Add test case comparing -O3 model simulation with -O2

[dweindl]

Would be good to add a test case comparing simulation results from -O3 compiled model to those from -O2 for a non-trivial model. For repeated simulations during parameter estimation, -O3 is clearly preferably, but may lead to issues when relying on undefined behavior or due to compiler bugs.

[FFroehlich]

Static analysis won't tell us where those issues are? If things work for -O3 (which is the default, right?) shouldn't they also work for -O2?

[dweindl]

Static analysis won't tell us where those issues are?

Everything besides compiler issues probably yes.

If things work for -O3 (which is the default, right?) shouldn't they also work for -O2?

If O3 is right, then O2 should be as well. But we are running tests on debug builds, don't we?

[FFroehlich]

Static analysis won't tell us where those issues are?

Everything besides compiler issues probably yes.

If things work for -O3 (which is the default, right?) shouldn't they also work for -O2?

If O3 is right, then O2 should be as well. But we are running tests on debug builds, don't we?

Oh, good point!