Floating point works with below 32 bits but the precision and range starts suffering drastically. Basically, the TRSEncodedBits class divides a floating point number into 1 bit Sign (PreserveSignBit = True), 1/4 Len Exponent, and 3/4 Len - 1 Mantissa. The class tries to fit the exponent into the +/-range of bits, e.g., if the exponent has 8 bits, the range of exponents can be -126 to 127. Exponent overflows can occur and are not flagged.
|
