I hadn't tried rerunning the benchmarking script since then. With these compiler flags, the benchmarking script errors for some of the test executables and stops. Removing the flag allows the runs to execute as expected.
Some preliminary investigation suggests the underflow occurs in the Bernoulli potential calculation when squaring very small velocities. This shouldn't stop execution of the model. I'm tempted to just remove the underflow flag.
https://github.com/edoddridge/aronnax/pull/169 added compiler flags to the test executables to catch underflow errors (amongst others):
https://github.com/edoddridge/aronnax/blob/d6e05bba75063654b069505eff66316e378a25f6/Makefile#L10
I hadn't tried rerunning the benchmarking script since then. With these compiler flags, the benchmarking script errors for some of the test executables and stops. Removing the flag allows the runs to execute as expected.
Some preliminary investigation suggests the underflow occurs in the Bernoulli potential calculation when squaring very small velocities. This shouldn't stop execution of the model. I'm tempted to just remove the underflow flag.