A few things could be optimised, off the top of my head:
- Heaps of graph-related work is repeats between processes and used again and again
- Could start launching solvers as the patch .inp files are generated, not after they're all generated.
- Fast cadence for outputting nodal displacments, slow cadence for everything else.
Then another profiling pass...
A few things could be optimised, off the top of my head:
Then another profiling pass...