[Shootout-list] fannkuch (timer resolution; HZ=1000?)

[Jon Harrop]

On Friday 20 May 2005 23:31, Isaac Gouy wrote:
> Sorry, I did misread your posting. I didn't notice that you stopped
> saying "using only tests which ran for >0.1s or even >1s but there is
> virtually no such data on the shootout", and started saying that 50% of
> the tests ran for >0.1s and 20% ran for >1s. That's a lot for virtually
> none.

Apples and bananas.

The positions of C# and Java on my graphs will be wrong as a consequence of 
these systematic errors. If N is increased then I think the results will be 
more reliable and both Java and C# should shift to better performance. 
Currently, they are both comparable to OCaml bytecode, which I don't believe 
is representative.

> For the current shootout there are 8 out of 28 tests <=0.1s

For <1s, I get 22/28 tests:

ack, count-words, fann, harmonic, heapsort, mandelbrot, nsieve, nsieve-bits, 
object, object-methods, pidigits, random, regex, reverse-complement, 
spectralnorm, spellcheck, startup, tcp-request-reply, tcp-stream, threads, 
threads-flow and word-frequency.

> Of those, 4 are remnants from the old Doug Bagley.

Several of those 22 are supposed to be numerically intensive: ack, harmonic, 
heapsort, mandelbrot, nsieve*, pidigits and spectralnorm.

> No doubt the max N for those 5 will be increased sometime.

I think it would be good to to increase the fastest running times to ~1s for 
the eight tests that I just listed.

> Shockingly there are several things much more broken than this which we
> need to fix.

Really?

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