idea-fasoc / openfasoc-sims Goto Github PK

Currently Ngspice takes a lot of time for -20c case for the temp-sense generator. Current workflow is running the sims for 7 hours and builds the csv files with already available sims data. The workflow must continue to run for more than 7 hours, quit at some point and build the csv file including the missing sim rpts too.
Use https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idtimeout-minutes to cancel the step after a certain number of hours (10 hrs?) and fill such results with the value "timeout" in the run_time column inside the csv file.

Need to add LDO generator to the regression flow. There are a bunch of unit test cases and the regression is a bit different from that of temp-sense generator.

Currently the workflow just does everything (from running the generator to running simulations). For Xyce, it didn't took much time (6 hours) to run prePEX and postPEX is completely failing for some reason. Ngspice is able to generate data at Xyce failing points (temps and configs).

The idea is modify the workflow (./.github/workflows/tempSense_sky130hd.yml) so that first the generator creates the spicedecks and stores them in /home/$USER/runner_results location on the runner GCP machine. Then all sims (except -20c point) is ran in parallel. Later all -20c are ran in parallel. Later, the workflow continues in the old way itself.
The idea is to save sim time and the proposed method is an experiment.

Need to update the workflow names for a better understanding of its operation. Need to update locations inside the workflow so that the csv data dump happens in the proper location inside this repo.

Enable sphinx docs and publish it on readthedocs.io with the explanation regarding the workflows that are developed in this repository.

Currently temp-sense regression sweeps headers from 5 to 10 with a step size of 2 (only odd numbers). But the number of headers should start from 3 and should be incremented by 1 till the count of 11

Mehdi wants to have the run time for each simulation along with the error, freq and power for each temp sim run. Need to figure out how to get that data. One solution could be to use the last updated timing info for every sim log file and determine the amount of time it took by taking the generation of its respective spice file.

Workflows are auto-cancelling after a certain period of time. Earlier that was not the case. Maybe some service in the background on the GCP VM runner is killing this job?

https://github.com/idea-fasoc/OpenFASOC-sims/actions/runs/4163987734/jobs/7231621117