On Friday 06 March 2009, Yamazaki R2 wrote: > **WOW** I'm an idiot I can't believe I didn't think of that! > I'll just add a wrapper script for my simulations to > preprocess the netlist and merge the transistors with the m > parameter. That should be a quick and easy solution until > gnucap automatically does this for you. If anybody is > interested I can post the script on the mailing list for > everybody else to use. I'll probably just use perl for regex > capability.
I don't think gnucap will ever explicitly combine devices like a preprocessor would. What is more likely is to use a cache to save model evaluation time, while keeping them apparently separate. If you really want a combined instance, that's what "m" is for. This discussion leads me to believe that the Gnucap choice of having probes like "gm" which are really the device's state variables, give the value for one of the paralleled devices is a reasonable choice. If you are thinking of methods of speeding up a simulation, that has always been a high priority for Gnucap. I have a circuit here that runs (transient analysis) in 1 minute in Gnucap, set for speed, about 2 minutes in Gnucap, set for "spice-equivalent accuracy", about 3 minutes in Gnucap, with default settings, which are usually more accurate than Spice, about 8 minutes with all options set as slow as possible. It takes 8 hours on NGspice. It's a linear circuit with about 500000 nodes. Spice run time is a quadratic function of circuit size. Gnucap scales linearly. _______________________________________________ geda-user mailing list geda-user@moria.seul.org http://www.seul.org/cgi-bin/mailman/listinfo/geda-user