add maxintfloat(::Type{BFloat16})

[milankl]

maxintfloat(T) is the largest integer n so that n+1 is not representable anymore with a given format T. For BFloat16 this is $256 = 2^{n_m+1}$ with $n_m = 7$ the number of mantissa bits.

julia> BFloat16(256)
BFloat16(256.0)

julia> BFloat16(257)
BFloat16(256.0)

julia> BFloat16(258)
BFloat16(258.0)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (1278fa9) 66.66% compared to head (e9117ea) 66.85%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #60      +/-   ##
==========================================
+ Coverage   66.66%   66.85%   +0.18%     
==========================================
  Files           3        3              
  Lines         180      181       +1     
==========================================
+ Hits          120      121       +1     
  Misses         60       60              

Files	Coverage Δ
src/bfloat16.jl	71.85% <100.00%> (+0.16%)	⬆️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[JeffreySarnoff]

lgtm

julia> function maxint(T, maxint=2^(1+precision(T)))
ints = 1:maxint
findfirst(iszero, ints .== Int.(BFloat16.(ints))) - 1
end;

julia> maxint(BFloat16)
256