long time ocean simulation w/o saving any waveform database

David,

Before I answer that, I need to ask some questions:

1. Is this an OCEAN XL script - i.e. is it running multiple simulations with measurements that are being taken after each simulation (as you'd do in ADE XL/Explorer/Assembler)?
2. If you don't want to save any results, how do you know the simulation was successful?

Regards,

Andrew.

hi Andrew,

this is a valid question. just single step sim yet. our testbench is constructed so that we know the expected output from given input set and self-checked. the lengthy regression is to check if there's any failing point after long random input. if we do see some failing point, we'll print when/what code so that we can just reproduce for that specific case instead the full blown one. 

so you can see for the lengthy case, I don't have to keep any dump (or later I'll selectively dump a few for FFT or any other post-processing). 

I see I can use save() and saveOption() to save a few things. but if I'm not using these command, it seems to dump good size intermediate files (if not necessary waveform) which I'm trying to avoid as I don't want to come back next day and see it crashed because of disk space full after just a couple hours run. 

thanks,

David

Hi David,

If you don't use save() or saveOption(), the default in spectre is save=selected. What that means is that if no signals are "selected" to be saved (i.e. with a spectre save statement, corresponding to the save() function in OCEAN), then all voltages (or at least all "public" or normally useful voltages; i.e. not those which are internal nodes within devices) are saved.

As soon as you save() some signals, only those signals are saved. This is done because if you don't ask to save something specific, the default assumption is that you want to look at something, and better to save all node voltages.

There is a save option in spectre save=nooutput which doesn't output any result database at all, but this isn't available from OCEAN (without an include file containing the option statement in spectre syntax). The closest is saveOption('save "none") which saves a single (random) signal ("none" is a bit of a misnomer, but it's related to how the output code is structured in spectre, where the act of opening the database requires something to be saved).

Hopefully that gives you some choices. Of course, you could use system("rm...") to remove the psf dir once your simulations have finished and you've read any results you need. OCEAN XL (like ADE XL, via the Options->Save form) has an option not to keep the simulation results, which is useful if you're doing lots of simulation.

Regards,

Andrew.

thanks Andrew,

I got it. that gives me enough hooks to play with whenever I need them saved or not. still a little bit more clarification as you mentioned it'll default to save=selected and later you mentioned it'll even default to save all. from my observation, if I just saved a coupled by save( 'v "/in1" "/in2"), I can still plot quite a lot more signals into waveform by plot command in the same Ocean script. does the simulator pre-screen the ocean script ahead so that it knows what signals are already selected to plot so better to save them ahead, even they are not explicitly requested to be saved? 

Andrew Beckett said:

If you don't use save() or saveOption(), the default in spectre is save=selected. What that means is that if no signals are "selected" to be saved (i.e. with a spectre save statement, corresponding to the save() function in OCEAN), then all voltages (or at least all "public" or normally useful voltages; i.e. not those which are internal nodes within devices) are saved.

thanks,

David

Hi David,

My mistake. Whilst the default in spectre is indeed save=selected, in ADE the default is save=allpub (you'll see this if you launch ADE and look at the Save All options form). So if you don't have a call to saveOptions() in your script, then it will save all voltage signals regardless of the presence of save().

So you should also add:

saveOption('save "selected")

to your script. Then it would only save in1 and in2.

Regards,

Andrew.

hi Andrew, yes I observed that as well. It still dumps bunch if I don't do the save "selected" option. now it grows much slower as I'm just dumping for two signals. 

from different side, I observed that AMS simulation defaults to save none so only explicitly saved signals could be plotted. I had to do save all for debugging purpose. something not quite consistent to each other. 

thanks,

David

Hi David,

The difference in defaults between AMS and spectre is mostly historical. Because AMS is based around the INCISIVE digital simulator (with the analog solver engine added), it's commonplace to have a huge number of digital signals and so saving all of those would be expensive; the analog solver with AMS follows that model.

Historically circuit simulators (such as spectre) were used with small circuits and so saving all signals was usually not that expensive. Nowadays the circuits simulated using APS can be very big (millions of nodes) - but we have the original default behaviour in spectre (and ADE) because most simulations are still of smaller circuits.

So the main reason is "that's the way it is!"

Andrew.

haha, yes, understand. thanks a lot Andrew as always.

David