Number type
Passing Float64 or Float32 to the grid constructor causes the grid to store all numbers with 64-bit or 32-bit floating point precision.
When not using Float64 be careful to not mix different precisions as it could introduce implicit type conversions which can negatively effect performance. You can pass the number type desires to many constructors to enforce the type you want: e.g. RectilinearGrid(CPU(), Float32; size=(16, 16, 16), extent=(1, 1, 1)) and ScalarDiffusivity(Float16; κ=1//7, ν=2//7).
While we run many tests with both Float32 and Float64 it is not clear whether Float32 is precise enough to provide similar accuracy in all use cases. If accuracy is a concern, stick to Float64.
We will be actively investigating the possibility of using lower precision floating point numbers such as Float32 and Float16 for fluid dynamics as well as the use of alternative number types such as Posits and Sonums.