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 | |
… | |
… | |
64 | #define ECB_PTRSIZE 4 |
75 | #define ECB_PTRSIZE 4 |
65 | #endif |
76 | #endif |
66 | #endif |
77 | #endif |
67 | |
78 | |
68 | /* work around x32 idiocy by defining proper macros */ |
79 | /* work around x32 idiocy by defining proper macros */ |
69 | #if __x86_64 || _M_AMD64 |
80 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
70 | #if __ILP32 |
81 | #if _ILP32 |
71 | #define ECB_AMD64_X32 1 |
82 | #define ECB_AMD64_X32 1 |
72 | #else |
83 | #else |
73 | #define ECB_AMD64 1 |
84 | #define ECB_AMD64 1 |
74 | #endif |
85 | #endif |
75 | #endif |
86 | #endif |
… | |
… | |
87 | #else |
98 | #else |
88 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
99 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
89 | #endif |
100 | #endif |
90 | #endif |
101 | #endif |
91 | |
102 | |
92 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
|
|
93 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
94 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
|
|
95 | #define ECB_CPP (__cplusplus+0) |
103 | #define ECB_CPP (__cplusplus+0) |
96 | #define ECB_CPP11 (__cplusplus >= 201103L) |
104 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
105 | |
|
|
106 | #if ECB_CPP |
|
|
107 | #define ECB_C 0 |
|
|
108 | #define ECB_STDC_VERSION 0 |
|
|
109 | #else |
|
|
110 | #define ECB_C 1 |
|
|
111 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
112 | #endif |
|
|
113 | |
|
|
114 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
115 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
97 | |
116 | |
98 | #if ECB_CPP |
117 | #if ECB_CPP |
99 | #define ECB_EXTERN_C extern "C" |
118 | #define ECB_EXTERN_C extern "C" |
100 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
119 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
101 | #define ECB_EXTERN_C_END } |
120 | #define ECB_EXTERN_C_END } |
… | |
… | |
134 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
153 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
154 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
136 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
155 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
137 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
156 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
138 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
157 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
158 | #elif __aarch64__ |
|
|
159 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
139 | #elif __sparc || __sparc__ |
160 | #elif (__sparc || __sparc__) && !__sparcv8 |
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
161 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
141 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
162 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
163 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
143 | #elif defined __s390__ || defined __s390x__ |
164 | #elif defined __s390__ || defined __s390x__ |
144 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
165 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
145 | #elif defined __mips__ |
166 | #elif defined __mips__ |
146 | /* GNU/Linux emulates sync on mips1 architectures, so we force it's use */ |
167 | /* GNU/Linux emulates sync on mips1 architectures, so we force its use */ |
147 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
168 | /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */ |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory") |
149 | #elif defined __alpha__ |
170 | #elif defined __alpha__ |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
151 | #elif defined __hppa__ |
172 | #elif defined __hppa__ |
… | |
… | |
165 | |
186 | |
166 | #ifndef ECB_MEMORY_FENCE |
187 | #ifndef ECB_MEMORY_FENCE |
167 | #if ECB_GCC_VERSION(4,7) |
188 | #if ECB_GCC_VERSION(4,7) |
168 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
189 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
169 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
190 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
191 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
192 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
170 | |
193 | |
171 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
194 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
172 | * without risking compile time errors with other compilers. We *could* |
195 | * without risking compile time errors with other compilers. We *could* |
173 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
196 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
174 | * around this shit time and again. |
197 | * around this shit time and again. |
175 | * #elif defined __clang && __has_feature (cxx_atomic) |
198 | * #elif defined __clang && __has_feature (cxx_atomic) |
176 | * // see comment below (stdatomic.h) about the C11 memory model. |
199 | * // see comment below (stdatomic.h) about the C11 memory model. |
177 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
200 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
201 | * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
202 | * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
178 | */ |
203 | */ |
179 | |
204 | |
180 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
205 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
181 | #define ECB_MEMORY_FENCE __sync_synchronize () |
206 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
207 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
|
|
208 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
209 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
210 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
211 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
212 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
182 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
213 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
183 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
214 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
184 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
215 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
185 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
216 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
186 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
217 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
206 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
237 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
207 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
238 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
208 | /* for most usages, or gcc and clang have a bug */ |
239 | /* for most usages, or gcc and clang have a bug */ |
209 | /* I *currently* lean towards the latter, and inefficiently implement */ |
240 | /* I *currently* lean towards the latter, and inefficiently implement */ |
210 | /* all three of ecb's fences as a seq_cst fence */ |
241 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
242 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
243 | /* for all __atomic_thread_fence's except seq_cst */ |
211 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
244 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
212 | #endif |
245 | #endif |
213 | #endif |
246 | #endif |
214 | |
247 | |
215 | #ifndef ECB_MEMORY_FENCE |
248 | #ifndef ECB_MEMORY_FENCE |
… | |
… | |
272 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
305 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
273 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
306 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
274 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
307 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
275 | #else |
308 | #else |
276 | #define ecb_attribute(attrlist) |
309 | #define ecb_attribute(attrlist) |
|
|
310 | |
|
|
311 | /* possible C11 impl for integral types |
|
|
312 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
313 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
314 | |
277 | #define ecb_is_constant(expr) 0 |
315 | #define ecb_is_constant(expr) 0 |
278 | #define ecb_expect(expr,value) (expr) |
316 | #define ecb_expect(expr,value) (expr) |
279 | #define ecb_prefetch(addr,rw,locality) |
317 | #define ecb_prefetch(addr,rw,locality) |
280 | #endif |
318 | #endif |
281 | |
319 | |
… | |
… | |
567 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
605 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
568 | #if 0 \ |
606 | #if 0 \ |
569 | || __i386 || __i386__ \ |
607 | || __i386 || __i386__ \ |
570 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
608 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
571 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
609 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
572 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
573 | || defined __s390__ || defined __s390x__ \ |
610 | || defined __s390__ || defined __s390x__ \ |
574 | || defined __mips__ \ |
611 | || defined __mips__ \ |
575 | || defined __alpha__ \ |
612 | || defined __alpha__ \ |
576 | || defined __hppa__ \ |
613 | || defined __hppa__ \ |
577 | || defined __ia64__ \ |
614 | || defined __ia64__ \ |
578 | || defined __m68k__ \ |
615 | || defined __m68k__ \ |
579 | || defined __m88k__ \ |
616 | || defined __m88k__ \ |
580 | || defined __sh__ \ |
617 | || defined __sh__ \ |
581 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
618 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
619 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
620 | || defined __aarch64__ |
582 | #define ECB_STDFP 1 |
621 | #define ECB_STDFP 1 |
583 | #include <string.h> /* for memcpy */ |
622 | #include <string.h> /* for memcpy */ |
584 | #else |
623 | #else |
585 | #define ECB_STDFP 0 |
624 | #define ECB_STDFP 0 |
586 | #include <math.h> /* for frexp*, ldexp* */ |
|
|
587 | #endif |
625 | #endif |
588 | |
626 | |
589 | #ifndef ECB_NO_LIBM |
627 | #ifndef ECB_NO_LIBM |
|
|
628 | |
|
|
629 | #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */ |
|
|
630 | |
|
|
631 | /* only the oldest of old doesn't have this one. solaris. */ |
|
|
632 | #ifdef INFINITY |
|
|
633 | #define ECB_INFINITY INFINITY |
|
|
634 | #else |
|
|
635 | #define ECB_INFINITY HUGE_VAL |
|
|
636 | #endif |
|
|
637 | |
|
|
638 | #ifdef NAN |
|
|
639 | #define ECB_NAN NAN |
|
|
640 | #else |
|
|
641 | #define ECB_NAN ECB_INFINITY |
|
|
642 | #endif |
|
|
643 | |
|
|
644 | /* converts an ieee half/binary16 to a float */ |
|
|
645 | ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; |
|
|
646 | ecb_function_ float |
|
|
647 | ecb_binary16_to_float (uint16_t x) |
|
|
648 | { |
|
|
649 | int e = (x >> 10) & 0x1f; |
|
|
650 | int m = x & 0x3ff; |
|
|
651 | float r; |
|
|
652 | |
|
|
653 | if (!e ) r = ldexpf (m , -24); |
|
|
654 | else if (e != 31) r = ldexpf (m + 0x400, e - 25); |
|
|
655 | else if (m ) r = ECB_NAN; |
|
|
656 | else r = ECB_INFINITY; |
|
|
657 | |
|
|
658 | return x & 0x8000 ? -r : r; |
|
|
659 | } |
590 | |
660 | |
591 | /* convert a float to ieee single/binary32 */ |
661 | /* convert a float to ieee single/binary32 */ |
592 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
662 | ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; |
593 | ecb_function_ uint32_t |
663 | ecb_function_ uint32_t |
594 | ecb_float_to_binary32 (float x) |
664 | ecb_float_to_binary32 (float x) |