On Mon, Nov 10, 2008 at 7:37 PM, Didier Verna <didier.verna@...> wrote:
> Now, I'm wondering if conducting these benches always in the same order,
> within the same "bench" function (and so within the same lisp image)
> could introduce a bias.
I don't think it's impossible, but since the bias would mostly due to
alignment (more of an issue on x86 than x86-64, since x86-64 aligns
loops better since 1.020.14) and memory issues I would expect any
given arrangement to hold its own bias.
If you do, however, discover that one of your benchmarks end up
performing noticeably better when run one way, then I would definitely
like to hear about it -- maybe something could be done to make things
Apropos: unless you are already using it, you may be interested in
SB-EXT:CALL-WITH-TIMING -- a lower level interface similar to TIME.