | |
|
|
867 | m = (m << s) & 0x3ff; /* mask implicit bit */ |
|
|
868 | e -= s - 1; |
|
|
869 | } |
|
|
870 | } |
|
|
871 | |
|
|
872 | /* e and m now are normalised, or zero, (or inf or nan) */ |
|
|
873 | e += 127 - 15; |
|
|
874 | |
|
|
875 | return s | (e << 23) | (m << (23 - 10)); |
|
|
876 | } |
|
|
877 | |
|
|
878 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
|
|
879 | ecb_function_ ecb_const uint16_t |
|
|
880 | ecb_binary32_to_binary16 (uint32_t x) |
|
|
881 | { |
|
|
882 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
|
|
883 | int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
|
|
884 | unsigned int m = x & 0x007fffff; |
|
|
885 | |
|
|
886 | x &= 0x7fffffff; |
|
|
887 | |
|
|
888 | /* if it's within range of binary16 normals, use fast path */ |
|
|
889 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
|
|
890 | { |
|
|
891 | /* mantissa round-to-even */ |
|
|
892 | m += 0x00000fff + ((m >> (23 - 10)) & 1); |
|
|
893 | |
|
|
894 | /* handle overflow */ |
|
|
895 | if (ecb_expect_false (m >= 0x00800000)) |
|
|
896 | { |
|
|
897 | m >>= 1; |
|
|
898 | e += 1; |
|
|
899 | } |
|
|
900 | |
|
|
901 | return s | (e << 10) | (m >> (23 - 10)); |
|
|
902 | } |
|
|
903 | |
|
|
904 | /* handle large numbers and infinity */ |
|
|
905 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
|
|
906 | return s | 0x7c00; |
|
|
907 | |
|
|
908 | /* handle zero, subnormals and small numbers */ |
|
|
909 | if (ecb_expect_true (x < 0x38800000)) |
|
|
910 | { |
|
|
911 | /* zero */ |
|
|
912 | if (ecb_expect_true (!x)) |
|
|
913 | return s; |
|
|
914 | |
|
|
915 | /* handle subnormals */ |
|
|
916 | |
|
|
917 | /* too small, will be zero */ |
|
|
918 | if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
|
|
919 | return s; |
|
|
920 | |
|
|
921 | m |= 0x00800000; /* make implicit bit explicit */ |
|
|
922 | |
|
|
923 | /* very tricky - we need to round to the nearest e (+10) bit value */ |
|
|
924 | { |
|
|
925 | unsigned int bits = 14 - e; |
|
|
926 | unsigned int half = (1 << (bits - 1)) - 1; |
|
|
927 | unsigned int even = (m >> bits) & 1; |
|
|
928 | |
|
|
929 | /* if this overflows, we will end up with a normalised number */ |
|
|
930 | m = (m + half + even) >> bits; |
|
|
931 | } |
|
|
932 | |
|
|
933 | return s | m; |
|
|
934 | } |
|
|
935 | |
|
|
936 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
|
|
937 | m >>= 13; |
|
|
938 | |
|
|
939 | return s | 0x7c00 | m | !m; |
|
|
940 | } |
|
|
941 | |
|
|
942 | /*******************************************************************************/ |
|
|
943 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
|
|
944 | |
|
|
945 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
|
|
946 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
|
|
947 | #if 0 \ |
|
|
948 | || __i386 || __i386__ \ |
|
|
949 | || ECB_GCC_AMD64 \ |
|
|
950 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
|
|
951 | || defined __s390__ || defined __s390x__ \ |
|
|
952 | || defined __mips__ \ |
|
|
953 | || defined __alpha__ \ |
|
|
954 | || defined __hppa__ \ |
|
|
955 | || defined __ia64__ \ |
|
|
956 | || defined __m68k__ \ |
|
|
957 | || defined __m88k__ \ |
|
|
958 | || defined __sh__ \ |
|
|
959 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
|
|
960 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
961 | || defined __aarch64__ |
|
|
962 | #define ECB_STDFP 1 |
|
|
963 | #else |
|
|
964 | #define ECB_STDFP 0 |
|
|
965 | #endif |
|
|
966 | |
|
|
967 | #ifndef ECB_NO_LIBM |
|
|
968 | |
|
|
969 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
970 | |
|
|
971 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
972 | #ifdef INFINITY |
|
|
973 | #define ECB_INFINITY INFINITY |
|
|
974 | #else |
|
|
975 | #define ECB_INFINITY HUGE_VAL |
| 512 | #endif |
976 | #endif |
| 513 | |
977 | |
|
|
978 | #ifdef NAN |
|
|
979 | #define ECB_NAN NAN |
|
|
980 | #else |
|
|
981 | #define ECB_NAN ECB_INFINITY |
|
|
982 | #endif |
|
|
983 | |
|
|
984 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
|
|
985 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
986 | #define ecb_frexpf(x,e) frexpf ((x), (e)) |
|
|
987 | #else |
|
|
988 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
|
|
989 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
|
|
990 | #endif |
|
|
991 | |
|
|
992 | /* convert a float to ieee single/binary32 */ |
|
|
993 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
|
|
994 | ecb_function_ ecb_const uint32_t |
|
|
995 | ecb_float_to_binary32 (float x) |
|
|
996 | { |
|
|
997 | uint32_t r; |
|
|
998 | |
|
|
999 | #if ECB_STDFP |
|
|
1000 | memcpy (&r, &x, 4); |
|
|
1001 | #else |
|
|
1002 | /* slow emulation, works for anything but -0 */ |
|
|
1003 | uint32_t m; |
|
|
1004 | int e; |
|
|
1005 | |
|
|
1006 | if (x == 0e0f ) return 0x00000000U; |
|
|
1007 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
|
|
1008 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
|
|
1009 | if (x != x ) return 0x7fbfffffU; |
|
|
1010 | |
|
|
1011 | m = ecb_frexpf (x, &e) * 0x1000000U; |
|
|
1012 | |
|
|
1013 | r = m & 0x80000000U; |
|
|
1014 | |
|
|
1015 | if (r) |
|
|
1016 | m = -m; |
|
|
1017 | |
|
|
1018 | if (e <= -126) |
|
|
1019 | { |
|
|
1020 | m &= 0xffffffU; |
|
|
1021 | m >>= (-125 - e); |
|
|
1022 | e = -126; |
|
|
1023 | } |
|
|
1024 | |
|
|
1025 | r |= (e + 126) << 23; |
|
|
1026 | r |= m & 0x7fffffU; |
|
|
1027 | #endif |
|
|
1028 | |
|
|
1029 | return r; |
|
|
1030 | } |
|
|
1031 | |
|
|
1032 | /* converts an ieee single/binary32 to a float */ |
|
|
1033 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
|
|
1034 | ecb_function_ ecb_const float |
|
|
1035 | ecb_binary32_to_float (uint32_t x) |
|
|
1036 | { |
|
|
1037 | float r; |
|
|
1038 | |
|
|
1039 | #if ECB_STDFP |
|
|
1040 | memcpy (&r, &x, 4); |
|
|
1041 | #else |
|
|
1042 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
1043 | int neg = x >> 31; |
|
|
1044 | int e = (x >> 23) & 0xffU; |
|
|
1045 | |
|
|
1046 | x &= 0x7fffffU; |
|
|
1047 | |
|
|
1048 | if (e) |
|
|
1049 | x |= 0x800000U; |
|
|
1050 | else |
|
|
1051 | e = 1; |
|
|
1052 | |
|
|
1053 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
|
|
1054 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
|
|
1055 | |
|
|
1056 | r = neg ? -r : r; |
|
|
1057 | #endif |
|
|
1058 | |
|
|
1059 | return r; |
|
|
1060 | } |
|
|
1061 | |
|
|
1062 | /* convert a double to ieee double/binary64 */ |
|
|
1063 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
|
|
1064 | ecb_function_ ecb_const uint64_t |
|
|
1065 | ecb_double_to_binary64 (double x) |
|
|
1066 | { |
|
|
1067 | uint64_t r; |
|
|
1068 | |
|
|
1069 | #if ECB_STDFP |
|
|
1070 | memcpy (&r, &x, 8); |
|
|
1071 | #else |
|
|
1072 | /* slow emulation, works for anything but -0 */ |
|
|
1073 | uint64_t m; |
|
|
1074 | int e; |
|
|
1075 | |
|
|
1076 | if (x == 0e0 ) return 0x0000000000000000U; |
|
|
1077 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
|
|
1078 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
|
|
1079 | if (x != x ) return 0X7ff7ffffffffffffU; |
|
|
1080 | |
|
|
1081 | m = frexp (x, &e) * 0x20000000000000U; |
|
|
1082 | |
|
|
1083 | r = m & 0x8000000000000000;; |
|
|
1084 | |
|
|
1085 | if (r) |
|
|
1086 | m = -m; |
|
|
1087 | |
|
|
1088 | if (e <= -1022) |
|
|
1089 | { |
|
|
1090 | m &= 0x1fffffffffffffU; |
|
|
1091 | m >>= (-1021 - e); |
|
|
1092 | e = -1022; |
|
|
1093 | } |
|
|
1094 | |
|
|
1095 | r |= ((uint64_t)(e + 1022)) << 52; |
|
|
1096 | r |= m & 0xfffffffffffffU; |
|
|
1097 | #endif |
|
|
1098 | |
|
|
1099 | return r; |
|
|
1100 | } |
|
|
1101 | |
|
|
1102 | /* converts an ieee double/binary64 to a double */ |
|
|
1103 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
|
|
1104 | ecb_function_ ecb_const double |
|
|
1105 | ecb_binary64_to_double (uint64_t x) |
|
|
1106 | { |
|
|
1107 | double r; |
|
|
1108 | |
|
|
1109 | #if ECB_STDFP |
|
|
1110 | memcpy (&r, &x, 8); |
|
|
1111 | #else |
|
|
1112 | /* emulation, only works for normals and subnormals and +0 */ |
|
|
1113 | int neg = x >> 63; |
|
|
1114 | int e = (x >> 52) & 0x7ffU; |
|
|
1115 | |
|
|
1116 | x &= 0xfffffffffffffU; |
|
|
1117 | |
|
|
1118 | if (e) |
|
|
1119 | x |= 0x10000000000000U; |
|
|
1120 | else |
|
|
1121 | e = 1; |
|
|
1122 | |
|
|
1123 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
|
|
1124 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
|
|
1125 | |
|
|
1126 | r = neg ? -r : r; |
|
|
1127 | #endif |
|
|
1128 | |
|
|
1129 | return r; |
|
|
1130 | } |
|
|
1131 | |
|
|
1132 | /* convert a float to ieee half/binary16 */ |
|
|
1133 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
|
|
1134 | ecb_function_ ecb_const uint16_t |
|
|
1135 | ecb_float_to_binary16 (float x) |
|
|
1136 | { |
|
|
1137 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
|
|
1138 | } |
|
|
1139 | |
|
|
1140 | /* convert an ieee half/binary16 to float */ |
|
|
1141 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
1142 | ecb_function_ ecb_const float |
|
|
1143 | ecb_binary16_to_float (uint16_t x) |
|
|
1144 | { |
|
|
1145 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
|
|
1146 | } |
|
|
1147 | |
|
|
1148 | #endif |
|
|
1149 | |
|
|
1150 | #endif |
|
|
1151 | |
