After reading at Speed Comparison - Programming Languages I took a look at the Racket code to see if there is something interesting. The last line of speed-comparison/src/leibniz.rkt at master · niklas-heer/speed-comparison · GitHub says (I removed unsafe. to make it easier to read)
(fl+ pie (fl/ x (fl- (fl* 2.0 (fx->fl i)) 1.0)))
It's obviously better to calculate
(fl+ pie (fl/ x (fx->fl (fx- (fx* 2 i) 1))))
but when I run the code I get almost the same time. I expect`a tiny difference, like a .5%, but I only get noise. It's actually too noisy.
Why??? Is the floating point fast enough now? Does the pipeline hide the difference? Am I missing something?