/* values used by math functions -- IEEE 754 float version */ #include "wctype.h" #include "xmath.h" _STD_BEGIN /* macros */ #define NBITS (16 + _FOFF) #if _D0 #define INIT(w0) {0, w0} #define INIT2(w0, w1) {w1, w0} #else #define INIT(w0) {w0, 0} #define INIT2(w0, w1) {w0, w1} #endif /* static data */ _CRTIMP2 const _Dconst _FDenorm = {INIT2(0, 1)}; _CRTIMP2 const _Dconst _FEps = {INIT((_FBIAS - NBITS - 1) << _FOFF)}; _CRTIMP2 const _Dconst _FInf = {INIT(_FMAX << _FOFF)}; _CRTIMP2 const _Dconst _FNan = {INIT(_FSIGN | (_FMAX << _FOFF) | (1 << _FOFF - 1))}; _CRTIMP2 const _Dconst _FSnan = {INIT(_FSIGN | (_FMAX << _FOFF))}; _CRTIMP2 const _Dconst _FRteps = {INIT((_FBIAS - NBITS / 2) << _FOFF)}; _CRTIMP2 const float _FXbig = (NBITS + 1) * 347L / 1000; #if defined(__CENTERLINE__) #define _DYNAMIC_INIT_CONST(x) \ (x._F = *(float *)(void *)(x._W)) _DYNAMIC_INIT_CONST(_FEps); _DYNAMIC_INIT_CONST(_FInf); _DYNAMIC_INIT_CONST(_FNan); _DYNAMIC_INIT_CONST(_FRteps); #endif _STD_END /* * Copyright (c) 1994 by P.J. Plauger. ALL RIGHTS RESERVED. * Consult your license regarding permissions and restrictions. */ /* 941029 pjp: added _STD machinery 950222 pjp: added signaling NaN, denorm minimum for C++ 950405 pjp: corrected _FSnan punctuation 950505 pjp: corrected _FDenorm spelling */