ref: 8decdc4ae4ab1ec629e1d2db83f97d5d2b571743
dir: /llt/dtypes.h/
#ifndef __DTYPES_H_ #define __DTYPES_H_ /* This file defines sane integer types for our target platforms. This library only runs on machines with the following characteristics: - supports integer word sizes of 8, 16, 32, and 64 bits - uses unsigned and signed 2's complement representations - all pointer types are the same size - there is an integer type with the same size as a pointer Some features require: - IEEE 754 single- and double-precision floating point We assume the LP64 convention for 64-bit platforms. */ #include "config.h" typedef int bool_t; /* unfortunately we can't make this an enum, since false is an invalid enum label in C++ (since it's a keyword) */ #define false (0) #define true (1) #if defined(__INTEL_COMPILER) && defined(WIN32) # define STATIC_INLINE static # define INLINE # ifdef BITS64 typedef unsigned long size_t; # else typedef unsigned int size_t; # endif #else # define STATIC_INLINE static inline # define INLINE inline #endif typedef unsigned char byte_t; /* 1 byte */ #if defined(WIN32) typedef short int16_t; typedef int int32_t; typedef long long int64_t; typedef unsigned char u_int8_t; typedef unsigned short u_int16_t; typedef unsigned int u_int32_t; #ifdef BITS64 typedef unsigned long u_int64_t; #else typedef unsigned long long u_int64_t; #endif #ifdef __INTEL_COMPILER typedef signed char int8_t; typedef short int16_t; typedef int int32_t; #endif #else #include <sys/types.h> #endif #ifdef BITS64 #define TOP_BIT 0x8000000000000000 #define NBITS 64 typedef unsigned long uint_t; // preferred int type on platform typedef long int_t; typedef int64_t offset_t; typedef u_int64_t index_t; typedef int64_t ptrint_t; // pointer-size int typedef u_int64_t u_ptrint_t #else #define TOP_BIT 0x80000000 #define NBITS 32 typedef unsigned long uint_t; typedef long int_t; typedef int32_t offset_t; typedef u_int32_t index_t; typedef int32_t ptrint_t; typedef u_int32_t u_ptrint_t; #endif typedef u_int8_t uint8_t; typedef u_int16_t uint16_t; typedef u_int32_t uint32_t; typedef u_int64_t uint64_t; typedef u_ptrint_t uptrint_t; #define ALIGN(x, sz) (((x) + (sz-1)) & (-sz)) // branch prediction annotations #ifdef __GNUC__ #define __unlikely(x) __builtin_expect(!!(x), 0) #define __likely(x) __builtin_expect(!!(x), 1) #else #define __unlikely(x) (x) #define __likely(x) (x) #endif #define DBL_MAXINT 9007199254740992LL #define FLT_MAXINT 16777216 #define U64_MAX 18446744073709551615ULL #define S64_MAX 9223372036854775807LL #define S64_MIN (-S64_MAX - 1LL) #define BIT63 0x8000000000000000LL #define BIT31 0x80000000 #define DBL_EPSILON 2.2204460492503131e-16 #define FLT_EPSILON 1.1920928e-7 #define DBL_MAX 1.7976931348623157e+308 #define DBL_MIN 2.2250738585072014e-308 #define FLT_MAX 3.402823466e+38 #define FLT_MIN 1.175494351e-38 #define LOG2_10 3.3219280948873626 #define rel_zero(a, b) (fabs((a)/(b)) < DBL_EPSILON) #define sign_bit(r) ((*(int64_t*)&(r)) & BIT63) #define LABS(n) (((n)^((n)>>(NBITS-1))) - ((n)>>(NBITS-1))) #define NBABS(n,nb) (((n)^((n)>>((nb)-1))) - ((n)>>((nb)-1))) #define DFINITE(d) (((*(int64_t*)&(d))&0x7ff0000000000000LL)!=0x7ff0000000000000LL) #define DNAN(d) ((d)!=(d)) extern double D_PNAN; extern double D_NNAN; extern double D_PINF; extern double D_NINF; extern float F_PNAN; extern float F_NNAN; extern float F_PINF; extern float F_NINF; typedef enum { T_INT8, T_UINT8, T_INT16, T_UINT16, T_INT32, T_UINT32, T_INT64, T_UINT64, T_FLOAT, T_DOUBLE } numerictype_t; #define N_NUMTYPES ((int)T_DOUBLE+1) #ifdef BITS64 # define T_LONG T_INT64 # define T_ULONG T_UINT64 #else # define T_LONG T_INT32 # define T_ULONG T_UINT32 #endif #endif