generalize our BigFloat(::Float64) method to more integer and floatingpoint types.

[oscardssmith]

#47546 made BigFloat(::Float64) about 2x faster, but basically the same optimization could be done for the common Integer types (Base.BitInteger64), Base.IEEEFloat and Rational (where the denominator is a power of 2).

[vortex73]

Hey i would like to try this out. Please provide me details on this .

[oscardssmith]

Sounds great! The first step is to read and understand the changes in #47546. From there the next step is probably to make that method accept Base.IEEEFloat (which is just Union{Float16, Float32, Float64}) and delete the current method for constructing BigFloat from Float16 and Float32.

[vortex73]

Sorry for the late response. I shall get working on this and inquire if i have any doubts!

[ghost]

Hi, I'm working on this issue. It's also my first issue, and I'm not already familiar with MPFR library. Lots of questions

• I could not find Base.BitInteger64 in the docs. Is it the same as Int64? Should I just keep the catch-all Integer method and create one that specifies a different type? Basically only line being ccall((:mpfr_set_z, libmpfr), Int64, (Ref{BigFloat}, Ref{BigInt}, MPFRRoundingMode), z, x, r)

• For the following code block,

function BigFloat(x::Rational, r::MPFRRoundingMode=ROUNDING_MODE[]; precision::Integer=DEFAULT_PRECISION[])
    setprecision(BigFloat, precision) do
        setrounding_raw(BigFloat, r) do
            BigFloat(numerator(x))::BigFloat / BigFloat(denominator(x))::BigFloat
        end
    end
end

How is the do keyword working without renaming the function?

I don't actually already know why a power of 2 Integer denominator on a rational would allow optimization, so e.g. '00...001000'. Is there another ccall function to use?

• How do I run the performance tests to check if there was a speedup?

• I'm not able to run make testall on my computer. Is it ok to just do `make -C test mpfr", or could the changes break other libraries?