TechnicalTests

Technical Tests

Short tests to check code integrity

• ACCESS runs should be bit reproducible when run with different numbers of processors. We have run 1 day tests with ACCESS1.3+CABLE-2.0 using 128, 64 and 16 processors. Using the .leave output each timestep of 'Absolute Norm' as an indicator, bit reproducibility is maintained across these cases.

• Uninitialised variable test (should be done but hasn't been. - Jhan has code to do this and will upload it to the trunk).

• In the first instance all developments to the source code should be made switchable. Thereby with the switch turned off, the model should reproduce exactly the same results as the base revision. This is a pre-requisite of any development being pushed to the trunk. This is discussed further [BasicTrunkTestIntro here]. In time, these switches may be removed.

---

Back to CableBenchmarking