… | |
… | |
54 | #else |
54 | #else |
55 | #define ECB_PTRSIZE 4 |
55 | #define ECB_PTRSIZE 4 |
56 | typedef uint32_t uintptr_t; |
56 | typedef uint32_t uintptr_t; |
57 | typedef int32_t intptr_t; |
57 | typedef int32_t intptr_t; |
58 | #endif |
58 | #endif |
59 | typedef intptr_t ptrdiff_t; |
|
|
60 | #else |
59 | #else |
61 | #include <inttypes.h> |
60 | #include <inttypes.h> |
62 | #if UINTMAX_MAX > 0xffffffffU |
61 | #if UINTMAX_MAX > 0xffffffffU |
63 | #define ECB_PTRSIZE 8 |
62 | #define ECB_PTRSIZE 8 |
64 | #else |
63 | #else |
… | |
… | |
83 | |
82 | |
84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
83 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
84 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
85 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
87 | #define ECB_CPP (__cplusplus+0) |
86 | #define ECB_CPP (__cplusplus+0) |
88 | #define ECB_CPP98 (__cplusplus >= 199711L) |
|
|
89 | #define ECB_CPP11 (__cplusplus >= 201103L) |
87 | #define ECB_CPP11 (__cplusplus >= 201103L) |
90 | |
88 | |
91 | /*****************************************************************************/ |
89 | /*****************************************************************************/ |
92 | |
90 | |
93 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
91 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
94 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
92 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
95 | |
93 | |
96 | #if ECB_NO_THREADS |
94 | #if ECB_NO_THREADS |
97 | # define ECB_NO_SMP 1 |
95 | #define ECB_NO_SMP 1 |
98 | #endif |
96 | #endif |
99 | |
97 | |
100 | #if ECB_NO_THREADS || ECB_NO_SMP |
98 | #if ECB_NO_SMP |
101 | #define ECB_MEMORY_FENCE do { } while (0) |
99 | #define ECB_MEMORY_FENCE do { } while (0) |
102 | #endif |
|
|
103 | |
|
|
104 | #ifndef ECB_MEMORY_FENCE |
|
|
105 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
106 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
107 | /* not just C11 atomics and atomic accesses */ |
|
|
108 | #include <stdatomic.h> |
|
|
109 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_acq_rel) |
|
|
110 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
111 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
|
|
112 | #endif |
|
|
113 | #endif |
|
|
114 | |
|
|
115 | #ifndef ECB_MEMORY_FENCE_RELEASE |
|
|
116 | #if ECB_GCC_VERSION(4,7) |
|
|
117 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_ACQ_REL) |
|
|
118 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
|
|
119 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
120 | #endif |
|
|
121 | #endif |
100 | #endif |
122 | |
101 | |
123 | #ifndef ECB_MEMORY_FENCE |
102 | #ifndef ECB_MEMORY_FENCE |
124 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
103 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
125 | #if __i386 || __i386__ |
104 | #if __i386 || __i386__ |
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
105 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
127 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
106 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
128 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
107 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
129 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
108 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
130 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
109 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
131 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
110 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
132 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
111 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
133 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
112 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
134 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
113 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
135 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
114 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
136 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
115 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
137 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
116 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
138 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
117 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
139 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
118 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
140 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
119 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
141 | #elif __sparc || __sparc__ |
120 | #elif __sparc || __sparc__ |
142 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
121 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
143 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
122 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
144 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
123 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
145 | #elif defined __s390__ || defined __s390x__ |
124 | #elif defined __s390__ || defined __s390x__ |
146 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
125 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
147 | #elif defined __mips__ |
126 | #elif defined __mips__ |
148 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
127 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
149 | #elif defined __alpha__ |
128 | #elif defined __alpha__ |
150 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
129 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
130 | #elif defined __hppa__ |
|
|
131 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
132 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
133 | #elif defined __ia64__ |
|
|
134 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
151 | #endif |
135 | #endif |
152 | #endif |
136 | #endif |
153 | #endif |
137 | #endif |
154 | |
138 | |
155 | #ifndef ECB_MEMORY_FENCE |
139 | #ifndef ECB_MEMORY_FENCE |
|
|
140 | #if ECB_GCC_VERSION(4,7) |
|
|
141 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
142 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
143 | #elif defined __clang && __has_feature (cxx_atomic) |
|
|
144 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
145 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
156 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
146 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
157 | #define ECB_MEMORY_FENCE __sync_synchronize () |
147 | #define ECB_MEMORY_FENCE __sync_synchronize () |
158 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
|
|
159 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
|
|
160 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
148 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
161 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
149 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
162 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
150 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
163 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
151 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
164 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
152 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
… | |
… | |
174 | #define ECB_MEMORY_FENCE __sync () |
162 | #define ECB_MEMORY_FENCE __sync () |
175 | #endif |
163 | #endif |
176 | #endif |
164 | #endif |
177 | |
165 | |
178 | #ifndef ECB_MEMORY_FENCE |
166 | #ifndef ECB_MEMORY_FENCE |
|
|
167 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
168 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
169 | /* not just C11 atomics and atomic accesses */ |
|
|
170 | #include <stdatomic.h> |
|
|
171 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
172 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
173 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
174 | /* for most usages, or gcc and clang have a bug */ |
|
|
175 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
176 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
177 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
178 | #endif |
|
|
179 | #endif |
|
|
180 | |
|
|
181 | #ifndef ECB_MEMORY_FENCE |
179 | #if !ECB_AVOID_PTHREADS |
182 | #if !ECB_AVOID_PTHREADS |
180 | /* |
183 | /* |
181 | * if you get undefined symbol references to pthread_mutex_lock, |
184 | * if you get undefined symbol references to pthread_mutex_lock, |
182 | * or failure to find pthread.h, then you should implement |
185 | * or failure to find pthread.h, then you should implement |
183 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
186 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
… | |
… | |
455 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
458 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
456 | ecb_inline void ecb_unreachable (void) { } |
459 | ecb_inline void ecb_unreachable (void) { } |
457 | #endif |
460 | #endif |
458 | |
461 | |
459 | /* try to tell the compiler that some condition is definitely true */ |
462 | /* try to tell the compiler that some condition is definitely true */ |
460 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
463 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
461 | |
464 | |
462 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
465 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
463 | ecb_inline unsigned char |
466 | ecb_inline unsigned char |
464 | ecb_byteorder_helper (void) |
467 | ecb_byteorder_helper (void) |
465 | { |
468 | { |
466 | const uint32_t u = 0x11223344; |
469 | /* the union code still generates code under pressure in gcc, */ |
467 | return *(unsigned char *)&u; |
470 | /* but less than using pointers, and always seem to */ |
|
|
471 | /* successfully return a constant. */ |
|
|
472 | /* the reason why we have this horrible preprocessor mess */ |
|
|
473 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
474 | /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ |
|
|
475 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
476 | return 0x44; |
|
|
477 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
478 | return 0x44; |
|
|
479 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
480 | retrurn 0x11; |
|
|
481 | #else |
|
|
482 | union |
|
|
483 | { |
|
|
484 | uint32_t i; |
|
|
485 | uint8_t c; |
|
|
486 | } u = { 0x11223344 }; |
|
|
487 | return u.c; |
|
|
488 | #endif |
468 | } |
489 | } |
469 | |
490 | |
470 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
491 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
471 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
492 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
472 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
493 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |