poor dogbolt performance

[psifertex]

Just wanted to reach out and ask if there's a way we could improve the performance of Ghidra on dogbolt.org.

You can see how we currently automate decompilation here:

https://github.com/decompiler-explorer/decompiler-explorer/blob/master/runners/decompiler/decompile_ghidra.py#L28-L38
and:
https://github.com/decompiler-explorer/decompiler-explorer/blob/master/runners/decompiler/tools/ghidra/DecompilerExplorer.java

We're seeing significantly worse performance than I'd expect and I'm not sure whether there are any obvious improvements to the automation we can make and any advice from the official devs would be appreciated!