Dump optimized graph when logging in already-optimized PE (#44315)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/44315
I find it more intuitive to dump the optimized graph if we have one;
when I first saw the unoptimized graph being dumped I thought we had failed to
apply any optimizations.
Test Plan: Observe output by hand
Reviewed By: Lilyjjo
Differential Revision: D23578813
Pulled By: bertmaher
fbshipit-source-id: e2161189fb0e1cd53aae980a153aea610871662a