Remove hard coded Float64

[dlfivefifty]

There should be nothing in the algorithms that depend on the parameters being Float64: BigFloat should in principle be allowed, though I suppose there may be assumptions on eps(T) for choosing cut-offs.

[ajt60gaibb]

I would need to think very carefully about this. I agree that there is in principle nothing in the algorithm that requires Float64, but there are some algorithmic parameters that do...

[dlfivefifty]

We could have the routines allow for general choices of the parameters.  In the case of Float64 it uses the current defaults, otherwise, users have to specify their own.

On 28 Mar 2016, at 3:00 AM, Alex Townsend notifications@github.com wrote:

I would need to think very carefully about this. I agree that there is in principle nothing in the algorithm that requires Float64, but there are some algorithmic parameters that do...

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub #22 (comment)

[ajt60gaibb]

I agree. It would also be exciting to add BigFloat versions of all the algorithms too.

[sflammia]

I am working on a problem that would benefit from a BigFloat implementation of both gausslegendre and gausslaguerre. I hope this comment will bump this issue up the priority queue!

[dlfivefifty]

I have proposed it as an undergrad summer project so hopefully it can come together.

[szabo137]

I’d like to bring this issue back to attention. Has there been any progress on supporting additional data types for at least gausslegendre?
From our side, Float32 support would be essential for GPU usage, while BigFloat would also be useful in other cases