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 (©) 2009-2014 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 0x00010004 |
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; |
… | |
… | |
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 | /* work around x32 idiocy by defining proper macros */ |
|
|
80 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
|
|
81 | #if _ILP32 |
|
|
82 | #define ECB_AMD64_X32 1 |
|
|
83 | #else |
|
|
84 | #define ECB_AMD64 1 |
|
|
85 | #endif |
|
|
86 | #endif |
|
|
87 | |
68 | /* many compilers define _GNUC_ to some versions but then only implement |
88 | /* many compilers define _GNUC_ to some versions but then only implement |
69 | * what their idiot authors think are the "more important" extensions, |
89 | * what their idiot authors think are the "more important" extensions, |
70 | * causing enormous grief in return for some better fake benchmark numbers. |
90 | * causing enormous grief in return for some better fake benchmark numbers. |
71 | * or so. |
91 | * or so. |
72 | * we try to detect these and simply assume they are not gcc - if they have |
92 | * we try to detect these and simply assume they are not gcc - if they have |
73 | * an issue with that they should have done it right in the first place. |
93 | * an issue with that they should have done it right in the first place. |
74 | */ |
94 | */ |
75 | #ifndef ECB_GCC_VERSION |
|
|
76 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
95 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
77 | #define ECB_GCC_VERSION(major,minor) 0 |
96 | #define ECB_GCC_VERSION(major,minor) 0 |
78 | #else |
97 | #else |
79 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
98 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
80 | #endif |
99 | #endif |
81 | #endif |
|
|
82 | |
100 | |
83 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
101 | #define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor))) |
84 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
102 | |
85 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
103 | #if __clang__ && defined(__has_builtin) |
|
|
104 | #define ECB_CLANG_BUILTIN(x) __has_builtin(x) |
|
|
105 | #else |
|
|
106 | #define ECB_CLANG_BUILTIN(x) 0 |
|
|
107 | #endif |
|
|
108 | |
|
|
109 | #if __clang__ && defined(__has_extension) |
|
|
110 | #define ECB_CLANG_EXTENSION(x) __has_extension(x) |
|
|
111 | #else |
|
|
112 | #define ECB_CLANG_EXTENSION(x) 0 |
|
|
113 | #endif |
|
|
114 | |
86 | #define ECB_CPP (__cplusplus+0) |
115 | #define ECB_CPP (__cplusplus+0) |
87 | #define ECB_CPP11 (__cplusplus >= 201103L) |
116 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
117 | |
|
|
118 | #if ECB_CPP |
|
|
119 | #define ECB_C 0 |
|
|
120 | #define ECB_STDC_VERSION 0 |
|
|
121 | #else |
|
|
122 | #define ECB_C 1 |
|
|
123 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
124 | #endif |
|
|
125 | |
|
|
126 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
127 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
88 | |
128 | |
89 | #if ECB_CPP |
129 | #if ECB_CPP |
90 | #define ECB_EXTERN_C extern "C" |
130 | #define ECB_EXTERN_C extern "C" |
91 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
131 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
92 | #define ECB_EXTERN_C_END } |
132 | #define ECB_EXTERN_C_END } |
… | |
… | |
125 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
165 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
166 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
127 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
167 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
128 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
168 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
170 | #elif __aarch64__ |
|
|
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
130 | #elif __sparc || __sparc__ |
172 | #elif (__sparc || __sparc__) && !__sparcv8 |
131 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
173 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
132 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
174 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
175 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
134 | #elif defined __s390__ || defined __s390x__ |
176 | #elif defined __s390__ || defined __s390x__ |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
177 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
136 | #elif defined __mips__ |
178 | #elif defined __mips__ |
|
|
179 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
|
|
180 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
137 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
181 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
138 | #elif defined __alpha__ |
182 | #elif defined __alpha__ |
139 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
183 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
140 | #elif defined __hppa__ |
184 | #elif defined __hppa__ |
141 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
185 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
186 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
143 | #elif defined __ia64__ |
187 | #elif defined __ia64__ |
144 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
188 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
189 | #elif defined __m68k__ |
|
|
190 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
191 | #elif defined __m88k__ |
|
|
192 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory") |
|
|
193 | #elif defined __sh__ |
|
|
194 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
145 | #endif |
195 | #endif |
146 | #endif |
196 | #endif |
147 | #endif |
197 | #endif |
148 | |
198 | |
149 | #ifndef ECB_MEMORY_FENCE |
199 | #ifndef ECB_MEMORY_FENCE |
150 | #if ECB_GCC_VERSION(4,7) |
200 | #if ECB_GCC_VERSION(4,7) |
151 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
201 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
152 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
202 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
153 | /*#elif defined __clang && __has_feature (cxx_atomic)*/ |
203 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
204 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
205 | |
|
|
206 | #elif ECB_CLANG_EXTENSION(c_atomic) |
154 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
207 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
155 | /*#define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)*/ |
208 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
209 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
210 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
211 | |
156 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
212 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
157 | #define ECB_MEMORY_FENCE __sync_synchronize () |
213 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
214 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
|
|
215 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
216 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
217 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
218 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
219 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
158 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
220 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
159 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
221 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
160 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
222 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
161 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
223 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
162 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
224 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
182 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
244 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
183 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
245 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
184 | /* for most usages, or gcc and clang have a bug */ |
246 | /* for most usages, or gcc and clang have a bug */ |
185 | /* I *currently* lean towards the latter, and inefficiently implement */ |
247 | /* I *currently* lean towards the latter, and inefficiently implement */ |
186 | /* all three of ecb's fences as a seq_cst fence */ |
248 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
249 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
250 | /* for all __atomic_thread_fence's except seq_cst */ |
187 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
251 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
188 | #endif |
252 | #endif |
189 | #endif |
253 | #endif |
190 | |
254 | |
191 | #ifndef ECB_MEMORY_FENCE |
255 | #ifndef ECB_MEMORY_FENCE |
… | |
… | |
241 | #define ECB_STRINGIFY_(a) # a |
305 | #define ECB_STRINGIFY_(a) # a |
242 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
306 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
243 | |
307 | |
244 | #define ecb_function_ ecb_inline |
308 | #define ecb_function_ ecb_inline |
245 | |
309 | |
246 | #if ECB_GCC_VERSION(3,1) |
310 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
247 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
311 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
|
|
312 | #else |
|
|
313 | #define ecb_attribute(attrlist) |
|
|
314 | #endif |
|
|
315 | |
|
|
316 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p) |
248 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
317 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
|
|
318 | #else |
|
|
319 | /* possible C11 impl for integral types |
|
|
320 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
321 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
322 | |
|
|
323 | #define ecb_is_constant(expr) 0 |
|
|
324 | #endif |
|
|
325 | |
|
|
326 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect) |
249 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
327 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
328 | #else |
|
|
329 | #define ecb_expect(expr,value) (expr) |
|
|
330 | #endif |
|
|
331 | |
|
|
332 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch) |
250 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
333 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
251 | #else |
334 | #else |
252 | #define ecb_attribute(attrlist) |
|
|
253 | #define ecb_is_constant(expr) 0 |
|
|
254 | #define ecb_expect(expr,value) (expr) |
|
|
255 | #define ecb_prefetch(addr,rw,locality) |
335 | #define ecb_prefetch(addr,rw,locality) |
256 | #endif |
336 | #endif |
257 | |
337 | |
258 | /* no emulation for ecb_decltype */ |
338 | /* no emulation for ecb_decltype */ |
259 | #if ECB_GCC_VERSION(4,5) |
339 | #if ECB_GCC_VERSION(4,5) |
260 | #define ecb_decltype(x) __decltype(x) |
340 | #define ecb_decltype(x) __decltype (x) |
261 | #elif ECB_GCC_VERSION(3,0) |
341 | #elif ECB_GCC_VERSION(3,0) |
262 | #define ecb_decltype(x) __typeof(x) |
342 | #define ecb_decltype(x) __typeof (x) |
|
|
343 | #elif ECB_CPP11 |
|
|
344 | #define ecb_decltype(x) decltype (x) |
|
|
345 | #endif |
|
|
346 | |
|
|
347 | #if _MSC_VER >= 1300 |
|
|
348 | #define ecb_deprecated __declspec(deprecated) |
|
|
349 | #else |
|
|
350 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
263 | #endif |
351 | #endif |
264 | |
352 | |
265 | #define ecb_noinline ecb_attribute ((__noinline__)) |
353 | #define ecb_noinline ecb_attribute ((__noinline__)) |
266 | #define ecb_unused ecb_attribute ((__unused__)) |
354 | #define ecb_unused ecb_attribute ((__unused__)) |
267 | #define ecb_const ecb_attribute ((__const__)) |
355 | #define ecb_const ecb_attribute ((__const__)) |
268 | #define ecb_pure ecb_attribute ((__pure__)) |
356 | #define ecb_pure ecb_attribute ((__pure__)) |
269 | |
357 | |
|
|
358 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */ |
270 | #if ECB_C11 |
359 | #if ECB_C11 |
271 | #define ecb_noreturn _Noreturn |
360 | #define ecb_noreturn _Noreturn |
272 | #else |
361 | #else |
273 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
362 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
274 | #endif |
363 | #endif |
… | |
… | |
291 | /* for compatibility to the rest of the world */ |
380 | /* for compatibility to the rest of the world */ |
292 | #define ecb_likely(expr) ecb_expect_true (expr) |
381 | #define ecb_likely(expr) ecb_expect_true (expr) |
293 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
382 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
294 | |
383 | |
295 | /* count trailing zero bits and count # of one bits */ |
384 | /* count trailing zero bits and count # of one bits */ |
296 | #if ECB_GCC_VERSION(3,4) |
385 | #if ECB_GCC_VERSION(3,4) \ |
|
|
386 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
|
|
387 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
|
|
388 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
297 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
389 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
298 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
390 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
299 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
391 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
300 | #define ecb_ctz32(x) __builtin_ctz (x) |
392 | #define ecb_ctz32(x) __builtin_ctz (x) |
301 | #define ecb_ctz64(x) __builtin_ctzll (x) |
393 | #define ecb_ctz64(x) __builtin_ctzll (x) |
… | |
… | |
432 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
524 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
433 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
525 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
434 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
526 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
435 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
527 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
436 | |
528 | |
437 | #if ECB_GCC_VERSION(4,3) |
529 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
438 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
530 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
439 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
531 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
440 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
532 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
441 | #else |
533 | #else |
442 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
534 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
… | |
… | |
459 | { |
551 | { |
460 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
552 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
461 | } |
553 | } |
462 | #endif |
554 | #endif |
463 | |
555 | |
464 | #if ECB_GCC_VERSION(4,5) |
556 | #if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable) |
465 | #define ecb_unreachable() __builtin_unreachable () |
557 | #define ecb_unreachable() __builtin_unreachable () |
466 | #else |
558 | #else |
467 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
559 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
468 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
560 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
469 | ecb_inline void ecb_unreachable (void) { } |
561 | ecb_inline void ecb_unreachable (void) { } |
… | |
… | |
475 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
567 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
476 | ecb_inline unsigned char |
568 | ecb_inline unsigned char |
477 | ecb_byteorder_helper (void) |
569 | ecb_byteorder_helper (void) |
478 | { |
570 | { |
479 | /* the union code still generates code under pressure in gcc, */ |
571 | /* the union code still generates code under pressure in gcc, */ |
480 | /* but less than using pointers, and always seem to */ |
572 | /* but less than using pointers, and always seems to */ |
481 | /* successfully return a constant. */ |
573 | /* successfully return a constant. */ |
482 | /* the reason why we have this horrible preprocessor mess */ |
574 | /* the reason why we have this horrible preprocessor mess */ |
483 | /* is to avoid it in all cases, at least on common architectures */ |
575 | /* is to avoid it in all cases, at least on common architectures */ |
484 | /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ |
576 | /* or when using a recent enough gcc version (>= 4.6) */ |
485 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
577 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
486 | return 0x44; |
578 | return 0x44; |
487 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
579 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
488 | return 0x44; |
580 | return 0x44; |
489 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
581 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
490 | retrurn 0x11; |
582 | return 0x11; |
491 | #else |
583 | #else |
492 | union |
584 | union |
493 | { |
585 | { |
494 | uint32_t i; |
586 | uint32_t i; |
495 | uint8_t c; |
587 | uint8_t c; |
… | |
… | |
543 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
635 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
544 | #if 0 \ |
636 | #if 0 \ |
545 | || __i386 || __i386__ \ |
637 | || __i386 || __i386__ \ |
546 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
638 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
547 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
639 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
548 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
549 | || defined __s390__ || defined __s390x__ \ |
640 | || defined __s390__ || defined __s390x__ \ |
550 | || defined __mips__ \ |
641 | || defined __mips__ \ |
551 | || defined __alpha__ \ |
642 | || defined __alpha__ \ |
552 | || defined __hppa__ \ |
643 | || defined __hppa__ \ |
553 | || defined __ia64__ \ |
644 | || defined __ia64__ \ |
|
|
645 | || defined __m68k__ \ |
|
|
646 | || defined __m88k__ \ |
|
|
647 | || defined __sh__ \ |
554 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
648 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
649 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
650 | || defined __aarch64__ |
555 | #define ECB_STDFP 1 |
651 | #define ECB_STDFP 1 |
556 | #include <string.h> /* for memcpy */ |
652 | #include <string.h> /* for memcpy */ |
557 | #else |
653 | #else |
558 | #define ECB_STDFP 0 |
654 | #define ECB_STDFP 0 |
559 | #endif |
655 | #endif |
560 | |
656 | |
561 | #ifndef ECB_NO_LIBM |
657 | #ifndef ECB_NO_LIBM |
|
|
658 | |
|
|
659 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
660 | |
|
|
661 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
662 | #ifdef INFINITY |
|
|
663 | #define ECB_INFINITY INFINITY |
|
|
664 | #else |
|
|
665 | #define ECB_INFINITY HUGE_VAL |
|
|
666 | #endif |
|
|
667 | |
|
|
668 | #ifdef NAN |
|
|
669 | #define ECB_NAN NAN |
|
|
670 | #else |
|
|
671 | #define ECB_NAN ECB_INFINITY |
|
|
672 | #endif |
|
|
673 | |
|
|
674 | /* converts an ieee half/binary16 to a float */ |
|
|
675 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
676 | ecb_function_ float |
|
|
677 | ecb_binary16_to_float (uint16_t x) |
|
|
678 | { |
|
|
679 | int e = (x >> 10) & 0x1f; |
|
|
680 | int m = x & 0x3ff; |
|
|
681 | float r; |
|
|
682 | |
|
|
683 | if (!e ) r = ldexpf (m , -24); |
|
|
684 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
685 | else if (m ) r = ECB_NAN; |
|
|
686 | else r = ECB_INFINITY; |
|
|
687 | |
|
|
688 | return x & 0x8000 ? -r : r; |
|
|
689 | } |
562 | |
690 | |
563 | /* convert a float to ieee single/binary32 */ |
691 | /* convert a float to ieee single/binary32 */ |
564 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
692 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
565 | ecb_function_ uint32_t |
693 | ecb_function_ uint32_t |
566 | ecb_float_to_binary32 (float x) |
694 | ecb_float_to_binary32 (float x) |
… | |
… | |
569 | |
697 | |
570 | #if ECB_STDFP |
698 | #if ECB_STDFP |
571 | memcpy (&r, &x, 4); |
699 | memcpy (&r, &x, 4); |
572 | #else |
700 | #else |
573 | /* slow emulation, works for anything but -0 */ |
701 | /* slow emulation, works for anything but -0 */ |
574 | ECB_EXTERN_C float frexpf (float v, int *e); |
|
|
575 | uint32_t m; |
702 | uint32_t m; |
576 | int e; |
703 | int e; |
577 | |
704 | |
578 | if (x == 0e0f ) return 0; |
705 | if (x == 0e0f ) return 0x00000000U; |
579 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
706 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
580 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
707 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
581 | if (x != x ) return 0x7fbfffffU; |
708 | if (x != x ) return 0x7fbfffffU; |
582 | |
709 | |
583 | m = frexpf (x, &e) * 0x1000000U; |
710 | m = frexpf (x, &e) * 0x1000000U; |
… | |
… | |
585 | r = m & 0x80000000U; |
712 | r = m & 0x80000000U; |
586 | |
713 | |
587 | if (r) |
714 | if (r) |
588 | m = -m; |
715 | m = -m; |
589 | |
716 | |
590 | if (e < -125) |
717 | if (e <= -126) |
591 | { |
718 | { |
592 | m &= 0xffffffU; |
719 | m &= 0xffffffU; |
593 | m >>= (-125 - e); |
720 | m >>= (-125 - e); |
594 | e = -126; |
721 | e = -126; |
595 | } |
722 | } |
… | |
… | |
610 | |
737 | |
611 | #if ECB_STDFP |
738 | #if ECB_STDFP |
612 | memcpy (&r, &x, 4); |
739 | memcpy (&r, &x, 4); |
613 | #else |
740 | #else |
614 | /* emulation, only works for normals and subnormals and +0 */ |
741 | /* emulation, only works for normals and subnormals and +0 */ |
615 | ECB_EXTERN_C float ldexpf (float x, int e); |
|
|
616 | |
|
|
617 | int neg = x >> 31; |
742 | int neg = x >> 31; |
618 | int e = (x >> 23) & 0xffU; |
743 | int e = (x >> 23) & 0xffU; |
619 | |
744 | |
620 | x &= 0x7fffffU; |
745 | x &= 0x7fffffU; |
621 | |
746 | |
… | |
… | |
623 | x |= 0x800000U; |
748 | x |= 0x800000U; |
624 | else |
749 | else |
625 | e = 1; |
750 | e = 1; |
626 | |
751 | |
627 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
752 | /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ |
628 | r = ldexpf (x * (1.f / 0x1000000U), e - 126); |
753 | r = ldexpf (x * (0.5f / 0x800000U), e - 126); |
629 | |
754 | |
630 | r = neg ? -r : r; |
755 | r = neg ? -r : r; |
631 | #endif |
756 | #endif |
632 | |
757 | |
633 | return r; |
758 | return r; |
… | |
… | |
642 | |
767 | |
643 | #if ECB_STDFP |
768 | #if ECB_STDFP |
644 | memcpy (&r, &x, 8); |
769 | memcpy (&r, &x, 8); |
645 | #else |
770 | #else |
646 | /* slow emulation, works for anything but -0 */ |
771 | /* slow emulation, works for anything but -0 */ |
647 | ECB_EXTERN_C double frexp (double v, int *e); |
|
|
648 | uint64_t m; |
772 | uint64_t m; |
649 | int e; |
773 | int e; |
650 | |
774 | |
651 | if (x == 0e0 ) return 0; |
775 | if (x == 0e0 ) return 0x0000000000000000U; |
652 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
776 | if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U; |
653 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
777 | if (x < -1.79769313486231470e+308) return 0xfff0000000000000U; |
654 | if (x != x ) return 0X7ff7ffffffffffffU; |
778 | if (x != x ) return 0X7ff7ffffffffffffU; |
655 | |
779 | |
656 | m = frexp (x, &e) * 0x20000000000000U; |
780 | m = frexp (x, &e) * 0x20000000000000U; |
… | |
… | |
658 | r = m & 0x8000000000000000;; |
782 | r = m & 0x8000000000000000;; |
659 | |
783 | |
660 | if (r) |
784 | if (r) |
661 | m = -m; |
785 | m = -m; |
662 | |
786 | |
663 | if (e < -1021) |
787 | if (e <= -1022) |
664 | { |
788 | { |
665 | m &= 0x1fffffffffffffU; |
789 | m &= 0x1fffffffffffffU; |
666 | m >>= (-1021 - e); |
790 | m >>= (-1021 - e); |
667 | e = -1022; |
791 | e = -1022; |
668 | } |
792 | } |
… | |
… | |
683 | |
807 | |
684 | #if ECB_STDFP |
808 | #if ECB_STDFP |
685 | memcpy (&r, &x, 8); |
809 | memcpy (&r, &x, 8); |
686 | #else |
810 | #else |
687 | /* emulation, only works for normals and subnormals and +0 */ |
811 | /* emulation, only works for normals and subnormals and +0 */ |
688 | ECB_EXTERN_C double ldexp (double x, int e); |
|
|
689 | |
|
|
690 | int neg = x >> 63; |
812 | int neg = x >> 63; |
691 | int e = (x >> 52) & 0x7ffU; |
813 | int e = (x >> 52) & 0x7ffU; |
692 | |
814 | |
693 | x &= 0xfffffffffffffU; |
815 | x &= 0xfffffffffffffU; |
694 | |
816 | |
… | |
… | |
696 | x |= 0x10000000000000U; |
818 | x |= 0x10000000000000U; |
697 | else |
819 | else |
698 | e = 1; |
820 | e = 1; |
699 | |
821 | |
700 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
822 | /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */ |
701 | r = ldexp (x * (1. / 0x20000000000000U), e - 1022); |
823 | r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022); |
702 | |
824 | |
703 | r = neg ? -r : r; |
825 | r = neg ? -r : r; |
704 | #endif |
826 | #endif |
705 | |
827 | |
706 | return r; |
828 | return r; |