TIP 
Use this information only as a last resort. The CommonPoint system supplies a set of portable functions to manipulate numbers.
The first technique is to cast a float value as an unsigned 32-bit integer. For example:
#define FLOAT_BITS(x) *((uint32_t *) &(x))
x should be a float value whose address may be taken. Use of FLOAT_BITS depends on the fact that the bits of a float, from most to least significant, are ordered as are the bits of an unsigned 32-bit integer.
NOTE
Here are two ways to extract the unbiased exponent of a float value
The situation is more complicated for a double or long double value, where the value is handled in 32-bit pieces. In this case, you must account for the byte order of the platform.x.
Use of the global function Logb is simpler and more portable than masking and shifting the exponent bits. If long unbiasedExp1 = ((int) ((FLOAT_BITS(x) & 0x7F800000) >> 23)) - 127;
float unbiasedExp2 = Logb(x);
x is subnormal, unbiasedExp2 is the (out of range) exponent as though x were normalized, while unbiasedExp1 is one less than the exponent of x as a subnormal value. The float value unbiasedExp2 has special values in case x is zero, infinite, or NaN. Note that applying integer operations to a floating-point value is not a particularly portable technique, as it requires the use of three magic numbers: the mask, the shift count, and the bias.
[Contents]
[Previous]
[Next]
Click the icon to mail questions or corrections about this material to Taligent personnel.