Skip to content

macro to change the default floating-point precision in Julia code

License

Notifications You must be signed in to change notification settings

JuliaMath/ChangePrecision.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

34 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ChangePrecision

CI

This package makes it easy to change the "default" precision of a large body of Julia code, simply by prefixing it with the @changeprecision T expression macro, for example:

@changeprecision Float32 begin
    x = 7.3
    y = 1/3
    z = rand() .+ ones(3,4)
end

In particular, floating-point literals like 7.3 are reinterpreted as the requested type Float32, operations like / that convert integer arguments to Float64 instead convert to Float32, and random-number or matrix constructors like rand and ones default to Float32 instead of Float64. Several other cases are handled as well: arithmetic with irrational constants like pi, linear-algebra functions (like inv) on integer matrices, etcetera.

The @changeprecision transformations are applied recursively to any include(filename) call, so that you can simply do @changeprecision Float32 include("mycode.jl") to run a whole script mycode.jl in Float32 default precision.

Code that explicitly specifies a type, e.g. rand(Float64), is unaffected by @changeprecision.

Note that only expressions that explicitly appear in the expression (or code inserted by include) are converted by @changeprecision. Code hidden inside external functions that are called is not affected.

This package is for quick experiments, not production code

This package was designed for quick hacks, where you want to experiment with the effect of a change in precision on a bunch of code. For production code and long-term software development in Julia, you are strongly urged to write precision-independent code — that is, your functions should determine their working precision from the precision of their arguments, so that by simply passing data in a different precision they compute in that precision.