Surprise...

[Brian Milby]

While this test may be valuable in comparing different versions of LC, I'm
not sure how useful the data is by itself.  I decided to run a test and see
what happened.  I defined 6 constants.  kA-kE for the size of each loop.
kF for the last line in tvar/element in the array.  To initialize, I just
made it simple and put 1,2,3...kX for the lines and T[1][1] = 1, T[1][2] =
2,... for the array.  To add an additional wrinkle, I added additional
lines to tvar such that the last line could be 5 or some other number.  I'm
on an iMac/Sierra running 9.0.1rc1.  Times are milliseconds.  First line is
the constant values.  First test is the original from the start of the
thread.

34 10 11 10 34 5
item: 4159 / array: 9718 / 0.427969

34 10 11 10 34 50
item: 5200 / array: 10346 / 0.50261

34 10 11 10 34 500
item: 36573 / array: 10298 / 3.551466

340 10 11 10 34 5
item: 81108 / array: 97007 / 0.836105

340 10 11 10 34 50
item: 213107 / array: 103755 / 2.053944

Notice not much relative difference between 5 and 50 lines in the original
test (but item did take 20% more time), but when I jumped it to 500 the
advantage shifted to array.  Notice how linear the array times are when the
outer loop is changed - compare test 1 to 4 and 2 to 5.  The number of
outer loops increased by a factor of 10 and the array time jumped by a
factor of 10.  Compare to the item where it was 20x and 40x.  This is where
the advantage of arrays lies - it is constant time to access, so the cost
is mainly a factor of the number of reads.

I also did the first and third test on 8.1.9 (slightly slower than 9):

34 10 11 10 34 5
item: 4828 / array: 10714 / 0.450625

34 10 11 10 34 500
item: 37806 / array: 11057 / 3.419191

If anyone wants the stack or code, let me know.

Thanks,
Brian

On Mon, Aug 20, 2018 at 4:03 PM, Curry Kenworthy via use-livecode <
use-livecode at lists.runrev.com> wrote:

>
> Hey Folks,
>
> Of course I'm interested in this thread. Always a big concern. Earlier
> this year, and last year, I was doing face-off competitions between
> LiveCode versions to test how optimization is faring.
>
> When I get time, after some other priorities and addon updates, I would
> like to do another round of cross-version testing to see who is the current
> champion. (IIRC, I think 6.7 was last time, for Windows.)
>
> Meanwhile....
>
> jbv:
>
> > the same data on LC 6.5.2 desktop Mac, and arrays run 20% faster
> > than items... On LC9 server I get opposite results : items are
> > 2 times faster then arrays.
>
> 1. Try LC 6.5 desktop vs LC 9 desktop: how does that fare? And maybe LC 9
> desktop vs LC 9 server? I would like to distinguish LC 6/9 from LC
> desktop/server. Apples to apples testing, then conclusions are possible.
>
> 2. Sample code stack would be great.
>
> Mark T.:
>
> > I wrote a stack that demonstrated this and other concepts for
> > LC Mark directly. His Global talks called ‘Optimizing Your Code’
> > are partially based on some things we have been discussing.
>
> Optimizing our code is a good thing, and something I am paranoid about and
> specialize in helping people with. Optimizing LC's own code is also a good
> thing, and really the bigger question here perhaps for this thread!
>
> So...I'll be quietly following this thread. No time to participate yet,
> but will later.
>
> Best wishes,
>
> Curry Kenworthy
>
> Custom Software Development
> LiveCode Training and Consulting
> http://livecodeconsulting.com/
>
>
> _______________________________________________
> use-livecode mailing list
> use-livecode at lists.runrev.com
> Please visit this url to subscribe, unsubscribe and manage your
> subscription preferences:
> http://lists.runrev.com/mailman/listinfo/use-livecode
>