4.10 Functions on floating-point numbers

Recall that a floating-point number consists of a sign s, an exponent e and a mantissa m. The value of the number is (-1)^s * 2^e * m.

Each of the classes cl_F, cl_SF, cl_FF, cl_DF, cl_LF defines the following operations.

type scale_float (const type& x, sintC delta)

type scale_float (const type& x, const cl_I& delta)

Returns x*2^delta. This is more efficient than an explicit multiplication because it copies x and modifies the exponent.

The following functions provide an abstract interface to the underlying representation of floating-point numbers.

sintE float_exponent (const type& x)

Returns the exponent e of x. For x = 0.0, this is 0. For x non-zero, this is the unique integer with 2^(e-1) <= abs(x) < 2^e.

sintL float_radix (const type& x)

Returns the base of the floating-point representation. This is always 2.

type float_sign (const type& x)

Returns the sign s of x as a float. The value is 1 for x >= 0, -1 for x < 0.

uintC float_digits (const type& x)

Returns the number of mantissa bits in the floating-point representation of x, including the hidden bit. The value only depends on the type of x, not on its value.

uintC float_precision (const type& x)

Returns the number of significant mantissa bits in the floating-point representation of x. Since denormalized numbers are not supported, this is the same as float_digits(x) if x is non-zero, and 0 if x = 0.

The complete internal representation of a float is encoded in the type decoded_float (or decoded_sfloat, decoded_ffloat, decoded_dfloat, decoded_lfloat, respectively), defined by

struct decoded_typefloat {
        type mantissa; cl_I exponent; type sign;
};

and returned by the function

decoded_typefloat decode_float (const type& x)

For x non-zero, this returns (-1)^s, e, m with x = (-1)^s * 2^e * m and 0.5 <= m < 1.0. For x = 0, it returns (-1)^s=1, e=0, m=0. e is the same as returned by the function float_exponent.

A complete decoding in terms of integers is provided as type

struct cl_idecoded_float {
        cl_I mantissa; cl_I exponent; cl_I sign;
};

by the following function:

cl_idecoded_float integer_decode_float (const type& x)

For x non-zero, this returns (-1)^s, e, m with x = (-1)^s * 2^e * m and m an integer with float_digits(x) bits. For x = 0, it returns (-1)^s=1, e=0, m=0. WARNING: The exponent e is not the same as the one returned by the functions decode_float and float_exponent.

Some other function, implemented only for class cl_F:

cl_F float_sign (const cl_F& x, const cl_F& y)

This returns a floating point number whose precision and absolute value is that of y and whose sign is that of x. If x is zero, it is treated as positive. Same for y.

This document was generated on August 27, 2013 using texi2html 5.0.