Re: funcall consing

[Tomas Hlavaty]

On Sat 01 Jan 2022 at 09:26, Eli Zaretskii <eliz@gnu.org> wrote:
> Note that one of the examples posted by the OP does exactly twice GC
> cycles that the other.  Which leads me to believe it had twice more
> opportunities for checking how much consing was done.
>
> So I think the caveat not to put too much faith in the _number_ of GC
> cycles is still something important to keep in mind.

In my example:

(benchmark-run 10 (dotimes (i 100000) (1+ i)))
;;(2.720941123 40 1.7525918699999998)
(let ((x (lambda (i) (1+ i)))) (benchmark-run 10 (dotimes (i 100000) (funcall x 
i))))
;;(4.9373091140000005 80 3.4835688719999958)

the third number is the time taken by garbage collection.  It is also
roughly doubled for the second case indicating that the second case
really generated more (say about twice as much) garbage.

Otherwise benchmark-run is not really useful for what I need.  In that
case, what is the preferred way to measure execution speed and consing?
I do not need absolute numbers, just something so that I can compare two
different implementations of the same thing and choose the better one or
focus on places to improve.