|
|
Our first ExaHyPE demonstrator will be an ADER DG approach for the linear seismic wave-equation in 3D, Peano does grid-management and AMR, and the computational kernels are optimized matrix-matrix operations.
|
|
|
|
|
|
|
|
|
Here are the drawings and specifications of our (Code-) KickOff in Garching on October 24, 2015:
|
|
|
|
|
|
This is the ExaHyPE master-plan.
|
|
|
It specifies the single tasks of a run, clarifies data dependencies and which data are read/written in each step. [Masterplan.docx](https://gitlab.lrz.de/gi26det/ExaHyPE/uploads/a80cc31a7bb6dc4756a2458291f886ea/Masterplan.docx)
|
|
|
[Masterplan.pdf](https://gitlab.lrz.de/gi26det/ExaHyPE/uploads/9a9a19fcd19fd7e437c9f127d209aaa2/Masterplan.pdf)
|
... | ... | @@ -29,3 +32,17 @@ Dimensionality of the tensors/fluxes/... |
|
|
![Data](https://gitlab.lrz.de/gi26det/ExaHyPE/uploads/1d91d0fcc339b22bf550244c79893984/Data.jpg)
|
|
|
|
|
|
|
|
|
|
|
|
Further aspects are:
|
|
|
. We don't want a flexible code with \ifdef and tons of variables, ifs, ....
|
|
|
. Wherever possible we will use const (even for the number of elements).
|
|
|
. For every application, if we introduce local timestepping, new numerics, ..., we will have a dedicated, tailored branch.
|
|
|
. A the moment a fine grain task model of the single computation steps is our favourite.
|
|
|
.
|
|
|
|
|
|
|
|
|
Next steps:
|
|
|
. DUR: Extension of Peano with Cell, Face Handling, Datastructures for DG
|
|
|
. TRE: A minimal/lean ADER DG code for the 3D linear seismic wave equations in FORTRAN
|
|
|
. TUM: Computational Kernels with Matrix-Matrix performance optimization
|
|
|
. All: Check timetables for Coding-Week in Munich, 1st (or 2nd) week of December 2015 |