16bit float:
- 1 bit for the sign
- 5 bits for the exponent, with an exponent bias of 16???
- 10 bits for the mantissa.
http://en.wikipedia.org/wiki/Half_precision_floating-point_format
IEEE 754 half precision binary floating-point format: binary16
- 1 bit for the sign
- 5 bits for the exponent, with an exponent bias of 15
- 10 bits for the mantissa. (11 bits of precision --> 1.mantissa)
沒有留言:
張貼留言