Vasily,

Congrats on the Good work!

I have a rather naive question.

I see you have used single precision for the calculations.

Is it “ok” to have solvers written in single precision?

Is double precision absolutely necessary?

Appreciate any help.

We ourselves are looking to GPUize a L-D-L(T) factorization. The questions are from that context. FYI.

Thanks,

Best Regards,

Sarnath