use an abstraction layer that gives a unified programming interface to distinctive computational backends – a so-named “overall performance
To show scaling under production problems of actual scenarios, we use an idealised rainfall runoff simulation over the Lessen Triangle area in
As added SERGHEI modules turn out to be operational, the HPC abilities will further empower simulations that are unfeasible with The existing era of
on the flume inlet gate). The simulation final results approximate experimental final results effectively. The mismatches are similar to those analysed by
A important obstacle that arises when protecting a number of programming designs during the supply code is performance portability, i.e., the peace of mind that the parallel effectiveness and scalability from the code is preserved throughout various HPC components. To the best of our know-how, Regardless of its great importance, efficiency portability hasn't yet been the main focus of your computational hydrology Neighborhood.
under various disorders. The field-scale exams showcase the applicability on the solver for serious difficulties, and allow for arduous computational
SERGHEI is usually a modular simulation framework fixing for surface and subsurface hydrodynamics (shallow h2o and Richards equations) along with the involved transportation of issue. SERGHEI can for that reason addresses several different applications, from flood modelling to simulating the cascade of make any difference throughout landscapes.
The simulated discharge hydrograph for the outlet is in comparison towards the analytical solution in Fig. eight. The numerical remedies
Exact wetting/drying: Just about every surface area mobile's exchange is computed according to its precise condition (wet or dry)
the place MYBACKEND need to get replaced by on the list of offered Kokkos backends, While using the corresponding Kokkos architecture keyword replacing MYARCH.
Table 2Analytical steady flows: summary of L norms for faults in drinking water depth; L norms for mistakes in device discharge are from the array 789CLUB of machine precision and omitted listed here.
This approach appropriately handles conditions the place: Partly wetted subsurface cells (some surface cells damp, some dry)
be compiled for both OpenMP programs and GPUs with Kokkos managing the low-degree parallelism on unique backends.
versions showcasing surface circulation have relied on kinematic or zero-inertia (diffusive) approximations because of their apparent simplicity