Stream: helpdesk (published)

Topic: 2pi as Float32

Júlio Hoffimann (Jan 19 2021 at 16:41):

What is the most appropriate method to produce 2pi in single precision? Do we need to explicitly convert from the double precision result?

Júlio Hoffimann (Jan 19 2021 at 16:42):

I have geometric algorithms with angles that need to do 2pi - theta and this is creating some type instabilities in single precision when theta is single

Júlio Hoffimann (Jan 19 2021 at 16:43):

So my question is: is there a method to convert the irrational pi to a single precision float directly?

Fredrik Ekre (Jan 19 2021 at 16:45):

Float32(pi)

Júlio Hoffimann (Jan 19 2021 at 16:50):

Yes, I guess that is the best option :smile:

Mason Protter (Jan 19 2021 at 16:52):

there's always

julia> 2f0π
6.2831855f0

Júlio Hoffimann (Jan 19 2021 at 16:58):

cool trick too :smile:

Júlio Hoffimann (Jan 19 2021 at 16:58):

but in this case I need to be generic with a type T(pi) inside a function

Mason Protter (Jan 19 2021 at 17:05):

You may be interested in ChangePrecision.jl

Mason Protter (Jan 20 2021 at 16:59):

@Júlio Hoffimann and @Fredrik Ekre would you mind if this was moved to #helpdesk (published)?

Júlio Hoffimann (Jan 20 2021 at 17:21):

Not at all :smile:

Notification Bot (Jan 20 2021 at 17:45):

This topic was moved here from #helpdesk > 2pi as Float32 by Mason Protter

Last updated: Nov 06 2024 at 04:40 UTC