Re: gEDA-user: gnetlist speed

[John Doty <jpd@space.mit.edu>]

Dan:

> Something bad is going on here.  I have a 200 net flat schematic.  I ran 
> through all of the gnetlist backends and they mostly take about 6.5 to
> 7.5 seconds except for bom2 which took around 12 seconds.  This is on a
> 440 MHz UltraSparc-ii based machine.  You about about 4x the nets, but
> a machine which should easily be faster than mine yet its taking nearly
> 40 times longer for you.
> 
> Just for kicks, how long does it take if you use '-g geda' (the geda
> backend)?

The time is pretty much the same independent of the backend. I've tried
"geda", "bom", "partslist1", and my own backend. "partslist1" takes the
time and then dies with a stack overflow (choked by 800 parts?). The
others appear to work OK (I just got to the netlist stage yesterday, so I
haven't done any but the most cursory checks).

The time may depend on other stats than just the net count. My design is
organized as four files (a different organization might be easier for the
geometry to topology transformation). I have a *lot* of components (~300
each of resistors and capacitors). I've done some big connectors as
multi-slot components (one pin per slot), which makes for a very large
number of symbols.

Ales:

> One way to check for sure where the time is being spent is to run
> gnetlist with the -v flag and watch the verbose output.  My guess all
> of this 4.5 minutes is being spent creating the internal representation.
>...

Yep. Over four minutes of stuff like "CpnnPnP". I don't consider this a
big deal: compared to the months it has taken to design this monster, the
netlisting time is negligible.

John Doty		"You can't confuse me, that's my job."
Home: jpd@w-d.org
Work: jpd@space.mit.edu