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 (©) 2009-2014 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: |
… | |
… | |
64 | #define ECB_PTRSIZE 4 |
64 | #define ECB_PTRSIZE 4 |
65 | #endif |
65 | #endif |
66 | #endif |
66 | #endif |
67 | |
67 | |
68 | /* work around x32 idiocy by defining proper macros */ |
68 | /* work around x32 idiocy by defining proper macros */ |
69 | #if __x86_64 || _M_AMD64 |
69 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
70 | #if _ILP32 |
70 | #if _ILP32 |
71 | #define ECB_AMD64_X32 1 |
71 | #define ECB_AMD64_X32 1 |
72 | #else |
72 | #else |
73 | #define ECB_AMD64 1 |
73 | #define ECB_AMD64 1 |
74 | #endif |
74 | #endif |
… | |
… | |
87 | #else |
87 | #else |
88 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
88 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
89 | #endif |
89 | #endif |
90 | #endif |
90 | #endif |
91 | |
91 | |
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) |
92 | #define ECB_CPP (__cplusplus+0) |
96 | #define ECB_CPP11 (__cplusplus >= 201103L) |
93 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
94 | |
|
|
95 | #if ECB_CPP |
|
|
96 | #define ECB_C 0 |
|
|
97 | #define ECB_STDC_VERSION 0 |
|
|
98 | #else |
|
|
99 | #define ECB_C 1 |
|
|
100 | #define ECB_STDC_VERSION __STDC_VERSION__ |
|
|
101 | #endif |
|
|
102 | |
|
|
103 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
|
|
104 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
97 | |
105 | |
98 | #if ECB_CPP |
106 | #if ECB_CPP |
99 | #define ECB_EXTERN_C extern "C" |
107 | #define ECB_EXTERN_C extern "C" |
100 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
108 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
101 | #define ECB_EXTERN_C_END } |
109 | #define ECB_EXTERN_C_END } |
… | |
… | |
134 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
142 | || 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") |
143 | #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__ \ |
144 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
137 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
145 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
138 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
146 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
147 | #elif __aarch64__ |
|
|
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
139 | #elif (__sparc || __sparc__) && !__sparcv8 |
149 | #elif (__sparc || __sparc__) && !__sparcv8 |
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
141 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
151 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
142 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
152 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
143 | #elif defined __s390__ || defined __s390x__ |
153 | #elif defined __s390__ || defined __s390x__ |
… | |
… | |
165 | |
175 | |
166 | #ifndef ECB_MEMORY_FENCE |
176 | #ifndef ECB_MEMORY_FENCE |
167 | #if ECB_GCC_VERSION(4,7) |
177 | #if ECB_GCC_VERSION(4,7) |
168 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
178 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
169 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
179 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
180 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
181 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
170 | |
182 | |
171 | /* The __has_feature syntax from clang is so misdesigned that we cannot use it |
183 | /* 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* |
184 | * 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 |
185 | * define our own ecb_clang_has_feature, but I just can't be bothered to work |
174 | * around this shit time and again. |
186 | * around this shit time and again. |
175 | * #elif defined __clang && __has_feature (cxx_atomic) |
187 | * #elif defined __clang && __has_feature (cxx_atomic) |
176 | * // see comment below (stdatomic.h) about the C11 memory model. |
188 | * // see comment below (stdatomic.h) about the C11 memory model. |
177 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
189 | * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
190 | * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
191 | * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
178 | */ |
192 | */ |
179 | |
193 | |
180 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
194 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
181 | #define ECB_MEMORY_FENCE __sync_synchronize () |
195 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
196 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
|
|
197 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
|
|
198 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
199 | #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier() |
|
|
200 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
201 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier() |
182 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
202 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
183 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
203 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
184 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
204 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
185 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
205 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
186 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
206 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
206 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
226 | /* 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 */ |
227 | /* 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 */ |
228 | /* for most usages, or gcc and clang have a bug */ |
209 | /* I *currently* lean towards the latter, and inefficiently implement */ |
229 | /* I *currently* lean towards the latter, and inefficiently implement */ |
210 | /* all three of ecb's fences as a seq_cst fence */ |
230 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
231 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
232 | /* for all __atomic_thread_fence's except seq_cst */ |
211 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
233 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
212 | #endif |
234 | #endif |
213 | #endif |
235 | #endif |
214 | |
236 | |
215 | #ifndef ECB_MEMORY_FENCE |
237 | #ifndef ECB_MEMORY_FENCE |
… | |
… | |
272 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
294 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
273 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
295 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
274 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
296 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
275 | #else |
297 | #else |
276 | #define ecb_attribute(attrlist) |
298 | #define ecb_attribute(attrlist) |
|
|
299 | |
|
|
300 | /* possible C11 impl for integral types |
|
|
301 | typedef struct ecb_is_constant_struct ecb_is_constant_struct; |
|
|
302 | #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */ |
|
|
303 | |
277 | #define ecb_is_constant(expr) 0 |
304 | #define ecb_is_constant(expr) 0 |
278 | #define ecb_expect(expr,value) (expr) |
305 | #define ecb_expect(expr,value) (expr) |
279 | #define ecb_prefetch(addr,rw,locality) |
306 | #define ecb_prefetch(addr,rw,locality) |
280 | #endif |
307 | #endif |
281 | |
308 | |
… | |
… | |
567 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
594 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
568 | #if 0 \ |
595 | #if 0 \ |
569 | || __i386 || __i386__ \ |
596 | || __i386 || __i386__ \ |
570 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
597 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
571 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
598 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
572 | || defined __arm__ && defined __ARM_EABI__ \ |
|
|
573 | || defined __s390__ || defined __s390x__ \ |
599 | || defined __s390__ || defined __s390x__ \ |
574 | || defined __mips__ \ |
600 | || defined __mips__ \ |
575 | || defined __alpha__ \ |
601 | || defined __alpha__ \ |
576 | || defined __hppa__ \ |
602 | || defined __hppa__ \ |
577 | || defined __ia64__ \ |
603 | || defined __ia64__ \ |
578 | || defined __m68k__ \ |
604 | || defined __m68k__ \ |
579 | || defined __m88k__ \ |
605 | || defined __m88k__ \ |
580 | || defined __sh__ \ |
606 | || defined __sh__ \ |
581 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 |
607 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
|
|
608 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
|
|
609 | || defined __aarch64__ |
582 | #define ECB_STDFP 1 |
610 | #define ECB_STDFP 1 |
583 | #include <string.h> /* for memcpy */ |
611 | #include <string.h> /* for memcpy */ |
584 | #else |
612 | #else |
585 | #define ECB_STDFP 0 |
613 | #define ECB_STDFP 0 |
586 | #endif |
614 | #endif |