next up previous contents
: OX Local Data : CMOexpressions for numbers and : CPU dependent double   ܼ


This subsection describes our format for bigfloat in terms of the int32. This format is identical to the internal format of mpfr on 32 bit CPU's.

Ref: OpenXM/src/mpfr/bfsize/bfsize.c

#define     CMO_BIGFLOAT32          52

The bigfloat32 is an array of int32 numbers of the following format.

, int32
, int32
, int32 a[1], ..., int32 a[k]

p=prec is the precision, s=sign is the sign(1 means positive, -1 (expressed by two's complement) is nevative), E=exp is the exponent, and the data above expresses the number

\begin{displaymath}s (a[k]/B + a[k-1]/B^2 + ... + a[1]/B^k) 2^E. \end{displaymath}

Here, $B=2^{32}$, $k = \lceil p/32 \rceil$.

Nobuki Takayama Heisei 28.8.27.