@rki looks like you found your answer

can confirm variance in results. had somewhat better results after translating Ehler's code manually
(although my implementation felt way too sluggish)