Can't plot transient signals until after simulation is complete

Dear forXTer,

A couple of questions if I may....

1. Have you navigated in UNIX to the psf directory to see if there is a "spectre.out" file - or are you just trying to locate it with the Results Browser?

2. Have you examined or changed any of your flushtime parameters? These parameters detail how often the data is written to disk. The more freuqently it is is written, the greater the overall simulation time. These options are under the "Options->Output tab" of the transient analysis and are documented at URL:

https://support.cadence.com/apex/techpubDocViewerPage?xmlName=spectreref.xml&title=Spectre%20Circuit%20Simulator%20Reference%20--%20Analysis%20Statements%20-%20Transient%20Analysis%20(tran)&hash=pgfId-1063371&c_version=19.1&path=spectreref/spectreref19.1/chap3.html#pgfId-1063371

or as shown below.

Shawn

Hi Shawn,

All of the flush parameters parameters in the output tab were blank, but I did finally see results after setting CDS_PSFXL_FLUSH_INTERVAL to 300. I think this is default, but I guess it must have forced a flush. I'll mess around with these a bit more, but I think this will fix my problem, thank you for the suggestion.

Hi Shawn,

All of the flush parameters parameters in the output tab were blank, but I did finally see results after setting CDS_PSFXL_FLUSH_INTERVAL to 300. I think this is default, but I guess it must have forced a flush. I'll mess around with these a bit more, but I think this will fix my problem, thank you for the suggestion.

Dear forXTer,

forXTer said:

, but I did finally see results after setting CDS_PSFXL_FLUSH_INTERVAL to 300.

Excellent - thank you for letting me know! Just for your information, the values we use follow:

$ env | grep FLUSH
CDS_PSFXL_FLUSH_INTERVAL=10
CDS_PSFXL_INIT_FLUSH_INTERVAL=10
CDS_PSFXL_MAX_FLUSH_INTERVAL=10

I can't say I see any noticeable simulation time increase with the maximum flush interval set to 10. However, I have not benchmarked simulations with the value varied from 10 to 300. I do know other team members experienced your problem and the default value was reduced from 300. I lost track of what it has been set to since that problem occurred a number of years back.

Shawn

Note that the flushpoints/flushtime/flushperiod parameters of a tran analysis don't help with PSF XL, you need to use the UNIX env vars that Shawn mentioned. However, I would caution against setting CDS_PSFXL_FLUSH_INTERVAL and CDS_PSFXL_MAX_FLUSH_INTERVAL so low, as if they are set very low, it limits the chunk size of the data that is written, which compromises read performance (that's one of the performance gains that PSF XL offers). I understand why you might want to have the flush interval lower, particularly near the beginning if a simulation to make sure it's starting OK, but if the simulation is taking hours, then after it's up and running, there's not so much need to see immediate feedback every 10 seconds. So that's why the CDS_PSFXL_INIT_FLUSH_INTERVAL is more reasonable to set, to allow initial quick updates, but then that increases over time so that you get the read-performance gains of bigger chunks being saved.

Andrew

Hi Andrew,

Thank you for adding your insights!

Andrew Beckett said:

I understand why you might want to have the flush interval lower, particularly near the beginning if a simulation to make sure it's starting OK, but if the simulation is taking hours, then after it's up and running, there's not so much need to see immediate feedback every 10 seconds. So that's why the CDS_PSFXL_INIT_FLUSH_INTERVAL is more reasonable to set, to allow initial quick updates, but then that increases over time so that you get the read-performance gains of bigger chunks being saved.

Your comment about changing the value of environmental variables after a validating that the simulation is progressing as expected is a very good one. Thank you, as always!

Shawn