/** * Atomic add */template <typename ValueT>inline ValueT add_and_fetch(volatile ValueT *ptr, ValueT val);template <>inline uint64_t add_and_fetch(volatile uint64_t *ptr, uint64_t val){#if LEVELDB_IS_SOLARIS return atomic_add_64_nv(ptr, val);#else return __sync_add_and_fetch(ptr, val);#endif}template <>inline uint32_t add_and_fetch(volatile uint32_t *ptr, uint32_t val){#if LEVELDB_IS_SOLARIS return atomic_add_32_nv(ptr, val);#else return __sync_add_and_fetch(ptr, val);#endif}#if defined(__APPLE__) || defined(__OpenBSD__) || (defined(__s390__) && !defined(__s390x__))template <>inline size_t add_and_fetch(volatile size_t *ptr, size_t val){ return __sync_add_and_fetch(ptr, val);}#endif});
解决方案
The C standard does not exactly define sizes of short
, int
, long
, long long
and their unsigned
versions. Only minimum sizes are enforced. For sake of example consider x86_64 architecture. long
on Linux is 64-bit whereas on 64-bit Windows it is 32-bit. Common approach to make code more portable is to use length-specific types like uint16_t
or int32_t
defined by C99's stdint.h
header file. Three kinds of integer types are defined there:
- with exactly specified size:
uint8_t
,uint16_t
,int32_t
, etc. - smallest type with at least specified size:
uint_least8_t
,uint_least16_t
,int_least32_t
, etc. - most efficient type with at least specified size:
uint_fast8_t
,uint_fast16_t
,int_fast32_t
, etc.
一句话就是为了代码的可移值性。