1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2013 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2015 Marc Alexander Lehmann <libecb@schmorp.de> |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
6 | * All rights reserved. |
6 | * All rights reserved. |
7 | * |
7 | * |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
9 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
… | |
… | |
23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
23 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
24 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
25 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
26 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
27 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
28 | * |
|
|
29 | * Alternatively, the contents of this file may be used under the terms of |
|
|
30 | * the GNU General Public License ("GPL") version 2 or any later version, |
|
|
31 | * in which case the provisions of the GPL are applicable instead of |
|
|
32 | * the above. If you wish to allow the use of your version of this file |
|
|
33 | * only under the terms of the GPL and not to allow others to use your |
|
|
34 | * version of this file under the BSD license, indicate your decision |
|
|
35 | * by deleting the provisions above and replace them with the notice |
|
|
36 | * and other provisions required by the GPL. If you do not delete the |
|
|
37 | * provisions above, a recipient may use your version of this file under |
|
|
38 | * either the BSD or the GPL. |
28 | */ |
39 | */ |
29 | |
40 | |
30 | #ifndef ECB_H |
41 | #ifndef ECB_H |
31 | #define ECB_H |
42 | #define ECB_H |
32 | |
43 | |
33 | /* 16 bits major, 16 bits minor */ |
44 | /* 16 bits major, 16 bits minor */ |
34 | #define ECB_VERSION 0x00010003 |
45 | #define ECB_VERSION 0x00010005 |
35 | |
46 | |
36 | #ifdef _WIN32 |
47 | #ifdef _WIN32 |
37 | typedef signed char int8_t; |
48 | typedef signed char int8_t; |
38 | typedef unsigned char uint8_t; |
49 | typedef unsigned char uint8_t; |
39 | typedef signed short int16_t; |
50 | typedef signed short int16_t; |
… | |
… | |
56 | typedef uint32_t uintptr_t; |
67 | typedef uint32_t uintptr_t; |
57 | typedef int32_t intptr_t; |
68 | typedef int32_t intptr_t; |
58 | #endif |
69 | #endif |
59 | #else |
70 | #else |
60 | #include <inttypes.h> |
71 | #include <inttypes.h> |
61 | #if UINTMAX_MAX > 0xffffffffU |
72 | #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
62 | #define ECB_PTRSIZE 8 |
73 | #define ECB_PTRSIZE 8 |
63 | #else |
74 | #else |
64 | #define ECB_PTRSIZE 4 |
75 | #define ECB_PTRSIZE 4 |
65 | #endif |
76 | #endif |
66 | #endif |
77 | #endif |
67 | |
78 | |
|
|
79 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
|
|
80 | #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
|
|
81 | |
68 | /* work around x32 idiocy by defining proper macros */ |
82 | /* work around x32 idiocy by defining proper macros */ |
69 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
83 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
70 | #if _ILP32 |
84 | #if _ILP32 |
71 | #define ECB_AMD64_X32 1 |
85 | #define ECB_AMD64_X32 1 |
72 | #else |
86 | #else |
73 | #define ECB_AMD64 1 |
87 | #define ECB_AMD64 1 |
74 | #endif |
88 | #endif |
… | |
… | |
79 | * causing enormous grief in return for some better fake benchmark numbers. |
93 | * causing enormous grief in return for some better fake benchmark numbers. |
80 | * or so. |
94 | * or so. |
81 | * we try to detect these and simply assume they are not gcc - if they have |
95 | * we try to detect these and simply assume they are not gcc - if they have |
82 | * an issue with that they should have done it right in the first place. |
96 | * an issue with that they should have done it right in the first place. |
83 | */ |
97 | */ |
84 | #ifndef ECB_GCC_VERSION |
|
|
85 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
98 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
86 | #define ECB_GCC_VERSION(major,minor) 0 |
99 | #define ECB_GCC_VERSION(major,minor) 0 |
87 | #else |
100 | #else |
88 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
101 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
89 | #endif |
102 | #endif |
|
|
103 | |
|
|
104 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
|
|
105 | |
|
|
106 | #if __clang__ && defined __has_builtin |
|
|
107 | #define ECB_CLANG_BUILTIN(x) __has_builtin (x) |
|
|
108 | #else |
|
|
109 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
110 | #endif |
|
|
111 | |
|
|
112 | #if __clang__ && defined __has_extension |
|
|
113 | #define ECB_CLANG_EXTENSION(x) __has_extension (x) |
|
|
114 | #else |
|
|
115 | #define ECB_CLANG_EXTENSION(x) 0 |
90 | #endif |
116 | #endif |
91 | |
117 | |
92 | #define ECB_CPP (__cplusplus+0) |
118 | #define ECB_CPP (__cplusplus+0) |
93 | #define ECB_CPP11 (__cplusplus >= 201103L) |
119 | #define ECB_CPP11 (__cplusplus >= 201103L) |
94 | |
120 | |
… | |
… | |
122 | #define ECB_NO_SMP 1 |
148 | #define ECB_NO_SMP 1 |
123 | #endif |
149 | #endif |
124 | |
150 | |
125 | #if ECB_NO_SMP |
151 | #if ECB_NO_SMP |
126 | #define ECB_MEMORY_FENCE do { } while (0) |
152 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
153 | #endif |
|
|
154 | |
|
|
155 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
|
|
156 | #if __xlC__ && ECB_CPP |
|
|
157 | #include <builtins.h> |
|
|
158 | #endif |
|
|
159 | |
|
|
160 | #if 1400 <= _MSC_VER |
|
|
161 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
127 | #endif |
162 | #endif |
128 | |
163 | |
129 | #ifndef ECB_MEMORY_FENCE |
164 | #ifndef ECB_MEMORY_FENCE |
130 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
165 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
131 | #if __i386 || __i386__ |
166 | #if __i386 || __i386__ |
132 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
167 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
133 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
168 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
134 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
169 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
135 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
170 | #elif ECB_GCC_AMD64 |
136 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
137 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
172 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
138 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
173 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
139 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
174 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
175 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
176 | #elif defined __ARM_ARCH_2__ \ |
|
|
177 | || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
|
|
178 | || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
|
|
179 | || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
|
|
180 | || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
|
|
181 | || defined __ARM_ARCH_5TEJ__ |
|
|
182 | /* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
141 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
183 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
142 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
184 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
|
|
185 | || defined __ARM_ARCH_6T2__ |
143 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
186 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
144 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
187 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
145 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
188 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
146 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
189 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
190 | #elif __aarch64__ |
|
|
191 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
147 | #elif (__sparc || __sparc__) && !__sparcv8 |
192 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
193 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
149 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
194 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
150 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
195 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
151 | #elif defined __s390__ || defined __s390x__ |
196 | #elif defined __s390__ || defined __s390x__ |
152 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
197 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
… | |
… | |
173 | |
218 | |
174 | #ifndef ECB_MEMORY_FENCE |
219 | #ifndef ECB_MEMORY_FENCE |
175 | #if ECB_GCC_VERSION(4,7) |
220 | #if ECB_GCC_VERSION(4,7) |
176 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
221 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
177 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
222 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
223 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
224 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
178 | |
225 | |
179 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
226 | #elif ECB_CLANG_EXTENSION(c_atomic) |
180 | * without risking compile time errors with other compilers. We *could* |
|
|
181 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
|
|
182 | * around this shit time and again. |
|
|
183 | * #elif defined __clang && __has_feature (cxx_atomic) |
|
|
184 | * // see comment below (stdatomic.h) about the C11 memory model. |
227 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
185 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
228 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
186 | */ |
229 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
230 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
187 | |
231 | |
188 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
232 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
189 | #define ECB_MEMORY_FENCE __sync_synchronize () |
233 | #define ECB_MEMORY_FENCE __sync_synchronize () |
190 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
234 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
191 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
235 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
… | |
… | |
220 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
264 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
221 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
265 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
222 | /* for most usages, or gcc and clang have a bug */ |
266 | /* for most usages, or gcc and clang have a bug */ |
223 | /* I *currently* lean towards the latter, and inefficiently implement */ |
267 | /* I *currently* lean towards the latter, and inefficiently implement */ |
224 | /* all three of ecb's fences as a seq_cst fence */ |
268 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
269 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
270 | /* for all __atomic_thread_fence's except seq_cst */ |
225 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
271 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
226 | #endif |
272 | #endif |
227 | #endif |
273 | #endif |
228 | |
274 | |
229 | #ifndef ECB_MEMORY_FENCE |
275 | #ifndef ECB_MEMORY_FENCE |
… | |
… | |
252 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
298 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
253 | #endif |
299 | #endif |
254 | |
300 | |
255 | /*****************************************************************************/ |
301 | /*****************************************************************************/ |
256 | |
302 | |
257 | #if __cplusplus |
303 | #if ECB_CPP |
258 | #define ecb_inline static inline |
304 | #define ecb_inline static inline |
259 | #elif ECB_GCC_VERSION(2,5) |
305 | #elif ECB_GCC_VERSION(2,5) |
260 | #define ecb_inline static __inline__ |
306 | #define ecb_inline static __inline__ |
261 | #elif ECB_C99 |
307 | #elif ECB_C99 |
262 | #define ecb_inline static inline |
308 | #define ecb_inline static inline |
… | |
… | |
276 | |
322 | |
277 | #define ECB_CONCAT_(a, b) a ## b |
323 | #define ECB_CONCAT_(a, b) a ## b |
278 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
324 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
279 | #define ECB_STRINGIFY_(a) # a |
325 | #define ECB_STRINGIFY_(a) # a |
280 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
326 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
327 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
281 | |
328 | |
282 | #define ecb_function_ ecb_inline |
329 | #define ecb_function_ ecb_inline |
283 | |
330 | |
284 | #if ECB_GCC_VERSION(3,1) |
331 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
285 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
332 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
333 | #else |
|
|
334 | #define ecb_attribute(attrlist) |
|
|
335 | #endif |
|
|
336 | |
|
|
337 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
286 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
338 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
287 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
288 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
|
|
289 | #else |
339 | #else |
290 | #define ecb_attribute(attrlist) |
|
|
291 | |
|
|
292 | /* possible C11 impl for integral types |
340 | /* possible C11 impl for integral types |
293 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
341 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
294 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
342 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
295 | |
343 | |
296 | #define ecb_is_constant(expr) 0 |
344 | #define ecb_is_constant(expr) 0 |
|
|
345 | #endif |
|
|
346 | |
|
|
347 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
|
|
348 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
349 | #else |
297 | #define ecb_expect(expr,value) (expr) |
350 | #define ecb_expect(expr,value) (expr) |
|
|
351 | #endif |
|
|
352 | |
|
|
353 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
|
|
354 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
|
|
355 | #else |
298 | #define ecb_prefetch(addr,rw,locality) |
356 | #define ecb_prefetch(addr,rw,locality) |
299 | #endif |
357 | #endif |
300 | |
358 | |
301 | /* no emulation for ecb_decltype */ |
359 | /* no emulation for ecb_decltype */ |
302 | #if ECB_GCC_VERSION(4,5) |
360 | #if ECB_CPP11 |
|
|
361 | // older implementations might have problems with decltype(x)::type, work around it |
|
|
362 | template<class T> struct ecb_decltype_t { typedef T type; }; |
303 | #define ecb_decltype(x) __decltype(x) |
363 | #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type |
304 | #elif ECB_GCC_VERSION(3,0) |
364 | #elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8) |
305 | #define ecb_decltype(x) __typeof(x) |
365 | #define ecb_decltype(x) __typeof__ (x) |
306 | #endif |
366 | #endif |
307 | |
367 | |
|
|
368 | #if _MSC_VER >= 1300 |
|
|
369 | #define ecb_deprecated __declspec (deprecated) |
|
|
370 | #else |
|
|
371 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
|
|
372 | #endif |
|
|
373 | |
|
|
374 | #if _MSC_VER >= 1500 |
|
|
375 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
|
|
376 | #elif ECB_GCC_VERSION(4,5) |
|
|
377 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
|
|
378 | #else |
|
|
379 | #define ecb_deprecated_message(msg) ecb_deprecated |
|
|
380 | #endif |
|
|
381 | |
|
|
382 | #if _MSC_VER >= 1400 |
|
|
383 | #define ecb_noinline __declspec (noinline) |
|
|
384 | #else |
308 | #define ecb_noinline ecb_attribute ((__noinline__)) |
385 | #define ecb_noinline ecb_attribute ((__noinline__)) |
|
|
386 | #endif |
|
|
387 | |
309 | #define ecb_unused ecb_attribute ((__unused__)) |
388 | #define ecb_unused ecb_attribute ((__unused__)) |
310 | #define ecb_const ecb_attribute ((__const__)) |
389 | #define ecb_const ecb_attribute ((__const__)) |
311 | #define ecb_pure ecb_attribute ((__pure__)) |
390 | #define ecb_pure ecb_attribute ((__pure__)) |
312 | |
391 | |
313 | #if ECB_C11 |
392 | #if ECB_C11 || __IBMC_NORETURN |
|
|
393 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
314 | #define ecb_noreturn _Noreturn |
394 | #define ecb_noreturn _Noreturn |
|
|
395 | #elif ECB_CPP11 |
|
|
396 | #define ecb_noreturn [[noreturn]] |
|
|
397 | #elif _MSC_VER >= 1200 |
|
|
398 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
|
|
399 | #define ecb_noreturn __declspec (noreturn) |
315 | #else |
400 | #else |
316 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
401 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
317 | #endif |
402 | #endif |
318 | |
403 | |
319 | #if ECB_GCC_VERSION(4,3) |
404 | #if ECB_GCC_VERSION(4,3) |
… | |
… | |
334 | /* for compatibility to the rest of the world */ |
419 | /* for compatibility to the rest of the world */ |
335 | #define ecb_likely(expr) ecb_expect_true (expr) |
420 | #define ecb_likely(expr) ecb_expect_true (expr) |
336 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
421 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
337 | |
422 | |
338 | /* count trailing zero bits and count # of one bits */ |
423 | /* count trailing zero bits and count # of one bits */ |
339 | #if ECB_GCC_VERSION(3,4) |
424 | #if ECB_GCC_VERSION(3,4) \ |
|
|
425 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
426 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
427 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
340 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
428 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
341 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
429 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
342 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
430 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
343 | #define ecb_ctz32(x) __builtin_ctz (x) |
431 | #define ecb_ctz32(x) __builtin_ctz (x) |
344 | #define ecb_ctz64(x) __builtin_ctzll (x) |
432 | #define ecb_ctz64(x) __builtin_ctzll (x) |
345 | #define ecb_popcount32(x) __builtin_popcount (x) |
433 | #define ecb_popcount32(x) __builtin_popcount (x) |
346 | /* no popcountll */ |
434 | /* no popcountll */ |
347 | #else |
435 | #else |
348 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
436 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
349 | ecb_function_ int |
437 | ecb_function_ ecb_const int |
350 | ecb_ctz32 (uint32_t x) |
438 | ecb_ctz32 (uint32_t x) |
351 | { |
439 | { |
|
|
440 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
441 | unsigned long r; |
|
|
442 | _BitScanForward (&r, x); |
|
|
443 | return (int)r; |
|
|
444 | #else |
352 | int r = 0; |
445 | int r = 0; |
353 | |
446 | |
354 | x &= ~x + 1; /* this isolates the lowest bit */ |
447 | x &= ~x + 1; /* this isolates the lowest bit */ |
355 | |
448 | |
356 | #if ECB_branchless_on_i386 |
449 | #if ECB_branchless_on_i386 |
… | |
… | |
366 | if (x & 0xff00ff00) r += 8; |
459 | if (x & 0xff00ff00) r += 8; |
367 | if (x & 0xffff0000) r += 16; |
460 | if (x & 0xffff0000) r += 16; |
368 | #endif |
461 | #endif |
369 | |
462 | |
370 | return r; |
463 | return r; |
|
|
464 | #endif |
371 | } |
465 | } |
372 | |
466 | |
373 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
467 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
374 | ecb_function_ int |
468 | ecb_function_ ecb_const int |
375 | ecb_ctz64 (uint64_t x) |
469 | ecb_ctz64 (uint64_t x) |
376 | { |
470 | { |
|
|
471 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
472 | unsigned long r; |
|
|
473 | _BitScanForward64 (&r, x); |
|
|
474 | return (int)r; |
|
|
475 | #else |
377 | int shift = x & 0xffffffffU ? 0 : 32; |
476 | int shift = x & 0xffffffff ? 0 : 32; |
378 | return ecb_ctz32 (x >> shift) + shift; |
477 | return ecb_ctz32 (x >> shift) + shift; |
|
|
478 | #endif |
379 | } |
479 | } |
380 | |
480 | |
381 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
481 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
382 | ecb_function_ int |
482 | ecb_function_ ecb_const int |
383 | ecb_popcount32 (uint32_t x) |
483 | ecb_popcount32 (uint32_t x) |
384 | { |
484 | { |
385 | x -= (x >> 1) & 0x55555555; |
485 | x -= (x >> 1) & 0x55555555; |
386 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
486 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
387 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
487 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
388 | x *= 0x01010101; |
488 | x *= 0x01010101; |
389 | |
489 | |
390 | return x >> 24; |
490 | return x >> 24; |
391 | } |
491 | } |
392 | |
492 | |
393 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
493 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
394 | ecb_function_ int ecb_ld32 (uint32_t x) |
494 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
395 | { |
495 | { |
|
|
496 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
497 | unsigned long r; |
|
|
498 | _BitScanReverse (&r, x); |
|
|
499 | return (int)r; |
|
|
500 | #else |
396 | int r = 0; |
501 | int r = 0; |
397 | |
502 | |
398 | if (x >> 16) { x >>= 16; r += 16; } |
503 | if (x >> 16) { x >>= 16; r += 16; } |
399 | if (x >> 8) { x >>= 8; r += 8; } |
504 | if (x >> 8) { x >>= 8; r += 8; } |
400 | if (x >> 4) { x >>= 4; r += 4; } |
505 | if (x >> 4) { x >>= 4; r += 4; } |
401 | if (x >> 2) { x >>= 2; r += 2; } |
506 | if (x >> 2) { x >>= 2; r += 2; } |
402 | if (x >> 1) { r += 1; } |
507 | if (x >> 1) { r += 1; } |
403 | |
508 | |
404 | return r; |
509 | return r; |
|
|
510 | #endif |
405 | } |
511 | } |
406 | |
512 | |
407 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
513 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
408 | ecb_function_ int ecb_ld64 (uint64_t x) |
514 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
409 | { |
515 | { |
|
|
516 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
517 | unsigned long r; |
|
|
518 | _BitScanReverse64 (&r, x); |
|
|
519 | return (int)r; |
|
|
520 | #else |
410 | int r = 0; |
521 | int r = 0; |
411 | |
522 | |
412 | if (x >> 32) { x >>= 32; r += 32; } |
523 | if (x >> 32) { x >>= 32; r += 32; } |
413 | |
524 | |
414 | return r + ecb_ld32 (x); |
525 | return r + ecb_ld32 (x); |
415 | } |
|
|
416 | #endif |
526 | #endif |
|
|
527 | } |
|
|
528 | #endif |
417 | |
529 | |
418 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
530 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
419 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
531 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
420 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
532 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); |
421 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
533 | ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
422 | |
534 | |
423 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
535 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); |
424 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
536 | ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) |
425 | { |
537 | { |
426 | return ( (x * 0x0802U & 0x22110U) |
538 | return ( (x * 0x0802U & 0x22110U) |
427 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
539 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
428 | } |
540 | } |
429 | |
541 | |
430 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
542 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); |
431 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
543 | ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) |
432 | { |
544 | { |
433 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
545 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
434 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
546 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
435 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
547 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
436 | x = ( x >> 8 ) | ( x << 8); |
548 | x = ( x >> 8 ) | ( x << 8); |
437 | |
549 | |
438 | return x; |
550 | return x; |
439 | } |
551 | } |
440 | |
552 | |
441 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
553 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); |
442 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
554 | ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) |
443 | { |
555 | { |
444 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
556 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
445 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
557 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
446 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
558 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
447 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
559 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
… | |
… | |
450 | return x; |
562 | return x; |
451 | } |
563 | } |
452 | |
564 | |
453 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
565 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
454 | /* so for this version we are lazy */ |
566 | /* so for this version we are lazy */ |
455 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
567 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
456 | ecb_function_ int |
568 | ecb_function_ ecb_const int |
457 | ecb_popcount64 (uint64_t x) |
569 | ecb_popcount64 (uint64_t x) |
458 | { |
570 | { |
459 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
571 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
460 | } |
572 | } |
461 | |
573 | |
462 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
574 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
463 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
575 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
464 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
576 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
465 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
577 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); |
466 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
578 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
467 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
579 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
468 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
580 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
469 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
581 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
470 | |
582 | |
471 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
583 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
472 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
584 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
473 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
585 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
474 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
586 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
475 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
587 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
476 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
588 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
477 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
589 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
478 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
590 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
479 | |
591 | |
480 | #if ECB_GCC_VERSION(4,3) |
592 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
|
|
593 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
|
|
594 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
|
|
595 | #else |
481 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
596 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
597 | #endif |
482 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
598 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
483 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
599 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
|
|
600 | #elif _MSC_VER |
|
|
601 | #include <stdlib.h> |
|
|
602 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
|
|
603 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
|
|
604 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
484 | #else |
605 | #else |
485 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
606 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
486 | ecb_function_ uint16_t |
607 | ecb_function_ ecb_const uint16_t |
487 | ecb_bswap16 (uint16_t x) |
608 | ecb_bswap16 (uint16_t x) |
488 | { |
609 | { |
489 | return ecb_rotl16 (x, 8); |
610 | return ecb_rotl16 (x, 8); |
490 | } |
611 | } |
491 | |
612 | |
492 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
613 | ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); |
493 | ecb_function_ uint32_t |
614 | ecb_function_ ecb_const uint32_t |
494 | ecb_bswap32 (uint32_t x) |
615 | ecb_bswap32 (uint32_t x) |
495 | { |
616 | { |
496 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
617 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
497 | } |
618 | } |
498 | |
619 | |
499 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
620 | ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); |
500 | ecb_function_ uint64_t |
621 | ecb_function_ ecb_const uint64_t |
501 | ecb_bswap64 (uint64_t x) |
622 | ecb_bswap64 (uint64_t x) |
502 | { |
623 | { |
503 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
624 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
504 | } |
625 | } |
505 | #endif |
626 | #endif |
506 | |
627 | |
507 | #if ECB_GCC_VERSION(4,5) |
628 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
508 | #define ecb_unreachable() __builtin_unreachable () |
629 | #define ecb_unreachable() __builtin_unreachable () |
509 | #else |
630 | #else |
510 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
631 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
511 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
632 | ecb_inline ecb_noreturn void ecb_unreachable (void); |
512 | ecb_inline void ecb_unreachable (void) { } |
633 | ecb_inline ecb_noreturn void ecb_unreachable (void) { } |
513 | #endif |
634 | #endif |
514 | |
635 | |
515 | /* try to tell the compiler that some condition is definitely true */ |
636 | /* try to tell the compiler that some condition is definitely true */ |
516 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
637 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
517 | |
638 | |
518 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
639 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
519 | ecb_inline unsigned char |
640 | ecb_inline ecb_const uint32_t |
520 | ecb_byteorder_helper (void) |
641 | ecb_byteorder_helper (void) |
521 | { |
642 | { |
522 | /* the union code still generates code under pressure in gcc, */ |
643 | /* the union code still generates code under pressure in gcc, */ |
523 | /* but less than using pointers, and always seems to */ |
644 | /* but less than using pointers, and always seems to */ |
524 | /* successfully return a constant. */ |
645 | /* successfully return a constant. */ |
525 | /* the reason why we have this horrible preprocessor mess */ |
646 | /* the reason why we have this horrible preprocessor mess */ |
526 | /* is to avoid it in all cases, at least on common architectures */ |
647 | /* is to avoid it in all cases, at least on common architectures */ |
527 | /* or when using a recent enough gcc version (>= 4.6) */ |
648 | /* or when using a recent enough gcc version (>= 4.6) */ |
528 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
529 | return 0x44; |
|
|
530 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
649 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
|
|
650 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
|
|
651 | #define ECB_LITTLE_ENDIAN 1 |
531 | return 0x44; |
652 | return 0x44332211; |
532 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
653 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
|
|
654 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
|
|
655 | #define ECB_BIG_ENDIAN 1 |
533 | return 0x11; |
656 | return 0x11223344; |
534 | #else |
657 | #else |
535 | union |
658 | union |
536 | { |
659 | { |
|
|
660 | uint8_t c[4]; |
537 | uint32_t i; |
661 | uint32_t u; |
538 | uint8_t c; |
|
|
539 | } u = { 0x11223344 }; |
662 | } u = { 0x11, 0x22, 0x33, 0x44 }; |
540 | return u.c; |
663 | return u.u; |
541 | #endif |
664 | #endif |
542 | } |
665 | } |
543 | |
666 | |
544 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
667 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
545 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
668 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
546 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
669 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
547 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
670 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
548 | |
671 | |
549 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
672 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
550 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
673 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
551 | #else |
674 | #else |
552 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
675 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
553 | #endif |
676 | #endif |
554 | |
677 | |
555 | #if __cplusplus |
678 | #if ECB_CPP |
556 | template<typename T> |
679 | template<typename T> |
557 | static inline T ecb_div_rd (T val, T div) |
680 | static inline T ecb_div_rd (T val, T div) |
558 | { |
681 | { |
559 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
682 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
560 | } |
683 | } |
… | |
… | |
577 | } |
700 | } |
578 | #else |
701 | #else |
579 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
702 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
580 | #endif |
703 | #endif |
581 | |
704 | |
|
|
705 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
|
|
706 | ecb_function_ ecb_const uint32_t |
|
|
707 | ecb_binary16_to_binary32 (uint32_t x) |
|
|
708 | { |
|
|
709 | unsigned int s = (x & 0x8000) << (31 - 15); |
|
|
710 | int e = (x >> 10) & 0x001f; |
|
|
711 | unsigned int m = x & 0x03ff; |
|
|
712 | |
|
|
713 | if (ecb_expect_false (e == 31)) |
|
|
714 | /* infinity or NaN */ |
|
|
715 | e = 255 - (127 - 15); |
|
|
716 | else if (ecb_expect_false (!e)) |
|
|
717 | { |
|
|
718 | if (ecb_expect_true (!m)) |
|
|
719 | /* zero, handled by code below by forcing e to 0 */ |
|
|
720 | e = 0 - (127 - 15); |
|
|
721 | else |
|
|
722 | { |
|
|
723 | /* subnormal, renormalise */ |
|
|
724 | unsigned int s = 10 - ecb_ld32 (m); |
|
|
725 | |
|
|
726 | m = (m << s) & 0x3ff; /* mask implicit bit */ |
|
|
727 | e -= s - 1; |
|
|
728 | } |
|
|
729 | } |
|
|
730 | |
|
|
731 | /* e and m now are normalised, or zero, (or inf or nan) */ |
|
|
732 | e += 127 - 15; |
|
|
733 | |
|
|
734 | return s | (e << 23) | (m << (23 - 10)); |
|
|
735 | } |
|
|
736 | |
|
|
737 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
|
|
738 | ecb_function_ ecb_const uint16_t |
|
|
739 | ecb_binary32_to_binary16 (uint32_t x) |
|
|
740 | { |
|
|
741 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
|
|
742 | unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
|
|
743 | unsigned int m = x & 0x007fffff; |
|
|
744 | |
|
|
745 | x &= 0x7fffffff; |
|
|
746 | |
|
|
747 | /* if it's within range of binary16 normals, use fast path */ |
|
|
748 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
|
|
749 | { |
|
|
750 | /* mantissa round-to-even */ |
|
|
751 | m += 0x00000fff + ((m >> (23 - 10)) & 1); |
|
|
752 | |
|
|
753 | /* handle overflow */ |
|
|
754 | if (ecb_expect_false (m >= 0x00800000)) |
|
|
755 | { |
|
|
756 | m >>= 1; |
|
|
757 | e += 1; |
|
|
758 | } |
|
|
759 | |
|
|
760 | return s | (e << 10) | (m >> (23 - 10)); |
|
|
761 | } |
|
|
762 | |
|
|
763 | /* handle large numbers and infinity */ |
|
|
764 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
|
|
765 | return s | 0x7c00; |
|
|
766 | |
|
|
767 | /* handle zero, subnormals and small numbers */ |
|
|
768 | if (ecb_expect_true (x < 0x38800000)) |
|
|
769 | { |
|
|
770 | /* zero */ |
|
|
771 | if (ecb_expect_true (!x)) |
|
|
772 | return s; |
|
|
773 | |
|
|
774 | /* handle subnormals */ |
|
|
775 | |
|
|
776 | /* too small, will be zero */ |
|
|
777 | if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
|
|
778 | return s; |
|
|
779 | |
|
|
780 | m |= 0x00800000; /* make implicit bit explicit */ |
|
|
781 | |
|
|
782 | /* very tricky - we need to round to the nearest e (+10) bit value */ |
|
|
783 | { |
|
|
784 | unsigned int bits = 14 - e; |
|
|
785 | unsigned int half = (1 << (bits - 1)) - 1; |
|
|
786 | unsigned int even = (m >> bits) & 1; |
|
|
787 | |
|
|
788 | /* if this overflows, we will end up with a normalised number */ |
|
|
789 | m = (m + half + even) >> bits; |
|
|
790 | } |
|
|
791 | |
|
|
792 | return s | m; |
|
|
793 | } |
|
|
794 | |
|
|
795 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
|
|
796 | m >>= 13; |
|
|
797 | |
|
|
798 | return s | 0x7c00 | m | !m; |
|
|
799 | } |
|
|
800 | |
582 | /*******************************************************************************/ |
801 | /*******************************************************************************/ |
583 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
802 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
584 | |
803 | |
585 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
804 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
586 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
805 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
587 | #if 0 \ |
806 | #if 0 \ |
588 | || __i386 || __i386__ \ |
807 | || __i386 || __i386__ \ |
589 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
808 | || ECB_GCC_AMD64 \ |
590 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
809 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
591 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
592 | || defined __s390__ || defined __s390x__ \ |
810 | || defined __s390__ || defined __s390x__ \ |
593 | || defined __mips__ \ |
811 | || defined __mips__ \ |
594 | || defined __alpha__ \ |
812 | || defined __alpha__ \ |
595 | || defined __hppa__ \ |
813 | || defined __hppa__ \ |
596 | || defined __ia64__ \ |
814 | || defined __ia64__ \ |
597 | || defined __m68k__ \ |
815 | || defined __m68k__ \ |
598 | || defined __m88k__ \ |
816 | || defined __m88k__ \ |
599 | || defined __sh__ \ |
817 | || defined __sh__ \ |
600 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
818 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
|
|
819 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
820 | || defined __aarch64__ |
601 | #define ECB_STDFP 1 |
821 | #define ECB_STDFP 1 |
602 | #include <string.h> /* for memcpy */ |
822 | #include <string.h> /* for memcpy */ |
603 | #else |
823 | #else |
604 | #define ECB_STDFP 0 |
824 | #define ECB_STDFP 0 |
605 | #endif |
825 | #endif |
… | |
… | |
619 | #define ECB_NAN NAN |
839 | #define ECB_NAN NAN |
620 | #else |
840 | #else |
621 | #define ECB_NAN ECB_INFINITY |
841 | #define ECB_NAN ECB_INFINITY |
622 | #endif |
842 | #endif |
623 | |
843 | |
624 | /* converts an ieee half/binary16 to a float */ |
844 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
625 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
845 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
626 | ecb_function_ float |
846 | #define ecb_frexpf(x,e) frexpf ((x), (e)) |
627 | ecb_binary16_to_float (uint16_t x) |
847 | #else |
628 | { |
848 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
629 | int e = (x >> 10) & 0x1f; |
849 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
630 | int m = x & 0x3ff; |
850 | #endif |
631 | float r; |
|
|
632 | |
|
|
633 | if (!e ) r = ldexpf (m , -24); |
|
|
634 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
635 | else if (m ) r = ECB_NAN; |
|
|
636 | else r = ECB_INFINITY; |
|
|
637 | |
|
|
638 | return x & 0x8000 ? -r : r; |
|
|
639 | } |
|
|
640 | |
851 | |
641 | /* convert a float to ieee single/binary32 */ |
852 | /* convert a float to ieee single/binary32 */ |
642 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
853 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
643 | ecb_function_ uint32_t |
854 | ecb_function_ ecb_const uint32_t |
644 | ecb_float_to_binary32 (float x) |
855 | ecb_float_to_binary32 (float x) |
645 | { |
856 | { |
646 | uint32_t r; |
857 | uint32_t r; |
647 | |
858 | |
648 | #if ECB_STDFP |
859 | #if ECB_STDFP |
… | |
… | |
655 | if (x == 0e0f ) return 0x00000000U; |
866 | if (x == 0e0f ) return 0x00000000U; |
656 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
867 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
657 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
868 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
658 | if (x != x ) return 0x7fbfffffU; |
869 | if (x != x ) return 0x7fbfffffU; |
659 | |
870 | |
660 | m = frexpf (x, &e) * 0x1000000U; |
871 | m = ecb_frexpf (x, &e) * 0x1000000U; |
661 | |
872 | |
662 | r = m & 0x80000000U; |
873 | r = m & 0x80000000U; |
663 | |
874 | |
664 | if (r) |
875 | if (r) |
665 | m = -m; |
876 | m = -m; |
… | |
… | |
677 | |
888 | |
678 | return r; |
889 | return r; |
679 | } |
890 | } |
680 | |
891 | |
681 | /* converts an ieee single/binary32 to a float */ |
892 | /* converts an ieee single/binary32 to a float */ |
682 | ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; |
893 | ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); |
683 | ecb_function_ float |
894 | ecb_function_ ecb_const float |
684 | ecb_binary32_to_float (uint32_t x) |
895 | ecb_binary32_to_float (uint32_t x) |
685 | { |
896 | { |
686 | float r; |
897 | float r; |
687 | |
898 | |
688 | #if ECB_STDFP |
899 | #if ECB_STDFP |
… | |
… | |
698 | x |= 0x800000U; |
909 | x |= 0x800000U; |
699 | else |
910 | else |
700 | e = 1; |
911 | e = 1; |
701 | |
912 | |
702 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
913 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
703 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
914 | r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126); |
704 | |
915 | |
705 | r = neg ? -r : r; |
916 | r = neg ? -r : r; |
706 | #endif |
917 | #endif |
707 | |
918 | |
708 | return r; |
919 | return r; |
709 | } |
920 | } |
710 | |
921 | |
711 | /* convert a double to ieee double/binary64 */ |
922 | /* convert a double to ieee double/binary64 */ |
712 | ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; |
923 | ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); |
713 | ecb_function_ uint64_t |
924 | ecb_function_ ecb_const uint64_t |
714 | ecb_double_to_binary64 (double x) |
925 | ecb_double_to_binary64 (double x) |
715 | { |
926 | { |
716 | uint64_t r; |
927 | uint64_t r; |
717 | |
928 | |
718 | #if ECB_STDFP |
929 | #if ECB_STDFP |
… | |
… | |
747 | |
958 | |
748 | return r; |
959 | return r; |
749 | } |
960 | } |
750 | |
961 | |
751 | /* converts an ieee double/binary64 to a double */ |
962 | /* converts an ieee double/binary64 to a double */ |
752 | ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; |
963 | ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); |
753 | ecb_function_ double |
964 | ecb_function_ ecb_const double |
754 | ecb_binary64_to_double (uint64_t x) |
965 | ecb_binary64_to_double (uint64_t x) |
755 | { |
966 | { |
756 | double r; |
967 | double r; |
757 | |
968 | |
758 | #if ECB_STDFP |
969 | #if ECB_STDFP |
… | |
… | |
776 | #endif |
987 | #endif |
777 | |
988 | |
778 | return r; |
989 | return r; |
779 | } |
990 | } |
780 | |
991 | |
781 | #endif |
992 | /* convert a float to ieee half/binary16 */ |
|
|
993 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
|
|
994 | ecb_function_ ecb_const uint16_t |
|
|
995 | ecb_float_to_binary16 (float x) |
|
|
996 | { |
|
|
997 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
|
|
998 | } |
782 | |
999 | |
783 | #endif |
1000 | /* convert an ieee half/binary16 to float */ |
|
|
1001 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
1002 | ecb_function_ ecb_const float |
|
|
1003 | ecb_binary16_to_float (uint16_t x) |
|
|
1004 | { |
|
|
1005 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
|
|
1006 | } |
784 | |
1007 | |
|
|
1008 | #endif |
|
|
1009 | |
|
|
1010 | #endif |
|
|
1011 | |