kicad-developers team mailing list archive

[Heikki Pulkkinen <hei6mail@xxxxxxxxx>]

Hi,

As an interest. I tested now with that A64-Olinuxino board.

Result is:

Old algo with parallel zone filling 8s. That 2 times fill algo.
New algo with parallel zone filling 20s. Only fill can be done parallel,
not insulated area cleaning. Filling take approx 2s.
Current master branch 37s.

And my machine is core2duo E6750, 8G.


Heikki




On Sun, Nov 19, 2017 at 6:31 PM, Tomasz Wlostowski <
tomasz.wlostowski@xxxxxxx> wrote:

> On 19/11/17 16:18, Heikki Pulkkinen wrote:
> > Hi,
> >
> > You can dowload it from:
> >
> > https://forum.kicad.info/t/testbench-board-for-kicad/1127
> >
> > I do some more test, and found that new algo would be very fast with my
> > parallell zone filling algo, but it get stuck after filled all 128 pours
> > below 10 seconds and started to removing insulated areas. Other boards
> > just works fine. Maybe it is too good to be true to get results like
> that.
>
> It's a board that I codesigned a long time ago (the White Rabbit Switch
> v3.0) and crudely converted from Altium to Kicad as a performance test.
> It's a very pathological test case, including split power planes
> converted to many polygons with extremely complex outlines and lots of
> tracks not centered on pads/vias (which Altium frequently does).
>
> It takes approx 3 mins to refill all zones on my machine ( i7-4700MQ, 16
> GB RAM).
>
> For comparison, two rather complex designs done from scratch in Kicad
> take much less time to refill.
> - A64-Olinuxino : 17 s
> - cible_ccd (JP's project) : 21 s
>
> Heikki, I'm interested in your parallel zone filling algorithm. Do you
> have it in your Github?
>
> Tom
> >
> > Rgards
> >
> > Heikki
> >
> >
> >
> > On Sun, Nov 19, 2017 at 4:39 PM, Tomasz Wlostowski
> > <tomasz.wlostowski@xxxxxxx <mailto:tomasz.wlostowski@xxxxxxx>> wrote:
> >
> >     On 19/11/17 15:35, Heikki Pulkkinen wrote:
> >     > Hi,
> >     >
> >     >
> >     > Sorry to tell that, but it seems that new connectivity algorithm
> is slow
> >     > with bigger boards. Doing some tests I noticed that  new algo is
> >     > speeding recalculating ratsnest, but it costs manual routing and
> >     > dragging performance. This video shows how big difference is. And
> that
> >     > board is just nothing big.
> >     >
> >     > Zones filling, that was really big difference. Old algo below 3
> mins.
> >     > New one almost 17  minutes. Old algo, has my parallelism algo in
> zone
> >     > filling, but it is doing it twice, with 2 core processor, and most
> of
> >     > the time it is calculating ratsnest before and after filling.
> >     >
> >     >
> >     >
> >     Heikki,
> >
> >     Can you send us (privately) the board that shows the drops in
> >     performance? I'd greatly like to optimize it, with your help if
> >     possible!
> >
> >     Best,
> >     Tom
> >
> >     PS. What's the CPU/RAM of your PC?
> >
> >
> >
> >     > Regards
> >     >
> >     > Heikki
> >     >
> >     >
> >     > https://youtu.be/JS57hRyzmdg
> >     >
> >     >
> >     >
> >     >
> >     > _______________________________________________
> >     > Mailing list: https://launchpad.net/~kicad-developers
> >     <https://launchpad.net/~kicad-developers>
> >     > Post to     : kicad-developers@xxxxxxxxxxxxxxxxxxx
> >     <mailto:kicad-developers@xxxxxxxxxxxxxxxxxxx>
> >     > Unsubscribe : https://launchpad.net/~kicad-developers
> >     <https://launchpad.net/~kicad-developers>
> >     > More help   : https://help.launchpad.net/ListHelp
> >     <https://help.launchpad.net/ListHelp>
> >     >
> >
> >
>
>