[Shootout-list] Re: OCaml harmonic benchmark

[Jon Harrop]

On Sunday 01 May 2005 18:07, Isaac Gouy wrote:
> --- Jon Harrop <jon@ffconsultancy.com> wrote:
> > I've reformatted the OCaml harmonic benchmark, making it a few lines
> > shorter.
> > Surprisingly, this also made it twice as fast on my computer.
>
> This is a situation where it's interesting to show both programs.

Actually I was wrong. If you compile them with the same command that the 
shootout uses then they have exactly the same running times. My new version 
runs as fast without needing the "-inline 100" switch, although I'm not sure 
why as I didn't think I'd inlined anything important...

Anyway, the time is hugely noisy because it is so short:

$ time ./harmonic.ocaml_run 10000000
16.695311366

real    0m0.184s
user    0m0.180s
sys     0m0.000s
$ time ./harmonic.ocaml_run 10000000
16.695311366

real    0m0.384s
user    0m0.200s
sys     0m0.010s
$ time ./harmonic.ocaml_run 10000000
16.695311366

real    0m0.259s
user    0m0.200s
sys     0m0.000s

For n = 10^8 instead, the times are insignificantly different for the old and 
new versions.

Incidentally, I was going to submit this:

  let sum = ref 1.0 and ifloat = ref 2.0;;
  for i = 2 to int_of_string Sys.argv.(1) do
    sum := !sum +. 1.0 /. !ifloat;
    ifloat := !ifloat +. 1.0
  done;;
  Printf.printf "%0.9f\n" !sum

but it is vastly slower, probably because ocamlopt fails to unbox the floats.

-- 
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
Objective CAML for Scientists
http://www.ffconsultancy.com/products/ocaml_for_scientists