Pyvsc Performance issue

[vandanabharath]

Hii
I'm running pyvsc constraints file which contains more than 2000 constraints, but it is taking more than 10mins to solve the complete file. How can I overcome this & improve simulation time.

[alwilson]

There are undocumented profiler and debug options that might be useful here.

You can set the VSC_PROFILE env var to list out the stats for all randomize calls. This includes how many randsets each call breaks down into, which are disjoint sets of variables and constraints. Would be interesting to improve that to show stats per randset so that large performance issues can be broken apart, if possible.

pyvsc/src/vsc/profile/__init__.py

Line 20 in 1734432

if "VSC_PROFILE" in environ.keys() and environ["VSC_PROFILE"]:

You can view how a randomize call is broken down and solved using randomize(debug=1). That might clue you in on how your constraints got converted into randsets. You could just stop your python process randomly and sample which randomize call and randset you're in as a profiler! 😆

Having a timeout option would be nice here. There's already functionality to print debug info whenever a randomize call fails. We could print out the randset that fails a timeout and possibly even extract the failing core back to a runnable python test case.

Alternatively py-spy is a great profiling tool for visualizing python profiling information in flamecharts. You would be able to see how long your program spends in each randomize call and where in pyvsc as well.