Kjetil Olsen Lye

Make SolverAdapter explicitly depend on the communication object

5

This fixes the MPI issues with CUDA/HIP/GPUISTL that were introduced with the `GhostLastMatrixAdapter` operator. The previous design of the SolverAdapter relied on the presence of the function `getCommunication` to determine...

Making BlackOilFluidState GPU friendly

4

This is in draft at the moment. Essentially this adds decorators to the functions and eventually will add support non-static fluidsystem-objects.

Work in progress: Primary variables on GPU

2

This is a very incomplete attempt at the moment. To get the DUNE vectors with GPU decorators, those classes have in essence been copied over to the repo and then...

Tests for BlackOilFluidState on GPU with the [opm-common PR 4472](https://github.com/OPM/opm-common/pull/4472)

This adds a much needed unit test for the `OPM_IS_INSIDE_HOST_FUNCTION` and `OPM_IS_INSIDE_DEVICE_FUNCTION` macros. This PR needs OPM/opm-common#4465

Added docker images for CUDA and ROCm

2

I am not claiming these Dockerfiles are perfect, but they seem to do the job (ROCm builds but running not tested yet, will test on Monday). Potentially we could combine...

Preprocess int64

1

Just a draft pull-request to measure the effects of moving the whole grid processing to int64

This is a draft PR that converts the clang-config file to the new YAML format.

Added git-clang-format as a pre-commit hook

14

I am opening this PR also to create a discussion. Before you go on reading, please note that this can be selectively disabled by exporting the `SKIP` environment variable to...

Made a standalone executable to dump grid information for partitioning

1

Maybe this is already easily available, but I am not aware anything like this at least. Currently only supports grid without wells. Does not support weights yet. Use case: Running...