[FLASH-USERS] questions about the FFT in FLASH, thanks in advance!

[Sheng Di]

Hi guys,
I have three questions about FFT used in FLASH and look forward to answers
here. We are studying if the FFT result is still valid by lowering the
precision of its input data. That is, we need to understand the impact of
the data distortion on the input of FFT to the result.

Question 1: Is FFT very sensitive to the precision of the input data? Does
FFT have to use double-precision in the application?
(We can't characterize it by ourselves because we don't know how to assess
its validity - see Question 2)

Question 2: How to assess the correctness/validity of the FFT's result? Is
there an evaluation metric to do so?

Question 3: Does the FFT in FLASH have very heavy communication overhead
(e.g., because of all-to-all operation or so)? Replacing the double input
data by single-precision input data may improve the performance, but don't
know if the result is still acceptable.

Many thanks in advance!

Best,
Sheng
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20180706/75275756/attachment.htm>