Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 1 addition & 5 deletions content/docs/configuration/basics/configuration-mapping.md
Original file line number Diff line number Diff line change
Expand Up @@ -143,7 +143,7 @@ To use this feature, build preCICE from source with [Kokkos-Kernels or Ginkgo en
| Mapping tag | Dependencies | `cuda` | `hip` | `sycl` | `openmp` | `cpu` | MPI-parallelization |
|-------------------------|--------------------------------------------------------------------------------------------------------------------|:------:|:-----:|:------:|:--------:|:-----:|---------------------|
| `rbf-pum-direct` | (Eigen, [mandatory](https://precice.org/installation-source-dependencies.html#required-dependencies) for preCICE) | — | — | — | — | ✓ | distributed |
| `rbf-pum-direct` | Kokkos-Kernels and Kokkos (preCICE >= v3.4.1) | ✓ | ✓ | ✓ | ✓ | — | distributed |
| `rbf-pum-direct` | Kokkos-Kernels and Kokkos (preCICE >= v3.5.0) | ✓ | ✓ | ✓ | ✓ | — | distributed |
| `rbf-global-direct` | Ginkgo and Kokkos (preCICE >= v3.2) | ✓ | ✓ | — | — | — | gather-scatter |
| `rbf-global-direct` | (Eigen, [mandatory](https://precice.org/installation-source-dependencies.html#required-dependencies) for preCICE) | — | — | — | — | ✓ | gather-scatter |
| `rbf-global-iterative` | Ginkgo and Kokkos (preCICE >= v3.2) | ✓ | ✓ | — | ✓ | — | gather-scatter |
Expand Down Expand Up @@ -171,10 +171,6 @@ Mapping configurations that follow a gather-scatter approach are always computed

By contrast, distributed mappings solve their rank-local problem in parallel. In practice, this means that the configuration snippet above assigns multiple MPI ranks to a single GPU (with device ID 0). Since oversubscribing a device is typically undesired, use `gpu-device-id="auto"` to assign MPI ranks in a round-robin fashion to the available GPUs. Also note -- due to the different parallelization strategies -- `n-threads` configures the number of threads per executing rank, i.e., for a distributed parallelization, `n-threads=10` assigns 10 OpenMP threads to each MPI rank, whereas for a gather-scatter parallelization `n-threads=10` assigns 10 OpenMP threads to the primary rank only.

{% note %}
Note that as of preCICE v3.4 the `rbf-pum-direct` executor only supports `consistent` constraints. Supporting a `conservative` is [work in progress](https://github.com/precice/precice/pull/2536).
{% endnote %}

More details on the feature can be found in [Schneider et al. 2023](https://doi.org/10.23967/c.coupled.2023.016).

## Geometric multiscale mapping
Expand Down
Loading