Closed maltezfaria closed 5 years ago
oh, one possible problem I found was that because we use Float32
, the result for the performance benchmark size n=16000
is actually different from a reference result which uses 64bit floats. Not sure if that will be problem. Should the result be "correct" for the performance benchmark size as well? If so the code should use Float64
then.
What does the other implementations use?
the julia version I departed from already used Float32
. The other versions I glanced at (c and c++ at the top 5) use double
We should probably switch to Float64 then.
on my machine this version is slightly faster. Using
@benchmark
I getChange is to check if abs2>4 only every few iterations. This idea was on one of the c++ codes I found on their website.