New IEEE Floating-Point Types

Oracle 10g introduces the new IEEE floating-point types BINARY_FLOAT and BINARY_DOUBLE. The types are extremely efficient for heavy floating point computations as the work is passed directly to the operating system. Literal assignments can be perfomed using the "f" and "d" suffixes or conversion functions TO_BINARY_FLOAT and TO_BINARY_DOUBLE:

DECLARE   l_binary_float   BINARY_FLOAT;   l_binary_double  BINARY_DOUBLE; BEGIN   l_binary_float  := 1.1f;   l_binary_double := 1.00001d;     l_binary_float  := TO_BINARY_FLOAT(1.1);   l_binary_double := TO_BINARY_DOUBLE(1.00001); END;

Rather than raise exceptions, the resulting values of computations may equate to the following constants that can be tested for:

• [BINARY_FLOAT|BINARY_DOUBLE]_NAN
• [BINARY_FLOAT|BINARY_DOUBLE]_INFINITY
• [BINARY_FLOAT|BINARY_DOUBLE]_MAX_NORMAL
• [BINARY_FLOAT|BINARY_DOUBLE]_MIN_NORMAL
• [BINARY_FLOAT|BINARY_DOUBLE]_MAX_SUBNORMAL
• [BINARY_FLOAT|BINARY_DOUBLE]_MIN_SUBNORMAL

The constants for NaN and infinity are also available in SQL.

Fonte: http://www.oracle-base.com/articles/10g/PlsqlEnhancements10g.php