From https://petsc.org/release/changes/318/ it appears that the pc_gamg_sym_graph option was removed (a couple of years ago) from petsc. I think the purpose of this was to communicate some symmetry of the problem to petsc, so it doesn't spend time trying to work it out itself.
We should update how this is done, which might be an optimisation.
This option appears in our compressible solver parameters