Daniel Arndt
I can reproduce on `odyssey`(https://systems.nic.uoregon.edu/internal-wiki/index.php?title=Compute:_Odyssey) with ``` $ hipcc --version HIP version: 7.0.51831-a3e329ad8 AMD clang version 20.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-7.0.1 25314 f4087f6b428f0e6f575ebac8a8a724dab123d06e) Target: x86_64-unknown-linux-gnu Thread model: posix InstalledDir: /opt/rocm-7.0.1/lib/llvm/bin Configuration file:...
``` Kokkos::RangePolicy policy(0, nrows + 1); ``` seems to fix it (haven't thought about that mkaing sense or not).
> Since the HIP backend was written from the Cuda one, could it be related with: [#8426](https://github.com/kokkos/kokkos/issues/8426)? That one is using `parallel_reduce` and small types, though.
https://github.com/dealii/dealii/blob/93160909dbf3bbfe986ad5320b675737f89d6e00/examples/step-86/CMakeLists.txt#L40-L52 says: ``` This tutorial requires a deal.II library that was configured with the following options: DEAL_II_WITH_PETSC = ON DEAL_II_PETSC_WITH_COMPLEX = OFF DEAL_II_WITH_P4EST = ON ```
This doesn't look like a `Kokkos` issue but more like a toolchain problem. Can you configure a dummy `CMake` project with the same compiler?
Casn you compile ``` int main(int argc, char *argv[]) { return argc - 1; } ``` with that compiler in the container?
Not a `Kokkos` issue.
Retest this please.
Retest this please.
Retest this please.