Question: 2 Google s bfloat 1 6 The bfloat 1 6 format, shown in Figure 1 , has similar fields as the IEEE 7 5 4
Googles bfloat
The bfloat format, shown in Figure has similar fields as the IEEE floatingpoint format. The
signbit field S indicates whether the number is positive ie S or negative ie S The
exponent field E encodes the value of the exponent as a biased signedvalue, where the bias B is
equal to ; assuming the unsigned value of the exponent field E is valE the value encoded by the
exponent field is:
exp valE
The field F represents the fractionmantissa The mantissa is encoded as an unsigned binary number
with the weight of the mostsignificant bit ie bit number in the format is
; subsequent bits have
the following weights:
To convert the binary fraction to decimal, one can use the
following equation:
F raction Ftimes
Ftimes
Ftimes
Ftimes
Ftimes
Ftimes
Ftimes
S E F
Figure : bfloat Format and EncodingThe bfloat format and encoding showing the different
fields; S: sign bit indicator, E: exponent value, F: fractionmantissa value.
Much like IEEE format, floatingpoint numbers can either be normalized or denormalized.
Normalized floatingpoint numbers have an implicit in the integer part, which is not represented by
the aforementioned fields. Denormalized floatingpoint numbers have an implicit in the integer part
of the number. The equation to calculate the value of the number encoded by the bfloat floatingpoint format is influenced by the value of the exponent field E and the value of the fraction field F
The following subsections outline the different equations based on the values of the exponent field E
and the fraction field F:
Step by Step Solution
There are 3 Steps involved in it
1 Expert Approved Answer
Step: 1 Unlock
Question Has Been Solved by an Expert!
Get step-by-step solutions from verified subject matter experts
Step: 2 Unlock
Step: 3 Unlock
